Blue is the account, green is QQQ, yellow is SPY

QQQ: 36.08%

QQQ Killer: 41.42%

S&P 500: 26.48%

Secular Opportunity Index: 23.45%

QQQ: 25.34%

SPY: 19.68%

Current Allocation:

QQQ Killer: 28.72%

QQQ: 23.38%

Current allocation

In 2014, AWS revenue was $4.6B, and this year (2025) it will likely end at a run rate near $130B. Operating income improved from $36.9B in 2023 to $68.6B in 2024, with operating margins increase from 6.4% to 10.8%.

Bezos famously ran Amazon to report zero operating profits on purpose. Why? Because if you don’t have profits then you don’t pay taxes. As long as you have attractive investments to make in the business paying taxes is a squandered opportunity.

These days I’m feeling a little bit like I’m in the twilight zone. Wall Street and the Xverse is abuzz with criticisms of the massive capital expenditures being laid out by Big Tech. Massive capex doesn’t just suck up cash today, it depresses earnings through depreciation into the future. It also increases operating expenses in perpetuity as more people are needed to manage the larger capital base. Still, shouldn’t we all be thrilled that these giant companies with stellar decades long track records believe they still have markets to go after that justify $80B annual capex budgets?

Amazon’s capex has increased from $4.9B in 2014 to $83B TTM and will likely be higher next year. This capex budget makes it possible to order products at night and get them delivered by morning in most big cities (by 2025 year’s end). It runs a substantial portion of the internet’s websites. The machines who are waking up are powered by it.

The only thing that worries me about Big Tech is that they are very obvious targets in a trade war. How easy would it be for Europe to just decide to charge them 5% of revenues generated on their turf? Or maybe 10%? Still, I don’t think such draconian measures would last long. I remain of the opinion that in the United States we live in a “Big Technocracy”. Despite Trump being the most powerful president of my 40 year lifetime - I do NOT think he is more powerful than the collection of tech companies who call America home.

Regarding Amazon specifically. If there is any single company ON EARTH that should be considered “anti-fragile” → that business is Amazon. I would feel more comfortable holding Amazon shares than any other through a zombie apocalypse. It’s business is extremely diversified. I put in my own figures for “True Margins” - which are simply what my best guess is they would average out to long term if Amazon made their objective maximizing operating margin - something they will never do as long as they are optimistic about growth opportunities.

Amazon - if it wanted - could have the most profitable advertising business on Earth. Their advertising business is defended by the 500 million + square feet of warehouse space. And a delivery infrastructure (I know you see the vans everywhere!) that is larger than anyone but the US Postal Service (and more advanced to boot).

The pandemic showed us that - in fact - ALL big tech companies are anti-fragile. But Amazon moreso than anyone. Over the course of just three years amazon DOUBLED the size of it’s fulfillment network. Having an ability to marshal and put to work that amount of resources is in itself an incredible skillset. Over the past twenty-four months Amazon has shown us that it can manage doubling capex from an already world-leading size at the same time as skyrocketing its operating margins.

Amazon describes itself as the “World’s most customer-centric company”. They are obviously deserved of this title. It’s easy to forget that Amazon essentially invented customer ratings, free no-questions asked returns, and reliable shipping. Before Amazon (I’m old enough to remember), if you ordered something on the internet or in a catalogue it would probably take “between 1 and 2 weeks” to arrive. If it didn’t arrive you would need to spend hours figuring out who to blame and chase for your money. And don’t you remember retailers being unwilling to accept returns if you lost the paper receipt?

Amazon turned their own experience selling retail products into Third Party Services - which I believe will someday be worth $1T.

It turned its own experience scaling digital infrastructure into AWS - which is arguably already worth $1T.

There is an enormous amount of optionality embedded in Amazon’s business. IF any company is going to crack healthcare in this country it is likely to be Amazon. Why not deliver medicine as a service to all prime members? Once autonomous vehicles are a thing why not make a Prime subscription option that comes with autonomous vehicles? Is anyone better positioned than Amazon to invest in automation and robotics? They employ some 1.5 million people - their payroll is now larger than any company on Earth except Walmart. That represents a lot of incentive to figure out how to automate. And what about satellites? Amazon wants to compete with SpaceX via their Kuiper business. Who else has a prayer? Why not include high speed internet with a Prime membership? My personal expectation is that SpaceX and Amazon will become an oligopoly that mops the floor with the big telecom companies. That’s a lot of revenue for the taking.

Below is an excerpt from Andy Jassy’s letter that speaks to optionality:

Fourth, speed disproportionately matters for every business, in every industry, at all times. It’s a false binary to argue that you can move fast or deliver high standards. If you want to be fast, you can be fast, and still be high quality. We’ve done it for many years (though we can still be faster). Speed is a leadership decision. The leadership team has to believe it’s a priority, reinforce it constantly, organize and remove structural barriers, and build in modular ways that enable pace. But, speed does not happen unless the entire company and culture embrace it. We have this persistent feeling, throughout the company and in every business in which we operate, that there are closing windows all around us. We operate in fiercely competitive market segments, with highly talented, well-funded, ambitious companies at every turn. Customers are always looking for something better. We spend a lot of time identifying how to unlock these experiences for them as quickly as possible, and know if we don’t, somebody else will.

This type of thinking is rare, particularly for large organizations. This type of thinking is completely extinct in Europe, Japan, and virtually everywhere except for the United States and China. Thinking like this is - literally - illegal in most parts of the world. How so? Because speed is impossible where labor laws are strict. Therefore - no one even practices the “mental muscle” of moving fast.

Economic chaos, the disruption caused by AI (e.g. to the jobs market), China taking an indominable lead in manufacturing scale AND prowess - there are many tectonic shifts underway that will destroy companies who are not nimble - but they will be extremely beneficial to companies who are. Amazon’s AI revenue is growing at over 100% Year on Year…

A few interesting tidbits to end on

-Amazon Prime started with unlimited, free, two-day delivery for 1 million products. It’s now grown to 300 million items and has 10s of millions available for same day delivery.

-Project Kuiper (the satellites) first target is the 400-500 million people around the world who don’t have access to broadband connectivity. Amazon expects the business to contribute meaningful operating income and be a high ROIC deployment.

-Lastly, this re-print from the 1997 letter by Bezos to shareholders showing just how far Amazon has come. 285,000 square feet of warehouse space in 1997 has turned into 500 million and growing!

Share

Isn’t it beautiful?

Moving forward I will be posting screenshots of the different indexes I write about on the 01Core_Indexes tab to make it easy to follow along and/or recreate the portfolios yourself (or modify them as you desire, etc). The second portfolio I’ve created will go live this week as soon as the funds clear. I’ll talk a little about the concept behind it below.

Opportunistic Secular Tailwinds

One of the equity research activities I do on a regular basis is use stock screeners. My four favorite screening criteria are:

1. Delete any company whose revenues haven’t paced with inflation (these companies are terminal zeros)

2. Select FOR companies that have rock solid balance sheets (various ways to accomplish this)

3. Select FOR companies whose forward multiples are lower than their historical EPS CAGR (I usually look at 3/5/10 year and want companies that have exceeded forward multiple in terms of EPS CAGR over multiple periods)

4. Select FOR steady and/or increasing operating margins

The criteria above are designed to avoid:

1. Companies whose terminal value is in question (those not keeping up with inflation)

2. Those who could suffer through economic troubles (bad balance sheets)

3. Those who are over valued (criteria #3)

4. Those who don’t have secular tailwinds (the most obvious sign of a secular tailwind is steadily increasing margins, and if your margins are decreasing then you obviously do NOT have secular tailwinds)

My recent obsession with homebuilders and semiconductor capital equipment companies (henceforth: semicap) is because they are the ONLY two groups of companies where most companies in the group meet the #3 criteria above (as well as all of the others).

Other companies that make it into the group are members of Big Tech, Nvidia, Arista Networks - some of the companies that you would expect to see there. But outside of the obvious candidates who all have idiosyncratic reasons for matching the criteria, only semicap and builders - as a group - meet the four criteria above.

I’ve explained in other posts on 01Core_XBytes why this is this case and why they should continue to be attractive investments.

Essentially - despite having massive secular tailwinds - both groups are cyclical - which results in pretty regular massive drawdowns like the ones we’re in now - which is unattractive to investors - which makes their average multiple from cycle to cycle far less than it would be were they to be given credit for their long term earnings growth trajectory.

Applied Materials is the perfect example. It goes through cycles where earnings are depressed, but over time the trend is up and to the right

In essence, you are being paid with reliable long-term share price outperformance - in return for being willing to eat the volatility.

Portfolio strategy

The idea with this index is to pick 3 or 4 themes that are currently attractive on a valuation basis and which also benefit from strong secular tailwinds - and allocate the entire portfolio equal weighted across the themes. Among all themes and stocks I look at (which is basically the S&P 500 /Nasdaq 100 and global tech behemoths) - the three I find most attractive right now are:

1. Big Tech

2. Semicap

3. Builders

This index will allocate 33.3% to each theme. The sub weightings within the themes are derived from a combination of:

1. Current valuation relative to historical

2. Market cap

3. Uniqueness of market position (moat)

The portfolio will be revisited at the end of 365 days so that IF any adjustments to the themes are made that require realizing gains they will be long term. Here is the starting portfolio.

Imho - it is an incredibly attractive time to begin investing in this index. The weighted drawdown of the entire portfolio is 35%. The forward multiple of the portfolio is lower than the historical EPS CAGR - which is an absolutely WILD fact given how many companies are in the portfolio and how solid their long term EPS growth has been.

For reference, the average company in the S&P typically trades around 18X forward with a historical EPS CAGR of 9% - meaning the FPE / CAGR is 2!!! This compared to being way less than 1 for our index above (see below).

This next table shows the historical EPS CAGR (10yr, 5yr), the forward PE, and a new metric I’ve created for purposes of looking at the portfolio on a look through basis - which is the weighted 10yr and 5yr EPS CAGR and the weighted FPE to calculate the weighted FPE/Historical CAGR (I use the average of the 5yr and 10yr CAGR).

Remember, you want the ratio of the [forward PE] divided by [historical EPS CAGR] to be as low as possible. This is analogous to saying you want the [price you pay] to be as low as possible relative to the [earnings you get].

The ratio for the S&P 500 historically is approximately 2. The ratio for this portfolio is .66 as of today. TBF - I believe estimates for this basket are likely to drop further than estimates for the S&P500 over the next year - but I also expect that they will recover faster thereafter.

Frankly - I will be stunned if this portfolio doesn’t mog the S&P 500 over the next 2-3 years.

Starting equity value: $10,000 on April 4th, 2025 (plus the cash/QQQ which I will remove and hence am not counting).

• 26EC means consensus earnings estimates for 2026

• 26ET means trough earnings estimate for 2026

• FPEC means forward PE based on consensus

• FPET means forward PE based on trough

• FPECW means the weighted impact on a look through basis of the consensus PE

• FPETW means the weighted impact on a look through basis of the trough PE

• 25EPS = current consensus estimates for 2025

• ImpGR = implied growth rate based on consensus estimates from 25 to 26

• Weighted growth rate is the growth rate multiplied by portfolio weighting

Total result is a portfolio trading at 22.32X 2026 consensus earnings and 27.45X trough 2026 earnings. Fwiw, both of these figures are CERTAIN to be lower than in reality if the tariffs remain in place in anything remotely similar to their current form. Still - as a basket the portfolio does not look expensive relative to the growth prospects (18% EPS growth based on consensus expectations).

This truly is one of the most beautiful portfolios that has ever existed. It is basically a conglomeration of the most attractive companies on Earth - which happen to almost all be operating in what I refer to as synthetic monopolies. I wrote a post about what a synthetic monopoly is here:

I’ve also explained why Big Tech is the best way to play the continued proliferation of AI due to it being a synthetic monopoly here:

If you considered the entirety of this index as a single business with each of the companies representing divisions, then we would have the following market shares:

1. Digital advertising: 66.4%

2. Computer operating systems: 91.1%

3. Mobile operating systems: 99.7%

4. Cyber Security: 17.4%

5. Logic Chip Design: 100%

6. Semiconductor capital equipment: 81.8%

7. Cloud Computing: 67.5%

8. Ecommerce: 43.3% (US)

9. Productivity Software (Office 365, etc): 87.2%

Some of the figures above like Cloud Computing are understated. There are really only 3 clouds in the sense that most people think of them - Azure, GCP, and AWS - only these companies are capable of acting as a “Base-Layer” (term I described in my post on The Cloud. So an alternative market share is 100% if we restrict “Cloud” to mean “Base-Layer”.

Similarly, if you were to use profit instead of revenue to calculate market share in digital advertising then the market share of this portfolio would be on the order of 86-87%.

Not only are the market shares above disgusting, they are disgusting in the most attractive business models that humans have ever created - with the largest and still expanding TAMs, with incredible secular tailwinds, and unparalleled optionality for the coming world of AI.

I’ve just wired the funds so hopefully they arrive today and I can get this started and post the beginning screen shot. Pretty wild time to be starting, should be exciting!

Obviously, this will be benchmarked against the QQQ.

Subscribe now

Share

139 bookmarks? Lfg baby!

I also drew some (fair) criticism

Waymo IS probably worth zero if Tesla cracks FSD with its existing suite of cameras. That said, I haven’t seen anything which leads me to believe they will (I say this as someone who has used FSD extensively). My current guess is that by the time Tesla achieves full and scalable autonomy it will need to have a lot more cameras and/or other sensors. The problem is if your autonomy only works in good weather people could get trapped by a sudden burst of rain forcing the cars to immediately pull over. I just don’t see how this gets solved without adding redundancy (which very well might mean just adding more cameras instead of LiDAR).

Gambrinus points out any analysis conducted this way will leave the last bit undervalued. He’s not wrong, though I don’t think my valuation estimates for the different businesses are far off (more on that below). Valuations today in private markets are insane, but even public market comps are high and to me indicate that a sum of the parts (SOTP) method ends up with Google looking cheap no matter how you dice it.

I’ll also say that while SOTP valuations are fun, they are frequently nothing more than mental masturbation. Sure there are people out there who think “it’s healthy”, but is it really helping you achieve any tangible goals? The purpose of any valuation is singular: help the investor make good decisions. Insofar as a SOTP makes that happen I think they add value. When it comes to Google I actually do think the SOTP is useful. But not because I believe it’s the best way to come up with a fair value for the equity (Sergey and Larry control the business and have zero interest in spinning off anything - they would only do so if forced). I think it’s useful mostly because it helps paint a picture of how diversified and strong Google really is.

In this post I’m going to talk about how I view Google.

What are we starting with?

I think this is how most people view Google today:

It is a formidable company but one that is being attacked on all sides, including by the company destined to become the next Big Tech Consumer business - OpenAI - who is also destined to render Search a relic of the past.

Another key part of this image is the fact that all of Google’s resources are fixed (a pile of gold coins). Rather than being “renewable resources” with lots of optionality, they are perceived as businesses whose best hope for generating profit is to defend their existing turf - which is a fixed size. Google is perceived as a company who will respond to disruption by spending all of their effort guarding their existing treasure (very similar to how Microsoft was perceived pre-Satya).

Quick aside - note how the OpenAI dragon is actually focused on Search? Idk if the AI was smart enough to do that but damn…anyway

I think a more accurate view of Google today is the following:

Notice the difference? There is no other dragon… The threats are soldiers - manageable.

Or better yet, this - the threats are so small as not to be noticeable compared to the opportunity to invest its profits.

I believe Google has a collection of assets that are:

1. Extremely profitable, and whose true profit generating capacity is not even understood by Google

2. Still in growing markets, with TAMs that are dramatically under estimated

3. NOT under attack by existential threats

4. Unmatched in terms of future optionality

In the image the dragon is using profits from its assets to invest in a new business venture - it is focusing its time on leveraging its resources to become even bigger! Not just to protect an existing asset base that is under threat…

I think there are two historical analogies that are apropos today.

Microsoft missing all the waves; Meta getting pounded by TikTok

Here’s a chart of Microsoft stock from 2006 to 2014

Microsoft traded sideways while it’s multiple compressed for a variety of reasons:

• Stagnant Revenue Growth Perception: While Microsoft’s earnings from Windows and Office remained robust, revenue growth slowed compared to the hyper-growth of Apple and later Google. The PC market matured, and shipments declined as tablets and smartphones took over. Investors saw little evidence of Microsoft replacing this with new revenue streams, leading to a lower growth premium.

• Missed Mobile Revolution: The iPhone’s success highlighted Microsoft’s inability to lead in mobile, a market that exploded in the late 2000s and early 2010s. Windows Phone’s failure to gain meaningful share (peaking at just over 3% globally) meant Microsoft wasn’t participating in the fastest-growing segment of tech, dampening enthusiasm for its stock.

• Competitive Pressure: Beyond Apple, competitors like Amazon (with AWS starting in 2006) and Google (with Android and search dominance) were capturing the future—cloud computing and mobile advertising—while Microsoft lagged. Its Azure cloud platform didn’t launch until 2010 and took years to scale, leaving it playing catch-up.

• Market Sentiment and Risk Aversion: Post-2008 financial crisis, investors favored companies with clear growth narratives. Apple’s iPhone-driven ascent made it a darling, while Microsoft’s steady but unglamorous profits from enterprise software and legacy products didn’t inspire the same excitement. Its P/E, which had been in the 20s in the early 2000s, fell into the low teens by 2012, reflecting a "value stock" rather than "growth stock" status.

• Leadership Doubts: Ballmer’s tenure became a lightning rod for criticism. His dismissal of the iPhone, combined with costly missteps like the $8.5 billion Skype acquisition in 2011 (seen as overpriced at the time), fueled a narrative of poor capital allocation and lack of vision. This eroded confidence in Microsoft’s ability to deliver outsized returns, compressing its multiple further.

From 2007 to 2012, Microsoft’s stock hovered between roughly $20 and $30, while its earnings per share grew modestly (from about $1.42 in 2007 to $2.00 in 2012). Its P/E dropped from around 20 to below 15. The S&P 500, despite the 2008 crash, began recovering by 2009, but Microsoft lagged, underscoring its “specific challenges”. By 2012, its market cap was overtaken not just by Apple but also briefly by others, like ExxonMobil, highlighting its fall from grace.

The narrative shifted slightly toward the end of this period as Microsoft showed early signs of adapting—Windows 8 and Surface launched in 2012, and Azure gained traction—but these were seen as too little, too late. It wasn’t until Satya Nadella took over in 2014, refocusing the company on cloud and subscriptions, that the stock broke out of its funk.

With the benefit of hindsight it seems obvious that Microsoft never should have traded at a low teens multiple. With hindsight it seems obvious that owning the operating system for PCs was an asset whose TAM would grow even if unit sales of devices stopped growing. With hindsight, it seems obvious that Microsoft was well-positioned to leverage its customer base to take share in the next transformative technology paradigm.

The Meta TikTok battle is also apropos. TikTok IS a competitor - that is obvious. Were it to disappear tomorrow Meta would gain lots of advertising dollars. One thing substituting for another is the literal definition of competition. But…TikTok essentially created a new addressable market that Meta wasn’t going after. That new market (short form video) ended up being huge. Meta and TikTok are both growing simultaneously at the expense of other ways people were spending free time (e.g. mobile gaming). Meta was able to take advantage of short-form because of its distribution. Having 3.5B daily customers gives you some breathing room to respond to new threats. For perspective, Meta was sent through a 70% drawdown b/c TikTok was growing from 500m to 1B+ users. Despite that wild user growth Meta’s users never dropped - time spent on their family of apps never dropped - and their revenues have mooned over the period (along with margins). We had a situation that looks like this:

Yes - market share dropped - but revenues and profits increased because they were still able to lock down a piece of a larger market.

Take a moment to think about where we would be if OpenAI hadn’t come along. I explained in a post in early 2023 why Google had never launched a product similar to ChatGPT. They already had the technology - but they were too afraid to release something that could “hallucinate” “lie” “be racist” etc…And when you have been CAGRing EPS 20%+ for 20 years why bother taking big risks?

OpenAI actually made Google - imo - a MORE VALUABLE COMPANY. Why? The value of a company is equal to the present value of its expected future cash flows. OpenAI released the kraken. It gave Google the cover to compete and enter new markets without the fear of becoming a political lightning rod and bringing down the gavel of government on its head. It accelerated the arrival of what will end up being massive new sources of cash flow growth. I think the before and after analogy for Google and OpenAI is more along the lines of this.

What’s more, the existence of OpenAI is probably a good thing for Google moving forward as well. Why? OpenAI and other LLMs will be forced to take risks that Google doesn’t want to take (for example, enabling higher levels of personalization in LLM response/behavior, more flexibility over imagegen, etc). If not for companies like OpenAI Google would not take these risks on its own. However, if other companies take the risk of public perception issues due to new features first, then Google is less worried about taking heat by simply releasing something someone else has already created.

At the end of the day they will always be able to rely on distribution to catch up, so they can strategically sit “behind the front line” - but not too far.

The Search Elephant

I’ll be the first to admit that LLMs have replaced Search for me almost entirely. My rough estimate is that I use Search 95% less than I did before LLMs came out. This is obviously a bad thing for Google. For decades “Google” has been a verb synonymous with the activity of “finding information” or “learning”. I remember many years ago asking my grandmother to quiz me about random historical facts - which I would then look up on Google and provide the answer for. Her mind was blown. Google was the company associated with making information free and available to everyone.

I’ll also note that the use of Google as a “verb” is all but dead in my own life and is absolutely 100% dead among Gen Z. For younger members of society Google is not only being replaced by ChatGPT - but also by TikTok. I never tell anyone to “Google” anything anymore. It is not remotely as good of an experience for learning about the world as using ChatGPT or LLMs - even with its new embedded AI responses (though Gemini 2.5 is absolutely top notch).

Interestingly, I believe TikTok, Amazon, and Meta are all bigger competitors to Search (and bigger “threats”) than OpenAI. Why? Because the things people search on Amazon (products), TikTok (the place GenZ goes to research restaurants or look for travel tips), and Meta (similar to TikTok but also now has What’sApp being used to communicate with businesses about their products, etc) - are the things that would generate Google ad revenue.

Ironically, all of the real threats to Google’s search REVENUE are the ones no one talks about. The part of Search that is most under threat from OpenAI (e.g. market research) never generated Google money anyway.

The real threat to search medium→long term is Agentic AI and new operating systems (more on that below). Rather than a human searching for a pair of sneakers to buy - they just ask the AI to do it and show them the top 3 choices.

Today you use ChatGPT to plan a vacation to Italy but then once you’re done you go right to Google to purchase the tickets, book the hotels, etc. Once ChatGPT is plugged into the internet and can manage the process of filtering through hotel options, rental car options, etc - it would become a serious threat.

But this use-case is something Google (and Apple) are in a far better position to provide…

The everything app

At the end of the day people will use as few apps as they can to accomplish their objectives. This is essentially a law of physics. The long-term end state of Agentic AI (what LLMs will evolve into) is EXTREMELY OBVIOUS.

It will be a personalized Agent - with an Avatar you select just like picking out a character’s appearance in a video game - with a personality that adjusts to yours over time (adjusts how flirty it is, what political views it has, etc). It will be connected to the operating system of your phone - and your interaction with apps will probably drop 50% because you will just ask the AI to do things. It will handle booking an Uber. It will handle ordering AND purchasing groceries and having them delivered autonomously. It will respond to work emails on your behalf. It will share screens with you and be tech support, copy-writing help, strategy whiteboarding partner, and so on…It will be intimately familiar with your personal and professional life…

It is absolutely hysterical to me - and I mean it literally makes me laugh out loud - every time I hear some idiot proclaim that they don’t want to give the AI their “personal information”.

Meanwhile they’re uploading pictures of what they had for lunch along with a location timestamp to Instagram just after posting a picture of their baby’s cute new outfit. Did I mention that the phone carriers literally know where they and their family are 24/7?

haahhahahahahhahhaahhahahahahahhahahaahaahahhahaha

The value people will derive from having AI be intimately familiar with their lives is so enormous we all will do it. We’ll share our darkest secrets with it - our biometric data with it - literally everything… And it will use that information to transform not only our health but also our happiness (wrote a little about that here).

The point of all this is that ONLY a company who owns an operating system can deliver this killer experience. You need control over the OS to…

1. Optimize latency

2. Offer “agency” to the AI securely

3. Force other apps to give the AI access

4. Make the entire experience seamless

Apple used “privacy” to cut off Meta’s access to app purchase data. It’s an absurd argument but it worked because our politicians and regulators are morons.

Given that precedent, how much easier will it be for Apple and Google to argue that they “can’t” provide other companies with access to the operating system for security reasons?

Piece of cake. “No Mr. Government, I’m sorry, giving ChatGPT access to people’s Wells Fargo account is just a step too far, there’s no way to do it securely”.

But there are other issues too.

Adobe obviously will not be thrilled about ChatGPT becoming the main way people use it. Remember - the future of graphic design is not sitting there dragging and dropping with a mouse - the future is talking to the AI, telling it what you want, and having it use the tool on your behalf.

Similarly, Bank of America will not want to let OpenAI be how people send money on Zelle (in the future, we’ll just ask our assistant to Zelle someone when we want to send money).

Effectively, Agents will become a new operating system that will operate in parallel to iOS or Android - but only Agents provided by the owner of the operating system will be capable of delivering an “everything app”. And even if someone else came along who could deliver a similar experience, Apple and Google will just shout “Security” and then knee-cap the functionality of competitors.

What’s funny is that both Sam Altman and Mark Zuckerberg AGREE WITH THIS AND SAY SO PUBLICLY ALL THE TIME! I feel like I’m screaming into the void when I point this out.

Sam’s vision is for OpenAI to offer the service that houses your memories, digital experience history, stores your logins, etc. He explicitly stated this in his recent interview with Ben Thompson. When I hear him talk it’s clear he wants to make their product a “4th component of the GUI” we use to interact with the digital world.

He is also known to be interested in (if not working on) an “AI First” device and alluded to this a few days ago in a response to a post on X

Mark Zuckerberg is trying to build an operating system for the next form factor (glasses) for interacting with the digital (or augmented) world - because he knows Apple and Google are who has the real control by owning the OS.

So again, we have a situation where people perceive OpenAI to be the biggest threat to Search - but in reality the bigger threat is actually the companies who are likely capable of competing with their operating system - not the “search application”.

To be fair - OpenAI just closed a $40B funding round and now has 500M+ users and as of yesterday was gaining users at the rate of 1 million per hour. This capital will probably be equivalent to $80B in the hands of Google or any other member of Big Tech (seriously, I think OpenAI is probably that much better at allocating capital - which is easier given their size).

Effectively, this new funding round gives OpenAI a war-chest that is on par with a Tech Giant. Combine that with 500 million+ users and you have what is obviously the most legitimate threat to Big Tech in history (outside of Big Tech being a threat to other members of Big Tech - which it is - but has never mattered because the TAMs keep growing so fast). To be clear, I’m saying OpenAI is a far bigger threat to Big Tech than TikTok ever was…

I believe we’ll find out that OpenAI is launching a smartphone device sometime late next year or during 2027. Why then? Because my view since the beginning of 2023 remains unchanged that delivering Agentic AI on a phone won’t be possible until late 2027 or 2028 given design cycles. You just can’t build a new device any faster when you are working with hardware that needs a fundamental update (chips) and not just software - and no one was even trying to work on AI first hardware until after ChatGPT came out.

This is the real threat to Search long-term - but it is a longer-term threat. And it is actually MORE of a threat to Apple than it is to Google - by far - because if Apple loses a smartphone customer it loses everything. Google has far less to lose given how much more diversified their business is. And - we just might see a super interesting twist where OpenAI builds on Android and becomes a Google Cloud customer…But again - these are longer-term predictions.

Remember, Amazon and Microsoft had a go at launching a smartphone. Meta has been pouring tens of billions of dollars per year into their new OS contender. OpenAI certainly has a shot at creating a new AI-first OS, but it’s a tall ask and the incumbency advantage is enormous.

To summarize the key points here:

1. Search is threatened by Agentic AI - not LLMs in current form

2. The killer “Agentic AI” app requires ownership of an OS

3. Start panicking about Search dying once new OS’s come out

Let’s talk about TAM

Everyone forgets that once upon a time phones were cheap. The Razor was the expensive version all the cool kids wanted. None of them cost half what a smartphone costs today. Only business people and rich people had Blackberry’s.

Smartphones added so much additional functionality that people adjusted their expectations around what a phone should cost. In fact, they offered SO MUCH value that carriers started subsidizing the cost en-masse - with zero interest financing - using them as a loss leader to get new subscribers.

Today some number of people are paying $20/mo for ChatGPT, but most people are not. My view is that five years from now EVERYONE will be paying big money for agents OR suffering through tons of ads if they want to use them for free. Maybe the cost of Agents is subsidized through Verizon and AT&T? Idk. But one way or another the providers of Agents will get paid for the value they deliver.

I am already paying $100+ for different AI apps per month. They all could be replaced by a single AI app with agency. I’ve talked a lot about how much value is delivered by agency, including in this post from January of 2023:

I won’t belabor the point, suffice it to say I think it’s reasonable to assume that every human in the developed world will be spending at least $20/month on Agents, with the top 10% spending $500+, and with somewhere around half spending $100+. This gives us the following

That’s $1.94Trillion of revenue opportunity to be split amongst Apple, Google, Microsoft, and whoever else can grab a slice. Microsoft and Meta probably duke it out in the small business space, with Apple and Google carving out nice chunks among their own user bases (with Apple gaining the lion’s share - should it execute well - because of it’s customer base essentially being the 15% wealthiest humans). For reference, TTM revenues for Apple, Microsoft and Google combined are around $1T.

These revenues probably start to kick-off in a meaningful way in 2028/2029. My belief is that over the next 10 years people will once again be surprised at the performance of big tech - and for the same reason that they have been surprised the last ten years - the TAM just keeps growing…

How big could the cloud get?

Funny you should ask. I wrote a post on this pre-ChatGPT in April 2022.

Obviously I have upped my expectations by orders of magnitude now that Agentic AI is right around the corner.

The TAM for the cloud was already heading toward $1T+ in annual revenues by 2035 before AGI. In a post-AGI world it’s impossible to estimate, but I can guarantee that no one’s estimates are bullish enough long term. Why? Because people estimate by extrapolating existing trends. The largest user of compute in a post-AGI world will not be humans - it will be computers. Until we see how much compute AI Agents use when they’re interacting with each other autonomously solving problems - we won’t have anything relevant to extrapolate insofar as predicting the TAM 10 years out.

Again - the takeaway is that “Big Cloud” (Google, Microsoft, Amazon) or what I refer to as the “Base-Layer” in prior posts - will have a TAM that keeps on surprising to the upside.

YouTube

Google’s other big business is YouTube. My estimate is that YouTube should trade at a premium to Netflix on a per-dollar of revenue basis. It is a better business model with a stronger moat and larger TAM. YouTube (YT) generates revenue from both ads and subscriptions. Google doesn’t breakout revenue for the subscriptions, but according to Google there were 125 million YT premium subscribers as of March 2025. At $10 per month this equates to $15B per year. I’m going to call it $13.2B per year because that’s the estimate ChatGPT came up with and it’s a little more conservative. Combining that with the publicly disclosed ad revenue gives total revenue for YouTube of $49B. I find it hard to imagine that YouTube is worth less than Netflix on a comparative basis - the moat seems stronger plus I think YT has more utility as a training source for new AI applications/more optionality. If we take YT revenue and divide it by Netflix revenue and then multiply by Netflix’ valuation then we get $49B/$39B = 1.27 X $400B = $506B of value for YT.

You can argue that Netflix is overvalued - that’s fine, I agree. But I also believe YT has far more optionality than Netflix and a more attractive business model with far less terminal value risk - so if Netflix is overvalued YT could still arrive at the same place with an imo well-deserved premium.

I think one of the most interesting things we’re going to learn in the next two years is what type of data is useful to train AGI - both AGI that is embodied physically and that which only exists in cyberspace. YouTube seems like it will be THE single most useful resource for this purpose - time will tell.

Okay, let’s value google SOTP style

First up the balance sheet

A company like Google does not need to carry a net cash balance. Hence what we can take off the balance sheet is: Cash + Marketable Securities + Non-marketable Securities (their VC portfolio, non-control assets, etc). I’ll net out the debt - which gives us the following:

1. $122.75 billion - they could kick this out as a dividend to investors without impacting the business

Waymo

Waymo was valued at $45B post-money in a $5.6B funding round last October. It was the first funding round since $2.5B in June of 2021 and $2.3B in March of 2020. The larger amount of capital is probably because it has cracked the nut on scaling a profitable ride-hailing service and is now ready to grow exponentially.

Rides per week went up 15X from May 2023 to October 2024. As of March the number of rides per week had already increased to over 200,000. Today Waymo is in Phoenix, San Francisco, and LA. By the end of 2026 it has already announced its intentions to be in the following cities

There is a fascinating and ongoing debate about the different approaches to autonomy taken by Waymo and Tesla. Relevant to coming up with a valuation for Waymo is considering the differences in potential business models. On one hand you have ride-hailing - a business model that is pretty well understood. Waymo’s ride-hailing service could operate like Uber without the share of funds needed to pay the driver, but with more expensive vehicles and back-end tech support.

The second potential business model is licensing their technology to auto OEMs who want to sell autonomous vehicles. Every month there are new rumors about car companies licensing Tesla’s technology. While I am as long-term bullish on Tesla as anyone, I do not believe their current sensor suite can accomplish autonomy. Hard rain, sun glare, and snow all cause FSD to glitch. I don’t see how you solve this problem without adding new cameras or sensors to create more redundancy. We’ll find out when their program launches in Austin. But - until Auto OEMs see Tesla’s driverless technology working at scale I don’t believe we’ll hear of any concrete plans to integrate the tech into their production lines. On the other hand Waymo has officially cracked the nut. Today GM could start retrofitting their models with Waymo and sell a vehicle that could operate with full autonomy in any of the cities where Waymo operates. I doubt any auto companies were even aware of the step-change in daily rides Waymo just experienced. But I do think that the big OEMs are now probably in active talks with Waymo about integrating their tech into production lines. My guess is we see our first signed deal by the end of this year, with at scale production scheduled for end of 2027/early 2028.

A third business model could emerge where Waymo becomes a direct Uber competitor - using their app not only for their own vehicles, but for other autonomous vehicles to plug into. For example, a new business model I expect to emerge near the end of this decade will be consumers purchasing autonomous vehicles so that they can still have total control over it, but then choosing to rent it out when it is available and not being used to earn some extra money and subsidize the cost of the vehicle. Or alternatively, they just want to have the ability to not focus on driving, or to share a single car in a family instead of needing multiple cars (pretty easy to do if the car drives itself). Even if they purchase an autonomous vehicle from GM, it will probably need to run on Waymo software.

I can only envision one scenario where Waymo is worth less than Uber by 2030 - and that is because it becomes obvious that Tesla is going to mop the floor with everyone else. Elsewise, it seems obvious that Waymo will overtake Uber in value because its business model is far superior.

Deepwater - famous for being bullish (it’s worked out very well for them historically) believes Waymo has three potential valuations circa 2030 (Share refers to % market share of ride-hailing):

I don’t see any world where they don’t solve autonomy given they’re now doing 200k rides per week and growing quickly. If they do hit a $0 valuation it will be because of Tesla. Still, until Tesla proves its hardware is capable of autonomy, investors will not see it as a threat to Waymo. Net-net, it seems very reasonable to value Waymo at $50B - which is just $5B more than their latest funding round. I bumped it up a bit because of the recent progress.

We now have the following for our sum of the parts (SOTP):

1. Cash + Investments: $122.75 billion

2. Waymo: $50 billion X 70% ownership = $35B

Youtube we already did, which gives us:

1. Cash + Investments: $122.75 billion

2. Waymo: $35 billion

3. YouTube: $506 billion

Android + Hardware

I’m going to lump other non-cloud subs into this category for ease. The big revenue items are Hardware sales and Play Store. Together with other non-cloud subs we have $26,800, with hardware probably around $12B of that and PlayStore probably around $10B (other subs/revenue hence at $4,800). According to documents from the Epic Games lawsuit against Google, the PlayStore has operating margins of ~70%. My guess is the operating margins on the other subs piece of the businesses are probably pretty good, but I’ll call them 50%. Hardware is probably not generating much if any profit. Most of the value from Android comes in the form of ecosystem control and being able to put whatever apps Google wants as defaults on the operating system. Interestingly, Pixel market share jumped at the end of last year with the new model launch:

That little blue line that jumped away from the rest of the laggards is Pixel. I think Android is the most under monetized asset on the planet. Android has 70%+ market share of mobile devices and somewhere between 3.5B and 4B active devices. Android is how Google will distribute AGI in the future. If we assume 70% margins on the PlayStore, 0% margins on devices, and 50% margins on other subs we get $9.4B in operating profit. Putting a 25 multiple on that gives us $235B. This seems very reasonable if not conservative given the optionality that comes with owning Android.

This gives us the following:

1. Cash + Investments: $122.75 billion

2. Waymo: $50 billion

3. YouTube: $506 billion

4. Android + Hardware: $235B

Cloud

Cloud includes GCP and products like Workspace. It grew 30% last year and it would have grown faster if they weren’t resource constrained. Margins are rocketing. Microsoft trades at 9X NTM revenue. Microsoft is one of the most profitable companies in the world with 35% net income margin, but it is also growing at less than half the pace of Google’s cloud business. If Cloud were valued at 9X NTM revenue we get a valuation of $503B. Alternatively if we assume 25% steady state margins and put a 30 multiple on it (seems conservative given growth rate) we come up with $420B. Let’s go with that to be conservative. Now we have the following:

1. Cash + Investments: $122.75 billion

2. Waymo: $35 billion

3. YouTube: $506 billion

4. Android + Hardware: $235B

5. Cloud: $420B

Drum roll please…

Add it all up and we’re at $1.319T. Google’s market value is only $1.925T. This leaves Search + Network being worth $606B. Mind you, Search + Network will generate $96B of operating profit next year (pre-tax). That implies Search + Network are trading at a multiple of - drum roll pleasssssssssssssssssse → 6.3X.

Alternatively, if you put a 15 multiple on Search operating profit you get $1.44T of value → combine that with the rest of it and we get an enterprise value of $2.759T, about 43% upside from here.

Closing thoughts

I don’t really think an SOTP is useful for purposes of setting a price target. What it is useful for is realizing how diversified Google is. Whatever assumptions you want to use, it seems hard to envision a scenario where Search is being attributed much more than half of their market cap - and a proportion that is shrinking with time.

If we reversed the activity we just did, put a 15X multiple on Search operating profits to get a $1.44T valuation - then we come up with only ~$485B of value for Cloud, Waymo, YouTube, DeepMind, Android + Hardware, and don’t forget the $123B of net cash. That seems ridiculous.

We haven’t even discussed the inherent optionality in things like the fact that Google has better data center technology than anyone on Earth (probably by a long shot). They can “deliver intelligence” more cheaply than anyone. They have the most advanced in-house chip business for AI.

It really is wild to me that people perceive Google as being a potential loser from AI. ANYONE could be a loser…But Google does not deserve some sort of special discount relative to other companies because it is more susceptible. If anything - it is less susceptible and deserves a premium.

Ben Thompson in bold.

I haven’t been watching all of Sam’s interviews lately, but this one is the first time where I hear him explicitly state that the model isn’t the moat. He caveats that by saying sustaining the most cutting edge model is a moat - if he had to take a stab at where “market cap” gets created I would bet this is something like what he would expect 10 years from now.

Today and for the next 5 years or so we’re likely to have a situation more like this

Note the green line for “cutting edge research” is essentially at zero. Today value is mostly accruing to the infrastructure layer. This line stood out to me

OpenAI and Microsoft parted ways because Satya and Sam started disagreeing on capital allocation. While I suspect they agree on where the market value is going to be created - I think Sam still wants to always have the leading edge model regardless of cost, both as a call option on super intelligence / fast takeoff post-AGI, but mostly because it’s part of the mission and culture of OpenAI, one that he built and shares.

If you’ve been reading my blog over the past two years you’ll know I’ve been predicting since January 2023 that the killer application for generative AI is personalized agents who are intimately familiar with our lives and connected to the operating systems and apps on our phones and computers. The list of people who have hinted at this as the future now include:

Sam, Zuck, Demis, Sundar, and many other lower profile individuals.

One of the key insights I feel like I’ve had thinking through how everything plays out with personalized Agents is that the companies who own operating systems are the ONLY companies who can deliver this vision. There are numerous reasons:

1. Security is a big deal. Apple already uses “privacy” as a reason to block certain types of data flowing through to app owners. How much more dangerous might an AI be that can control your OS and open apps (e.g. bank account) on its own?

2. The AI needs to be able to talk to all apps, the settings, etc - will all apps be happy letting just anyone (e.g. OpenAI) be the new point of contact through which people use them? I doubt it. Only the OS owner will have the negotiating leverage to make apps comply.

3. Business. They can easily come up with a story around why it’s too risky to have personalized AI (e.g. security, social hacking, etc) - so only they can protect consumers.

Now I’m going to play devil’s advocate. My view is that only the OS providers can deliver the killer app. But maybe Sam is right and new devices will pop up, new browsers will get a foothold. If that happens - what enabled it?

What is Apple’s moat?

I’m a PC user and always have been. I hate Macs. Still, I have used an iPhone since the year they came out, and purchase Airpods ~once per year (they break, but I love them and haven’t ever found a replacement). For about 9 months I tried out a new Pixel. I wanted to give it enough time to really learn the interface and adapt before placing judgment. In the end I went back to the iPhone because “it just worked”. There were too many instances of having apps glitch on the Pixel, the software just not seeming to work right, getting weird lags in the interface…This was only 3-4 years ago mind you.

For hardcore Apple users they have an iPad, a Mac and a Phone - and the devices make it extremely convenient to share information between them.

Arguably the strongest part of Apple’s moat is the ecosystem - which is at this point mostly built around apps that work (almost) flawlessly. In second place would be brand. I don’t believe brand can be a sustainable moat for large companies. Looking out I am bullish on the moat of small brands, but bearish on the giant ones - all of them. That’s a topic for a different post.

You could also name things like “Innovation and Design” and “Supply Chain and Scale” - but that’s simply not accurate anymore. There are numerous high quality sleek phones with comparable design and lower costs than iPhones.

In truth, gun to my head, I’d say the second strongest part of Apple’s moat after ecosystem lock-in is good old fashioned inertia. Their install base is huge. The phones work. The apps work. The cost is subsidized by carriers…

What’s the point in switching?

If a new device was going to succeed (phone or other form factor) in penetrating the moat, it would have to break the inertia with something that offered a strong enough valuation proposition to outweigh the ecosystem lock-in.

When we say: “it just works” - what we usually mean is that the user interface designed by a (human) person is very good and intuitive to use. Another example would be how easy it is to connect Turbo Tax to your brokerage account to pull-in financial statements.

What “it just works” really means is given the limitations of the current graphical user interface and [the knowledge base of an average human] - “it just works”.

Now consider the following…

Today if I want to send someone money I can open the Venmo mobile app - the interface is intuitive, it already syncs with my contacts, there’s a button in the middle of the screen that jumps out at you “Pay/Request”. Any idiot can use it. How well it works IS part of Apple’s moat. But what if in the future instead of doing all of the different steps of opening an app, adding a phone number, sending $1 as a test to make sure you have the right contact info, etc…What if instead you simply say to your personalized Agent: “Hey Grok, send $20 to Juan.”

Grok may need to ask some clarifying questions if you have multiple “Juans” in your phone, but outside of that he’ll take care of it. He’ll be able to handle the verification, do the equivalent of a $1 test send, etc.

Start thinking through how you use most (utility) apps on your phone and it will immediately become obvious that IF you had an AI with access to the apps THEN the amount of time you would spend maneuvering through the user interface would fall precipitously - maybe by 75%+?

The Agent becomes another operating system - sitting there side by side with iOS/Android.

Remember, someone doesn’t have to build something capable of competing with Apple’s ecosystem in a vacuum. They get to start with Android - which is open! Google would be perfectly happy to provide the software…

Besides, the cost of “writing code” is dropping precipitously. It will be easier than ever to make sleek apps AND to ensure that they integrate seamlessly with everything.

All of the “integrations” between phones and computers and iPads that make the experience so magical are not necessary if you have an AI that can make everything work seamlessly on its own. Consider the issue of image sharing. It’s wonderful to have photos you take on your phone automatically appear in your iCloud account which is accessible by your computer.

But how easy would it be for an AI on an android phone to create the same automatic, happening-in-the-background experience? It would be trivial.

Before smartphones people mostly spent $200 on a phone (maybe $300 inflation adjusted). Now they regularly shell out >$1,000. What changed is the amount of value being delivered.

ChatGPT revenue - driven by consumer subscriptions, is going to grow from $3.7B last year to >$12B this year. I spend $20/mo on Claude, ChatGPT, more than that on Grok, and also pay for subscriptions for other family members and employees. $20 seems like an absolute steel relative to the value I’m deriving from it.

If you plugged ChatGPT into my GUI? Let it manage the mouse, share my screen, and simply execute on all of the things I’m always asking it how to do (like set up a web domain, format a file, change settings in Shopify, etc)…I would pay hundreds of dollars per month for that.

Remember, step one of training an AI is simply having it do all of the things you need to do all of the time. It won’t be long thereafter until it can simply become an employee or act like an actual personal assistant.

My belief is that ALL humans will pay $20+ (or start getting bombarded with ads) for an AI assistant within 24 month of the assistants getting access to the GUI. I think at least 1 billion humans will pay $100+ per month. The value delivered is just so large that this seems obvious.

Today, the iPhone is the indispensable piece of equipment. But - I think this flips sometime around 2029/2030 (which is 2 years after when I expect personalized autonomous agents to hit smartphones). The new indispensable piece of equipment will be Agents (which aren’t a device!).

With a potential revenue stream of $100B+ per month just from the top tier users, and that much again easily from the rest - we’re talking about a multi-trillion dollar addressable market.

The difference between Agents and what Apple is good at is that Agents will be software driven. Apple is world class (maybe the best) at creating chip architectures to make phones work well, at managing an ecosystem to ensure that apps integrate flawlessly, and melding software and hardware together. But no one would accuse it of being the best at writing applications.

Developing the “personalized Agent” app will be maybe the most challenging software problem of our time. It will need to rapidly customize its personality and communication style to the user. Adapt over time. And - it will need to have a customizable appearance just like in the beginning of role-playing video games where the user gets to choose what their character looks like.

Can Apple build this? Would they even be willing to?

Imagine an iPhone user having a conversation with their “Agent” - the agent is wearing a US flag bikini and ranting about how deranged some member of congress is. The user might love it and post the experience online. Would Apple really be willing to provide a product like this that exposes their brand to potentially catastrophic criticism?

We might end up discovering that it just takes a new company to provide this type of “product” - and this might be the hole in the armor a company like OpenAI or another upstart could pierce.

Remember, this was a game

My real perspective is that personalized agents are the operating system owners’ game to lose. Apple is in a better position than anyone to take advantage of them for the consumer use case (maybe Google is tied?). Their customer base is the 15% wealthiest people on Earth. Much of the hard work around making Agents a pleasant user experience will be clever hardware and software working together to eliminate latency so that communication feels human. Figuring out what to cache on device and off. This is exactly Apple’s skillset.

There is also a good chance that shortly after these things launch - IF they are truly customizable - that people will be held responsible for the actions of their Agents - not the companies providing them.

People will stop blaming Apple or Google for things the agents do as soon as they themselves use an agent and see how quickly the agent adapts to reflect themselves. It’s a cultural shift that will absolutely take place - but it’s hard to say when. Extreme personalization and (likely) avatars being rendered in real-time are both pre-requisites.

Share

Subscribe now

I love this Jeff Bezos quote on long term planning: “Well, certainly, in 10 years many things will evolve; technology will change. Machine-learning technology, in particular, will evolve very significantly over the 10-year time horizon. But I would always encourage people, when they think about 10 years, to ask the question, what won’t change? That’s actually the more important question. You can build strategies around things that will be stable in time. In that 10-year vision, there are a bunch of things at Amazon that are not going to change. One of them, maybe the most important one, is that we will stay customer-obsessed instead of competitor-obsessed. We will work on maintaining that culture.”

Warren Buffett was born on August 30^th, 1930. He is 94 years old. Were you to transport someone from that date through time and plop them down in the year 2025 they would likely suspect that Aliens had visited Earth and kindly shared their advanced technology with us common Homo Sapiens. They would also be shocked at how society itself had changed.

In 1930 the world was in the middle of the Great Depression. Most babies were still born at home, not in hospitals. 6% of White babies and 10% of black babies died before their first birthdays. It was only five years prior that half of US homes had electricity installed. Women had only been voting for ten years. Homosexuality was illegal in most of the United States and Europe. Traveling between New York and London was done by boat. Doctors were regularly featured in ads for cigarettes. The home television had not been invented. World War II hadn’t started. The largest companies in the United States sold commodities. And there were still laws throughout the country segregating whites and blacks.

92 years before 1930 was 1838. Slavery was still legal in the United States. Out of every 1,000 babies born, only 570 would make it to their 5^th birthday. That’s not a typo. Barely half of children made it to five years old.

92 years before that was 1746, thirty years before the United States declared independence from Great Britain. Native Americans still outnumbered immigrants.

I’m paraphrasing what Warren Buffett refers to as the “lifetime” anecdote, which he uses to remind people how quickly progress is made. 276 years sounds like a long time ago, but “three lifetimes” doesn’t.

But you don’t need to look back 92 years to see an enormous amount of change take place. Consider the summer of 2012.

Facebook had just IPOd, and for the second election in a row Obama was being lauded for how expertly he used social media to connect with people. The mainstream media was explicitly positive on the topic of using the “private data” of Facebook users not only to win campaigns in the United States - but also to facilitate regime change all over the world (remember the “Arab Spring”).

The biggest economic threat the world faced was a breakup of the EU. Greek government bond yields had crossed 35% on the ten-year while German two-year bonds had gone negative for the first time. North Korea was the country everyone worried about – not Russia (except for Mitt Romney) or China. Amazon’s AWS revenue was only $1.5 billion (it will be over $120B this year). Oculus, the VR company that Facebook purchased for $2billion to launch its metaverse effort had just filed its incorporating documents (now the founder of Oculus - Palmer Luckey - has already started a new business building weapons, which is worth $30B+ and counting).

Russia was holding its equivalent of Davos - which - most of the world attended. Called the St. Petersburg International Economic Forum, Goldman Sachs, McKinsey, Pepsi, Mercedes-Benz and “everyone who is anyone in the energy business” were there. Mr. Putin delivered a vigorous speech about making Russia more business friendly. Romney was being laughed at and ridiculed for referring to Russia as a geopolitical ally. The Economist closed the article by saying:

Western companies and politicians remain optimistic that relations with Russia could normalize in much the same way as things have between the West and China.

LMFAO - so the goal of US foreign policy in 2012 was to replicate how marvelously things were going with China - in Russia…

How times change…

Cognitive dissonance?

One of the things that frustrates me most are smart people who totally grok that the future is unpredictable with respect to thing A, but who then become completely triggered when you point out that it is also unpredictable with respect to thing B. Climate change and financial markets are the quintessential example.

Most technology and finance bros agree that you cannot predict stock markets. However, most also agree that we can use “science” to forecast that if temperatures rise by 2 degrees then 50 million people will die 75 years from now.

One of the primary data points for predicting future death is [future wealth] - which you would predict largely by forecasting the performance of financial markets.

Do you see the logical error?

Besides, even if you could - with perfect accuracy - predict exactly what the weather was going to be like in the distant future, this is not the same thing as being able to predict what the future itself will be like (e.g. how humans have adapted). In order to predict deaths from climate change you would also have to be able to accurately predict:

1. Economic growth between now and then for purposes of estimating how many resources we will have to address climate change

2. Migratory patterns

3. How resources are allocated between now and then

4. What new technologies will get invented

5. How expensive electricity or other forms of energy would be

6. What infrastructure has been built and where

And the list goes on.

If I could give one gift of a understanding to the entire planet - I’d be very tempted to make that gift the innate comprehension of the inherent unpredictability of complex systems.

If anyone could predict the future of financial markets they would be a trillionaire. For reference, over 90% of mutual funds managers and over 95% of hedge fund managers under perform the S&P 500 over a 20 year basis. Their performance is worse than dart throwing monkeys. This is due mostly to fees - but obviously they are not - as a group - relatives of Nostradamus. And if you’re wondering - the average person performs far worse due to poorly timed purchases and sales (buying high selling low).

Predicting the future of a complex system is impossible. Why? The degrees of freedom (variables) interact with one another. Cultures evolve. Economic opportunities change. And yes - in cases - the weather changes too!

The more variables there are in a system the more difficult it is to figure out which ones combine to create signal. At a certain amount of complexity the signal disappears entirely. Remember, [time] needs to be multiplied by all variables because the path variables take is as impactful as where they arrive, making the total number of variables grow exponentially as time passes.

If you wanted to predict how many people will die from climate change 50 years from now - one of the most important variables to be able to forecast would be the performance of financial markets between now and then…If financial markets have negative returns between now and then, we would need to adjust our expectations around technological sophistication, the amount of infrastructure we can afford to build, etc. Similarly, if global stock markets compounded at 10%+, then we can infer that humans in 2075 are rich as Crassus and have likely invented all sorts of wondrous new technologies and built infrastructure that makes us look like a space faring civilization. It is also quite possible that we’re damn near done figuring out how to control the weather itself (if not already controlling it).

Here’s the takeaway. Financial markets are one of many complex systems that together make up the larger complex system of “human life”. If you can’t predict financial markets (or any other complex system that is part of the whole) then you can’t predict anything about any complex systems that count financial markets as a building block.

ChatGPT summarized complex systems thus:

The nonlinear relationships among components create feedback loops and emergent properties, where the whole behaves differently than the sum of its parts. These intricate interactions amplify uncertainties, ensuring that small errors or incomplete knowledge quickly compound into significant unpredictability over time.

The techies who most clearly illustrate this cognitive dissonance are (currently) the hardware bros. Here’s Gill suggesting that in order to “save the environment” - whatever that means - we need to climb the Kardashev scale (explanation below if you aren’t familiar with the term).

There are a group of people who believe (as I do) that we are going to have humanoid robots with human level intelligence - by the billions - able to help us build infrastructure - as soon as the late 2030s. But still they’re worried about the environment? What part of it specifically I can’t fathom. If we’re going to be landing humanoid robots on Mars we can probably allocate a few of them to plant some trees…

Share

Subscribe now

Consider how remarkably accurate Ray Kurzweil’s predictions of future compute capabilities (both demand and power per $) have been.

Decades ago he predicted 2023 would be the year when widely available computers would be capable of human level cognition. Given the capabilities of LLMs today that seems pretty remarkable. 2023 is unquestionably the year that machines felt like they “woke up” via mastering human language.

One of my predictions that has panned out accurately is that the end of Moore’s law wouldn’t slow anything down - and that the pace of technological advance would continue to accelerate despite the slowing rate at which we can fit more transistors on a chip.

I did not predict Nvidia’s relevance to AI nor did I predict that AI/ML would collectively lead to the greatest capex cycle in history. I didn’t predict the insane improvement from Hopper to Blackwell to Rubin (omg, nuts, read the latest Semianalysis piece linked here). But - what I did was extrapolate - as Kurzweil has - a trend that has been in place for a very long time (see this post for the deep dive on “The Exponential Growth of Progress”.

In the above post I commented that for most of history (4.4 billion years) progress - which I define as “information growth” - was driven by evolution. For example, two pterodactyls contain more information than one, but adding a new species via evolution- a velociraptor for example - does far more to increase total information than adding even another 10 pterodactyls.

All information growth was caused by evolution until humans came along. Humans were able to spread information from one to another through advanced communication + memory. For a few hundred thousand years humans were the dominant cause of information growth. Then we started to invent tools. Tools are anything useful insofar as increasing total information. Language is a tool. Computers are a tool. Farming techniques are a tool. Over time the contribution (“Market Share”) of tools to information growth has increased.

Crucially, for 99.999999% of the past 4 billion years (since the first anaerobic bacteria started multiplying) intelligence has been the limiting factor for progress:

If you want to do a gut check about when tools became more important to progress than human (“labor”) - I’d say to contemplate when the nerds started to take over. Today they account for the vast majority of new wealth creation (Musk, Zuck, Bezos, etc).

Nerds leverage the value of tools far better than do other humans - Elon being the preeminent example. Moving on, what has happened over the past 30 years is that tools became far more impactful to progress than labor. 50 years ago tools might have been +/- 10 points, but today they are obviously dominant. Consider how common it is to have companies with 100,000 employees that are worth more than the bottom 3 billion people on Earth put together. Similarly, consider how increasingly common it is to have individuals running businesses that make $1-10m per year? I see a new example pop up almost daily.

For the first time ever - thanks to AI - progress will not be [human intelligence] constrained. I think this happens within the next 10 years, and will obviously be the case post 2035.

The ramifications of this are profound. Let’s unpack it.

I need to clarify that when I say “natural resources” will be the limiting factor - I mean our ability to extract them from the earth and sky. I also want to stress that so far we don’t have a resource extraction problem - we have a red tape problem. China for example is able to build energy infrastructure at a rate - literally - 10X faster than we can. The bottlenecks we have in the United States around building more energy (whether it be nuclear or natural gas or the transformers needed to take energy from the grid and put it into a data center) are entirely self-inflicted. But I think this changes eventually and it simply won’t be physically possible to build all of the energy we want to use (until the physical robots start making new physical robots - analogous to digital intelligence that is recursively self-improving).

The price of resources (e.g. natural gas) are set through a combination of supply and demand. Historically, demand for commodities has always fluctuated wildly - that’s why we call them “cyclicals”. Prices fluctuate because demand increases - so the price goes up - so more projects to extract more of the resource become economically justified - so capacity increases - then capacity overshoots and price comes down again. A tale as old as time.

Here’s the thing - I posit that the reason capacity overshoots is because intelligence has been a “limiting resource”. Capacity overshoots because we don’t come up with a new source of demand before the old source of demand S curves. We have something like the following:

The reason price comes down is because we don’t get a new S curve to take place of the one before it (or more realistically, we don’t get many new S curves to take place of the many burning out):

Share

Subscribe now

I’ve overlayed what I expect to happen to cyclical capex (both for chips and natural resources) on top of Ray Kurzweil’s chart of the intelligence explosion:

I think we’re already seeing cycles shrink in terms of both [peak-trough amplitude] and frequency. While the cycle has always been following this trend of up and to the right - might we eventually just get to a point where the cycles are barely noticeable if at all?

When people talk about a “Fast Takeoff” - what they’re referring to is intelligence that becomes capable of improving itself. It can update its code, make itself more energy efficient, make itself smarter, etc. Another term for this is recursive self-improvement. Recursive self-improvement is what will effectively once and for all drive the cost of intelligence asymptotically to zero. I made this chart for a different post and it plots the cost of intelligence against time.

In economics there is a concept called price elasticity. It describes how changes in price impact demand. Some goods are completely inelastic - it doesn’t matter what the price is no one will change their purchasing habits. An example of a good with zero price elasticity is my brother’s dog’s shit. It has zero value no matter what price it is because the demand for it is zero.

At the exact opposite end of the spectrum is a patriot missile. The government will pay whatever it costs to purchase one, regardless of price.

A typical price elasticity chart as used in college textbooks looks something like the following.

The chart is a tad misleading because it seems to imply that the demand for goods will go near infinity long before price hits zero, but you get the gist.

In the real world demand for everything is limited. There’s only so much humans can use. But again - I posit this is due to intelligence being a limiting resource. Consider the following:

The most price-elastic good imaginable - where reducing the price toward zero would theoretically drive demand toward infinity - is something with:

1. No inherent consumption limits

2. Universal desirability

3. Minimal storage or usage costs

A classic theoretical example is money itself, or something extremely close to it, like risk-free financial assets (e.g. Treasury bills). If you were to offer money at a discount (selling a dollar for less than a dollar), rationally, everyone would infinitely demand it.

Another clear example could be Energy (e.g. electricity). If energy were truly free and limitless, people would immediately find infinite ways to use it, such as cryptocurrency mining, artificial intelligence computation, climate control, desalination, vertical farming, or large-scale data centers. As the cost approaches zero, the demand could theoretically spike indefinitely, limited only by physical constraints, not desirability.

Artificial Intelligence that has human-level agency (which is my definition of AGI) also deserves a spot on this list. Why?

1. Infinite usefulness.

   1. It will enhance productivity, problem-solving, and innovation across every human activity. It will also come up with the solution to rendering humans “pointless” insofar as economic productivity goes.

   2. There’s no upper bound tot he utility of increased intelligence - every increment unlocks additional possibilities, compounding its desirability.

2. Universal Demand.

   1. Every individual, business, government or institution will want as much as they can get their hands on - particularly once intelligence becomes embodied

   2. Unlike niche products, intelligence has applicability across all fields, from medicine and education to finance and engineering to companionship

3. Near-zero Marginal Cost

   1. Software-based AGI could potentially scale with very low marginal costs - duplicating intelligence is far cheaper than replicating physical commodities

   2. If the price approaches zero, the lack of meaningful constraints on distribution (let’s go open source!!!) could drive exponential demand

4. Accelerating Returns

   1. AGI amplifies itself (recursively self-improving). This creates a snowball effect.

   2. More intelligence = more value creation = more resources to invest in further intelligence.

Intuitively this feels right, but let’s get tangible

It’s one thing to predict that demand for intelligence will be infinite. Most people probably agree that if energy and intelligence were free demand would be infinite. What savvy people wonder is whether the cycles are really going to shrink in amplitude and frequency.

Very smart people are worried that we are in an AI bubble - Nvidia will not hit its long term revenue targets - the Big Tech capex boom is unsustainable, etc. My view on this is that I don’t know, and to borrow a wimpy prediction from Larry Summers (recently forecasted the likelihood of recession was 50/50) - I feel the same about whether we’ll see another cycle.

What seems obvious to me is that if we get digital AGI then intelligence is no longer a limitation - therefore the only limitation is our ability to marshal physical resources - therefore any parts of the physical resource supply chain that can’t keep up with demand will stop experiencing price cyclicality.

Note that we could (and will!) still see cyclicality within the ecosystem. For example - maybe we’re able to extract oil or build solar panels faster than we can build transformers. If this is the case and transformers become the bottleneck, then prices for solar panels and oil could still fall if we have tons of production that can’t be put to use. The same could be true about the cyclicality of transformer prices if we are able to scale production of them far faster than the energy required to make them useful. But, these could be thought of as “mini-cycles” within the broader cyclicality of energy demand. What seems obvious is that when intelligence is no longer a limitation the cyclicality of energy demand itself will end.

What about chip demand? Interestingly, I think we end up in the same place post-AGI. The cyclicality in demand for semiconductors will end, but within the ecosystem bottlenecks could occur in semiconductor capital equipment, energy production, data center construction, etc. What will change is that the mega-cycles around demand for “intelligence” (e.g. semiconductors) will end.

One implication of not having what I’ll call “macro cycles” is that we can expect capex booms and busts to be smaller in the future (when aggregated). The busts in one area will be counter-balanced by the booms in another area - for example a bust in E&P (exploration and production) might be accompanied by a boom in expanding the grid. Similarly, a “bust” in logic/GPU chips might be accompanied by a boom in demand for memory and analog.

There is another implication which I think is more actionable - and that is the margins of the basket should increase over time. Let me explain.

Predicting the future is tough, but it is far easier to predict micro than macro. If the macro is guaranteed (e.g. increased demand for energy at a rate > production can expand) then what executives really need to be able to forecast is how long it will take a given bottleneck (a “micro”) to get resolved.

Obviously the macro is not precisely forecastable, but it is more forecastable than it has ever been in the past (and will become reasonably forecastable once we get AGI). Less variables impacting future demand make it easier for businesses to plan which in turn makes it easier to improve margins.

There is also another factor which seems to already be in play. The people who want more energy have a higher and better use for it than everyone who came before. Nvidia’s margins are what they are because the people buying their chips perceive them as having enormous potential value. Big Tech feels similarly about the value of data centers. The cost of Nvidia’s chips and the cost to construct a data center are - so far - perceived as being small relative to the potential upside.

Quick aside: I’d argue that Meta has already proven sustainable high ROI on their Nvidia spend.

Hence the improving margins of companies like Sterling Infrastructure and Eaton:

Again - maybe we’re overbuilding now, I don’t know. But, what I do know is that AGI is a sufficiently “high and good” use case that much higher margins will be supported across the ecosystem of companies enabling its proliferation.

My expectation moving forward is that margins will rise for all companies, and they’ll rise more quickly for companies in the United States than elsewhere due primarily to looser labor laws - American companies can basically fire at will - this is not true in ANY OTHER major economic zone.

To sum it up, I expect cyclicality to decrease at the same time as margins are increasing. The margin increases will be larger in percent terms for the companies who are moving to a less-cyclical future. The net result of less cyclicality and higher margins is that valuations should increase over time.

I went back and read this post from the start until this point and thought of better ways to organize and present these thoughts, alas I don’t have infinite time and I’d rather get more thoughts on paper than improve the presentation of what I’ve got so far. Also, back to my comment about feeling like an LLM lately - this post is sort of like that. I just started writing and went till I finished - analogous to the LLM doing some “thinking” in the background - except instead of doing the final edit I’m just publishing the thinking part.

Now back to the post.

Exponentials on top of exponentials

I’m not going to go through the work or address these topics in detail b/c this post is getting long enough, but I wanted to mention another thing I’ve spent a ton of time thinking about: how much data will we be using in ten years - and how will that usage break down?

In 2006 an average “techie” with a fast internet connection probably used around 25GB of data per month.

I just looked up how much I use per month and it’s ~1,600GB.

That represents a 6,300% increase over ~20 years.

It seems reasonable to me that the average increase over the next 10 years will be twice as much as it was collectively over the past 20 years (this actually feels extremely conservative). Meaning, we’ll have a ~12,700% increase in the amount of data a person is using per month.

This equates to 204 Terabytes per month or almost 7 per day.

For reference and just to get your mental juices flowing - most data consumed on the internet 20 years ago was peer to peer file sharing. Post smartphones it became content related - creating, viewing, delivering, etc + the compute spent on ML to turn that content into signal for purposes of serving ads.

Interestingly, the biggest source of data usage for me is Twitter (specifically on my phone). At first I thought this was strange because I never watch videos on Twitter (or rarely). How could Twitter be using more data than YouTube - which I am constantly streaming and spend more time using? What gives?

Agentic AI = more compute

I realized what’s happening is that Twitter is regularly caching content so that it’s easily accessible without needing to load. The reason this is interesting to me is because the number one use of data in the future will be - imho - personalized agents caching data so that they can talk to us about that data with zero latency - like in “Her”.

Let me explain. Today if someone uploads a video to YouTube that video generates data every time it is streamed to a person. For example: 100 million views of a 4GB video = 400 million GB of data streamed. Further, the video data is probably stored in 15-20 data centers around the world. Each video will on average also have meta-data stored (makes AI be able to understand it, used to serve ads optimized around it, etc) that is greater than the video size itself. So for this example we have the following:

Now - in order for “Her” to become a reality (and it will, likely around 2028 because the market TAM for Her is many trillions and 2028 is when hardware should be getting close to being able to handle it on a mobile device) - she has to be able to communicate with us with zero latency. That means lots of caching needs to be done - either on device or in a “cloud” that has a more limited search space - maybe a “personal cloud”.

Literally everything we come in contact with - what we purchase, who we talked to recently, what movies we’ve watched, what we’re working on - everything. The AI will essentially need to continuously predict what we’re most likely to inquire about - and then have that information at the ready so it doesn’t have to spend an awkward amount of time seeking it out before responding to an inquiry about it. This will require enormous amounts of dynamic caching. In order for the AI to talk to us about the movie - for example - it will likely need to have cached a significant portion of the meta data. We might end up with something like the following…

Note bandwidth demand increases by 50%, but storage demand increases by 125,000,000% (125 million percent). Each agent needs to have its own copy of the meta data. Now - I’m sure in practice the percent increase in demand for storage will be lower b/c as demand grows novel solutions will be found to lower the cost of delivery. But how much will demand for memory grow? 1 order of magnitude over 10 years? 2? 3?

And this is just one of many examples I can think of that could cause demand for compute / chips to skyrocket.

“Thinking AI” = more compute

Jensen commented at GTC on Tuesday that inference demand moving forward will be 100X+ greater than Nvidia had thought it would be at this time last year. Why? Because LLMs last year were just taking a single crack at coming up with an answer to a problem. Now they are “deep thinking”. The thinking part is generating on average 100X the number of tokens per question (you don’t see all of the tokens it generates, just the ones that are distilled into the final answer). I gut checked this with ChatGPT and Grok - testing out the difference in time to respond - and it passes the sniff test.

When stocks moon like Nvidia has lately - most people think that the “bold” prediction is that it will keep mooning. Statistically speaking this is incorrect. The bold prediction is never that the trend will continue - that is the “expected case”. When stocks run far longer than anyone expects it is because expectations don’t keep up with reality. This is why Apple didn’t go up 1000% in a day when it released the iPhone - instead it compounded at 20%+ spread out pretty evenly over the next 15 years. I am not making a prediction about where Nvidia will go - just pointing out that the odds are always - statistically speaking - in favor of a trend continuing. Note this is NOT the same as predicting a continuing trend for a company like Gamestop which has deteriorating fundamentals and no growth. There is a fundamental limit to the price people are willing to pay for an asset that has no ability to create demand for itself (by using profits to repurchase shares).

There is no such limit for companies who can create their own demand using profits. Nvidia’s forward PE is 25, lower than it has been since its 10X run started - because profits have increased faster than its share price.

I’m going to stop the post with a list of things for you to think about if you want to try and wrap your head around how much compute we’re going to need in the future:

1. Frontier AI model training

2. Inference scaling - particularly when Agents are asking Agents to do things

3. Synthetic data generation and digital simulations

4. Cybersecurity (imagine 1 billion agents trying to hack things all the time 24/7 - the best way to try and defend against something like that would be to have 10 billion “good agents” doing the same thing and identifying bugs faster)

5. Real-time rendering for augmented reality

6. Autonomous robots (vehicles, drones, humanoids, etc)

7. HPC / Scientific simulations (weather prediction)

8. Edge computing and proliferation of IoT devices

9. Space-based sensing and earth observation data

One final closing thought - in a world where resources become extremely cheap we can do things like stop hurricanes in their tracks. How? Imagine you have 1,000 giant ships with Cones on them. They ride around the outside of the hurricane and use the cones to funnel the wind power into on-board energy, simultaneously sucking the energy out of the hurricane. The problem is these ships are too expensive to build today - but they won’t be tomorrow!

The future is going to be wild. It’s coming so fast. Get pumped, stay healthy, investigate deep ketosis.

-Ben

Share

Subscribe now

Pre-Waymo cutting edge LiDAR systems cost around $75,000 per unit. Uses were mostly limited to military and geographical mapping. If you wanted to build a transport-related solution then you needed at least 3 units. Fast forward to today and these systems are tracking to drop below $1k in total in China. Here’s a chart of what that cost decline looks like

Here’s a log chart to make it clear that costs are still declining rapidly in percent terms from year to year:

LiDAR costs follow what is known as a Wright’s law cost decline. Essentially, for every doubling in production volume the cost per unit declines by an approximately equal percentage. Battery costs have done the same thing:

And the cost of solar panels has done the same thing…

It’s 2025 and I’m sure the vast majority of people are unaware of how precipitously prices for these technologies are falling. The impact of these prices plummeting is either obfuscated from or irrelevant to the daily life of most people.

Mobile Phones have also followed a Wright’s law cost decline, but the decline is hidden in the form of ever better technology. The same thing has happened to computers and AI servers. The cost per unit doesn’t actually drop, but the performance per unit rapidly increases.

What’s exciting about today - 2025 - is that these cost declines are about to become visible to the public. Waymo rides per week are growing exponentially. By the end of…

• 2023: 10,000 rides per week

• 2024: 100,000

• February 2025: 200,000

Forecasts are all over the place, but extrapolating fleet size based on production rumors gives us something like this. They’re managing 200k rides per week with a fleet not much larger than 1,000 vehicles.

For perspective, there are currently >1,000 Waymos sitting in the lot of their production facility

Link to image source

The cost of a Waymo has dropped from $250k to $150k in the past five years. These costs will fall further when they start using Hyundai in addition to Jaguar to build the vehicles on which their sensors are retrofitted.

If you compare the cost of a vehicle per hour of daily usage, Waymos would therefore be less expensive than a new Toyota Corolla:

Now consider the amount of value delivered to society by freeing up the time spent driving to do other activities. Let’s call it $15/hr. That gives us the following:

What we have is a vehicle that is already less expensive when amortized on a per hour basis than a Corolla - that also re-allocates $1m worth of time over the course of 10 years. This $1m could end up flowing through to digital advertising (people playing games or shopping while they ride), Netflix, or increased productivity for a business. There could also be surprising externalities like improved mental health and decreased demand for anxiety medication (commuting in traffic is stressful, I’m not being hyperbolic, this is a genuine possibility).

If we examine the economics including operating costs of a Waymo vs. an Uber driver in a Corolla the ratio remains surprisingly similar:

Costs are lower than in practice b/c I didn’t look up estimated gas and/or electricity / standard maintenance because they’re all equivalent, but the point is to compare the price between autonomy and human driven rather than to calculate an absolute cost.

This cost difference is why I expect to see a surge in Waymo vehicle production. Costs will only drop as time goes on.

Let’s assume that the number of Waymos on the road doubles every year for the next 2 years and then increases by 50% per year for the 3 years after that (this seems conservative to me). Let’s further assume that Insurance/Connectivity/Tech; Operational Support; and Maintenance (of advanced systems) fall 10% in cost for every doubling of production. That gives us the following in 2030:

If I had to guess, I expect Waymo technology to be doing on the order of 10 million rides per day in 2030. Backing into that figure implies 20 rides per day per vehicle, with 500,000 vehicles on the road. This seems plausible, and would represent approximately .91% of daily trips taken by people around the country (includes bus trips, taxis, any trip taken by a vehicle).

Subscribe now

Share

Here is Lennar’s normalized EPS growth since 2014:

Lennar always has been and always will be cyclical. However, their cyclicality will also reliably be around a line that moves up and to the right. EPS doubled from the top of the last cycle in 2018 ($6.35) to the bottom of this cycle ($12.28), and as surely as the sun rises it will go on to new highs.

Idk if the next cycle will start in 2026, nor do I know when it will end. What I do know is that at $12.28 of EPS Lennar is currently at an Earnings yield of almost 12% if you strip out the cash. What I do know is that at some point it will have a new upcycle and earnings will hit new highs. Meanwhile, it has $4B of net cash on a $30.5B market cap - plenty of room to buy back stock while we wait for the good times.

Let’s do another. Here is AMAT’s (Applied Materials - the second largest semiconductor capital equipment maker by revenue after ASML, and the most diversified) historical and forward EPS estimates:

Semiconductor capital equipment companies are all supposed to be cyclical, and they are, but the secular tailwinds at this point are so strong that we’re barely even seeing EPS drops during the “down cycles”.

For EPS increased from $.90 to $3.23 during the first cycle; from $2.86 to $8.61 in the second cycle; and will start the third cycle at $8.24. There’s actually pretty good visibility into 2026 revenue already so the $9.92 they’re expecting probably gets hit.

Meanwhile, both homebuilders and semicap (abbreviation for semiconductor capital equipment companies) are run by veterans who understand how to take advantage of cycles to purchase shares at a discount.

AMAT shares outstanding are down by 32% over the period.

Margins for both semicap and homebuilders used to be far more variable than they are today. What’s changed? In both cases the past two decades have seen the markets consolidate. Homebuilders got out of the land speculation business. Semicap install base has grown and with it steady services revenue now accounts for a bigger piece of the total. Both groups of businesses have secular tailwinds.

DR Horton has grown it’s market share from approximately 0% to over 14% - building 1 out of every 7 homes in the US:

Lennar has seen similar and remarkable growth. Looking across the basket, major builders have consolidated the market to the point where they account for 60%+ of total new homes built. Their size combined with optioning rather than purchasing lots has turned them into capital light manufacturing businesses that just happen to “manufacture homes”.

DR Horton has higher margins than most due to its scale - but virtually all homebuilders - even the smaller regional players - have margin profiles that look more like cyclical manufacturers than the homebuilders of yore:

(In case the letters are too small - over the past 10 years DHR trough gross margins were 20.12% and peak were 30.08%).

Homebuilders will not grow as fast in the future as they did in the past, primarily because they already have so much of the market that there are fewer smaller builders to rollup and hence less consolidation ahead.

Semicap finished consolidating a decade ago - though that hasn’t stopped their phenomenal EPS growth thanks to the secular tailwind of humans needing ever more chips.

These are AMAT’s operating margins (not gross margins):

It’s absolutely wild that they maintained a 23% operating margin at the bottom of the last cycle, and even more wild that they will probably bottom out above 28% this year (another low). The biggest risk to semicap - by far without a close second - is China. Not only do they get a huge portion of their revenue from China - the more the US puts technology export bans on their equipment - the more incentivized China is to subsidize local competitors. Still, it seems ridiculous to me to believe that given the acceleration we’re seeing in AI that this rising tide will not result in still remarkable EPS growth moving forward.

The companies I’ve named are not anomalies. You can throw a dart at any company in either group and the story is similar. Here’s - my personal favorite among the semicap names (as a business, not at these prices though):

And how about those operating margins?

We’ve hit the point in semicap where the margins at the cyclical lows are still world-class. Once again - here’s what shares outstanding have done over the past ten years:

Here’s another homebuilder’s EPS growth - America’s third largest by completed homes:

And what a share cannibal they have been:

And here are Pulte’s historical and forecasted EBIT margins:

The largest holding of my [homebuilder/construction basket] is a company called St. Joe - but it’s a unique animal so I won’t get into in this post, but the second and third largest holdings are not actually homebuilders at all - but suppliers to the industry (in addition to supplying non-residential). They are Owens Corning and Carlisle.

has seen phenomenal EPS growth over the past ten years as it takes market share and rolls up smaller companies at low EBITDA multiples - which it immediately improves by centralizing operations and using their purchasing power to negotiate COGS down.

Here’s OC’s operating cash flow:

And yet again (it’s a theme, I know) we have a share cannibal:

Operating margin growth has been a thing of beauty:

The basic theme of this post should be clear by now. We have two cyclical industries (homebuilding/construction and semicap) - both with huge secular tailwinds. Margins, EPS, Cash flow etc are all cyclical - but cyclical around lines that are going up and to the right. Over time the cyclicality has decreased, and that should lead to multiple expansion in the future, meanwhile you have incredibly shareholder friendly management who will repurchase shares opportunistically (as opposed to companies who just buy no matter what the valuation is).

One thing an investor needs to keep an eye on is how good management is at predicting their own future. There is NO MANAGEMENT on Earth (probably?) who is as bad at predicting their own future as Snapchat. It’s hysterical listening to them talk about their business - you can just assume that whatever they think about the future the exact opposite will happen.

Across semicap and the homebuilder/construction basket, most management are top notch, grounded in reality, and pretty good at forecasting. In 2018 Carlisle laid out their “Vision 2025”:

Let’s see how they did. Revenue first. In 2018 Revenue was ~$4.5B. It broke down as follows:

Of that $4.5 → $3B was in the construction materials business. Their revenue for 2025 is expected to clock in around $5.3B - all of which comes from the construction business b/c the other divisions were sold:

They didn’t hit the 2X, but 76% growth isn’t too shabby, and look what’s happened to margins and EPS:

Carlisle’s goal was $15 in EPS by 2025, they hit that goal in 2022. Their new objective for 2030 is to hit $40+ in EPS:

Carlisle is at $336 today. If they hit $40 in 2030 and trade at 15X you’re looking at ~12% share CAGR before taking into account shareholder distributions - which should clock in around 6% this year.

IMO the remarkable performance (shareholder returns) of these companies - both semicap and homebuilders/construction cos - is made possible by the cyclicality/uncertainty associated with their respective businesses. The market refuses to give them a multiple that would result in their long-term go-forward returns being more in line with the S&P500. For investors this is a gift, because the cyclicality isn’t the same as it is in the oil or coal industries - where downcycles can last decades and margins don’t go up from cycle to cycle.

Consider as an example the margins of Exxon Mobile:

Or worse, a coal company like Alpha Metallurgical Resources:

Or even worse, an offshore drilling company like Transocean:

I’m not saying you can’t make money in companies like Transocean, Exxon and AMR. Some very savvy people who I respect have large positions in these companies. But what I can tell you for sure is that I am unwilling to put large amounts of money into companies whose cyclicality is around a flat line, rather than a line that goes straight up and to the right.

Given their current valuations it seems extremely likely that both groups will beat the market moving forward, and given their cash-gushing shareholder friendly nature, the downside seems limited after the recent pullback. All companies in each group are down approximately 30% from recent highs.

Subscribe now

Share

When you put it all together, ChatGPT is building out a very detailed profile not only me as an operator and as a person, but also of how accurately I perceive myself and my own abilities.

Already I’ve experienced some moments that gave me chills where ChatGPT was able to identify connections that I had failed to see. For example, I had been trying for months to figure out how to re-create days where I felt like my brain was functioning at a vastly superior level - what I call a “brain on fire day” (energy, speed, recall, reflexes in tennis, all excellent). I had been trying to correlate it with sleep, diet, exercise, etc - to no avail. ChatGPT noticed that I used certain verbiage to describe how I felt more frequently (e.g. Amazing, Extreme Energy, etc) ONLY when I had recently told it that I was in deep ketosis (2.0 mmol/L+).

Sure enough, so long as I maintain my ketones at that level I feel incredible, even if my sleep isn’t that good… I can now - reliably - create “brain on fire” days.

Take a moment to think about how incredible this is. ChatGPT has helped me unlock more energy - literally giving me more time per day! In essence, ChatGPT just extended my functional lifespan!

One of the toughest things in life generally speaking is simply executing on the best advice anyone has ever given anyone else: do more of what works and less of what doesn’t. If you can follow this advice over decades you are sure to become happy, healthy and wealthy. Unfortunately, even with herculean amounts of self-discipline and keen introspective observance - it’s extremely difficult and time consuming to monitor “what is working” and “what isn’t working” across all facets of our daily lives. Often times - as with my quest to figure out how to reliably experience “brain on fire” days - it is simply beyond our ability to identify the relevant causal relationships.

LLMs are going to change this.

One thing you will notice everywhere once you start keeping an eye out for it is how frequently people update their “personal belief history”. They will claim not to have believed something which they did in fact believe. They will claim to have predicted something that was in fact the polar opposite of what they thought would happen.

Twitter is an amazing resource for seeing this in action because people regularly post tweets that go viral where public figures or big accounts are saying something in 2025 that is directly contrary to something they said in the recent past.

Similarly, Twitter also calls out people like Ray Dalio and Howard Marks who have been calling for bear markets every year since the financial crisis - meanwhile markets are up many hundreds of percent…

The most pervasive (and likely one of the most pernicious) examples of this in action where it really matters is people [on the left and right] “predicting” that under the new administration the world will fall apart. People thinking Obama would turn the US into a communist country, for example. People today thinking that Trump is going to crash our economy and turn us into a dictatorship (and run for a third term).

Having these fears is not only irrational - it has a real impact on life. You will make bad money choices if you are irrationally pessimistic about the future. You will have less mental energy if you are worried that the country is about to implode.

But unless you are aware that you are prone to these types of errors you won’t be able to correct them. And, spoiler alert - everyone is prone to these types of errors. Evolution meant updating your belief systems to be in line with the group - elsewise you were at risk of being ostracized - which meant certain death 100,000 years ago.

Today - most beliefs are not rationally derived probabilistic views of the world - they are the result of media or algorithms optimizing monetization of thought bubbles - or worse…

This is why many great investors keep journals and write long letters - it forces them to track their historical decision making processes so that they don’t fall prey to “re-writing” history. It keeps them rational and accountable, and facilitates improvement over time.

I don’t think I’m being hyperbolic to say that improving decision making by some percentage (10%? 15%?) would have a profound impact when compounded over time. Already I’m finding myself getting 1-2+ hours of extra focused work per day by having identified that deep ketosis fills me with energy.

LLMs are naturally (innately) optimistic - because optimism is the rational perspective given the corpus of human history we have to extrapolate from. So - by using LLMs not only as a daily journal and a clever statistician, but also as a trusted counselor - it will be impossible not to benefit via improved decision making and by having a clearer picture both of the world as it exists today and your ability to accurately diagnose your own strengths and weaknesses.

My goal moving forward is to upload as much information about myself, my business, my finances, my health - and so on - as possible. Imagine being able to query an LLM about your poor investment decisions… Alternatively, imagine being able to query an LLM about the similarities between the decisions that turned out great…

Imagine literally figuring out how to biohack yourself - how to feel stronger, with more energy, more mental clarity…

Consider the following:

-1 extra hour of energy per day is equivalent to a 12.5% increase in productivity (assuming an 8 hour day—9/8 = 1.125) - which is equivalent to getting an entire extra month PER YEAR!

-Earning an extra 1% per year on your portfolio is worth 30%+ in the size of your nest egg compounded over 30 years…

Share

Subscribe now

What got me my first two thousand twitter followers and my first 1,000 Substack subscribers were articles defending Facebook as an investment after their stock tanked 25% during the first quarter 2022 earnings call (reviewing Q4 2021). I believed at the time (still do) that the moats of all “Big Tech” companies are more impregnable than most people think. The argument in favor of Facebook going out of business in 2022 went something like the following:

It has no moat. It already lost Gen Z. Apple’s new privacy rules will knee-cap its targeting ability. TikTok dominates short-form video which will be the new way everyone spends time. The user base has already peaked. It’s terminal value is zero.

I concluded my first and third posts about Facebook with the following:

1st post

Third post

Meta finished 2022 with $116B in revenue and is tracking to hit almost $190B this year and $215B next year, meanwhile EPS in 2025 is almost triple where they finished 2022.

The more I’ve thought about it the more similarities there seem to be between Facebook’s battles with TikTok and Apple, and Google’s “battles” (if you can even call them that yet) with the new LLMs.

To be clear, as I wrote in my articles about Facebook, TikTok was a real threat to Facebook (still is a formidable competitor) and LLMs have the potential to become a real competitor with Google. However, similarly to how Facebook’s battle with TikTok played out - I think we’ll look back in hindsight and consider it a wonderful thing that other companies forced Google to pick up this new LLM “mechanic” - because it will both monetize better and dramatically expand the TAM for advertising (and add entirely new TAMs).

Idk if anyone knows what Google’s moat is?

What X/Twitter seems confident of is: “Because I am Googling 95% less Search is going to die. Google = Search so Google is going to die. End of story.” I will address this line of thinking, but first let me point out how broad Google’s moat is:

1. 5 Trillion+ searches annually

2. 7 products with 2 billion+ users (soon to be 8 given the rapid growth of AI ‘expanded search’ feature inside Google)

3. 10+ years of custom silicon and unmatched data center tech leading to lower cost of cloud / AI delivery

4. Waymo is a form of real-world “embodied” intelligence and it’s taking off

5. YouTube is pulling ahead of the competition

6. Android install base of 4 billion devices

Between the data generated by Search + Gmail + YouTube + Chrome Google has - by far without a close second - the largest funnel of valuable data that has ever been created (useful for training AI). And don’t forget maps + photos - both multi-billion user products.

This usually comes as a shock to people - but Search will account for less than 55% of revenue next year. What’s more? The only reason it isn’t an even smaller portion of revenue is that it keeps on growing faster than anyone expects!

2025 Revenue breakdown:

1. Search 54.3%

2. YouTube + Subs + Platforms + Devices: 21.8%

3. Cloud: 16.8%

4. Network: 6.6%

5. Other .5%

LLMs have been out since the end of 2022, meanwhile:

1. Google 2023 Search revenue growth rate: 8%

2. 2024 Search growth rate: 14%

3. 2025 Search growth rate (Forecast): 10%

The attack of the Google “verb”

It feels like I’ve seen at least 20 people make the joke: “I don’t Google things anymore, I Grok them”. Let’s break down where this matters and where this doesn’t.

Obviously it is not a good thing for Google to have other products with hundreds of millions of users start to get associated as “places you go to find information”. For the past 20 years Google has held a >90% mindshare as the first place you think of if you need to learn something. For 20 years, it didn’t matter if you wanted to learn about quantum physics or if you wanted to buy a plane ticket - the place you went to discover anything was Google.

Quick aside: there is a strong argument to be made that TikTok is actually a more direct competitor of Google than Meta - b/c TikTok competes with YouTube and it also is the number one place users of Gen Z start a “Search” - whether it be for a product or activity.

Today, anyone savvy will use an LLM to learn about quantum physics - they will probably also use an LLM if they want to plan out a vacation…But when it comes to purchasing the plane tickets and booking the hotels they will use Google.

The industries generating the most revenue for Google Search include:

1. Finance and Insurance

2. Retail and ecommerce

3. Travel and hospitality

4. Tech (hardware and software subscriptions)

5. Healthcare

I count myself in the basket of people who use Google 90%+ less than they used to. However - that’s because I - along with other “knowledge workers” - mostly used Google for things that never generated it any revenue in the first place (e.g. research). Rest assured, I still use Google for probably 99% of the activities that generate money for it - and so does everyone else!

What’s more - people on X/Twitter are an anomaly. Most people have never used an LLM. Only 1% of the world knows how to code. ChatGPT has 400 million weekly users (so I’ve read) - which makes it the only AI product thus far with a possibility of becoming a sustainably consumer facing company (Ben Thompson calls it the Accidental Consumer Tech Company - article here worth reading for anyone interested).

Perplexity? Claude? DeepSeek? Great products, I use Claude all the time and it’s my favorite LLM for certain tasks. But these products have zero chance of addressing the consumer market - which is the biggest by multiple orders of magnitude - and they represent ZERO threat to Google.

According to Sundar (Google’s CEO) - Google’s AI search overviews are now being used by 1 billion + monthly users - this for a product only launched in May of 2024. Zuck says Meta’s AI product now has over 700 million monthly users up from 600 million in December. Considering a typical ratio for weekly/monthly users of ~50% for products like Google and Meta’s - it’s probable that their LLMs already have more users than ChatGPT.

So, begs the question, in what situations does it matter that Grok/ChatGPT become new verbs and replace Google? The answer is Agents. Until the agents arrive the primary threats to Google will be what they always have been, namely, companies with massive amounts of distribution becoming an alternative channel for ad-generating searches. Apple could be a threat here, Amazon is already their biggest direct competitor due to product searches starting on Amazon.com, Meta is a threat here. What isn’t a threat to the Search business is LLMs.

Agents will change everything though…

The Agents are Coming!

I’ve written extensively about agents, I’ll link to those posts along with the Facebook posts at the end of this article. The relevant definition for an “Agent” here is something that can take actions on your behalf just like a highly tech-savvy virtual (human) assistant would. For example, it can respond to emails, book travel, do market research, etc.

LLMs are far from human virtual assistants today, but they won’t be for long. My guess is that by the end of next year you will be able to have an AI Butler purchase items on your behalf. HOWEVER, and this is a HUGE caveat. There are only a handful of companies who will be able to deliver agents, and NONE of the new LLM companies have any prayer of doing so (wrt to mass market consumers). There are 3 reasons why:

1. Something that has the ability to take actions on your behalf needs to be next level secure. If it can open your email and web browser it can also open your bank account, password file, etc. It is plainly obvious that none of the operating system owners (Apple w/ iOS; Microsoft w/ Windows; Google w/ Android) are going to provide this level of access to other applications - much less their most dangerous competitors.

2. People don’t want to talk to an agent for every app. Because the operating system owners won’t let any random app talk to the operating system itself or the other apps - the ONLY way to have a single agent for all apps will be for it to come from the operating system owner itself.

3. Apple, Google, Microsoft etc know they have this “Ace in their pocket”. They know that at any point they can absolutely knee-cap the LLMs by launching their own AI assistants that have operating system access - and not permitting that access to other apps. They’ll be able to invent whatever reasons around security, privacy, etc for ensuring that only they are capable of providing true Agentic AI. Even if there were no security issues, they would never start the precedent of giving such deep access to their operating systems.

It doesn’t matter how good ChatGPT is - the killer application that will be used by literally everyone can only be delivered by companies who control the operating system.

Where does this leave us with respect to Google? Right where we started. As has been the case for the past 20 years, the biggest threats are what they always have been - other Tech Giants - not new rabble rousers.

Two quick asides:

#1 If you are thinking “what if Apple used ChatGPT as the backbone for their intelligence instead of Google or an in-house model?” - I would say, yes absolutely that’s a threat - but it’s the same threat that has existed to Search for ever - namely - an attack from another member of Big Tech w/ massive distribution.

#2 I wrote a post explaining why only Big Tech can really threaten other Big Tech back in February of 2024 - I also explained in that post why cannibalization of one Big Tech’s services by another member of Big Tech is likely to have little impact or be margin accretive to the winner+loser’s combined flow-through P&L. The takeaway is to buy the basket - because the basket is impermeable and won’t cannibalize itself.

No one knows exactly what Search will look like in a world where Agents are doing the searching and making the purchases. The tech hasn’t been figured out. The business model hasn’t been figured out. However - one thing is very likely:

The TAM of digital advertising will dramatically increase. As a rule of thumb, here is a ranking of the most valuable information for a company trying to target us with digital ads.

The TAM for digital advertising grows as companies get more and higher quality data about us. LLMs will deliver to tech companies an entirely new quality of information. We’re heading to a world where LLMs:

1. Become our trusted confidants and advisers

2. Know everything about our personal physical and mental health

3. Know everything we purchase, everything our friends purchase, etc

Here are two examples of where the future is going:

And here’s one from me - I wrote a post about using ChatGPT to analyze and improve mental health (linked here if anyone is interested):

When Facebook first came out parents scoffed at the idea of their college students “putting pictures of themselves online” - or - horror of horrors - “letting predators know where they’ve been or where they are”. Now no one cares, and the number one preferred career of Gen Z is “Influencer” - essentially living your life on the internet for everyone to see. And have you heard of OnlyFans? Lmfao…

The goal post keeps moving. To anyone claiming they don’t want to divulge everything about themselves to AI - good luck with that. The future is obvious - and the future is sharing literally all of your deepest darkest secrets with an “Agent” that becomes one of your closest friends.

In digital advertising there are different concepts like “Click through rate”; “Cost per thousand impressions”; “Cost per action”. Today Google may need to show 200 ads to get a person to buy a product. The person whose product gets sold pays Google $2.36 for the click that led to the sale. They may need to get 10 people to click on an ad in order to get 1 sale - which would make their “Cost per acquisition” = $23.60.

In a world where LLMs know everything about us - just imagine how much more effective targeting will be?

More knowledge about us = better targeting = higher ROI for advertisers = bigger TAM for digital ad companies.

Consider two possible use cases for LLM targeting in the future:

1. A woman has her period and asks the LLM for quick tips to improve cramps. The LLM suggests things she can do and also asks if she’d like to have some pads delivered later that day. The woman says yes, the LLM orders them, the vendor (maybe a combination of Procter and Gamble + Amazon?) pays the LLM a fee for originating the transaction.

   1. What we don’t know is what’s going on behind the scenes - how did the LLM select which product to buy? If it just bought the same product as last time does any ad revenue get generated? Lots to figure out here.

2. Here’s one relevant to me: You have been having conversations with an LLM for over a month about health and fitness. The LLM correctly identifies that enough time has passed that it’s probably worth taking a new blood test, Vo2Max and Dexascan (body composition).

   1. Again, how does the LLM pick the product? If it has a discussion with you about what you want (e.g. the closest testing location) - does the business pay for the lead? Idk…

I’ve seen some people comment that if Agents are doing the searches then there will be fewer searches and fewer eyeballs seeing ads and hence ad revenue will go down. I disagree. More effective ads = more ad revenue. This is obviously the case and always has been (in the digital ad space), people who think otherwise just haven’t really grokked the real drivers of digital advertising’s success over the past 20 years.

Summary:

It’s frankly wild to me how pervasive the idea is that Google stands to be one of AI’s biggest losers. The financials don’t back up this view. The user data doesn’t back up this view. The different paths the future of AI might take don’t back up this view. Google is trading as if Search has already entered terminal decline. It’s trading at a lower multiple than the S&P 500 with twice the growth and a variety of rapidly expanding TAMs. It’s absurd.

The biggest threats to Google today are the same as they have been for the past 20 years. LLMs don’t pose a threat unless they get Agency - and once Agents become technically feasible the only companies capable of delivering them will be the operating system owners.

Just like with Meta and TikTok - LLMs will expand the TAM for advertising by a lot - as well as create entirely new TAMs like AI butlers…and will end up with the benefit of hindsight as being obviously a very good thing for Google.

People don’t realize that while LLMs are now a better place to do research, research never generated Google any $$$ in the first place.

People don’t realize that even IF LLMs had agency, the way they will get into most people’s hands is through operating systems (iOS + android). Distribution is the moat.

Subscribe now

Share

Posts referenced above for further reading…

Here are the Facebook articles if anyone wants to go revisit them:

The narrative that drove Meta’s stock down 70% in 2022 was that Apple’s privacy measures would kneecap Meta’s ability to serve ads, AND that TikTok had completely captured the next generation - thereby effectively neutering any potential user growth in the future - and worse still - would start to suck away users and total user time. Investors also feared that if TikTok could make a killer algorithm that drew people away from Meta’s properties - so could someone else…

Meta’s moat wasn’t what people thought it was…

My own usage of Google is down probably 95% since LLMs came out. But, the 5% of stuff I still do on google probably represents almost 100% of the stuff I do that would generate revenue and ad-clicks (e.g. buying a plane ticket). I do believe LLMs are a threat to Google (just as I have always believed that TikTok is as formidable competitor to Meta as any), and it is obviously a very bad thing that hundreds of millions of people are now associating LLMs with being the best place to find information. Google’s motto since its founding has been: “Organize the world’s information and make it accessible”. 12 Links and a wiki page are a laughable solution to most of what I need to research these days when compared to LLMs like Claude and ChatGPT (and more recently Grok).

Fast forward a few years and the TikTok threat isn’t talked about anymore. Usage on Meta’s properties has increased (as have user counts) - in large part thanks to their adopting TikTok’s short-form video product and distributing it across their user base (which accounts for ~2/3 of non-Chinese ON EARTH!). Short-form video dramatically expanded the TAM for advertising, and with the benefit of hindsight it was a wonderful thing that TikTok showed Meta how large of an opportunity there was in short-form video.

I believe the same thing is going to happen at Google. LLMs are going to DRAMATICALLY expand the TAM for digital advertising. Remember, the value of a digital ad is directly proportional to its ability to influence an action (e.g. a purchase decision).

If you think it’s creepy how much data companies have about you today - just imagine how much data they’re going to have about you after you spend all day having conversations with them about your health, interests, relationship struggles, etc…This data will make the ad space (volume) increase because you will be spending more time on the service - and it will also make the value per ad delivered be higher because of improved targeting.

Remember - ONLY the companies who own operating systems can deliver the an AI product that has the agency to control your device. This is a big deal. This is the only deal that matters. It is a “get out of jail free card” they can use at any time to render competitors instantaneously obsolete. This is why Apple isn’t worried about getting behind in the AI race. They don’t need to own any intelligence themselves - they just need to control the distribution and be the only company that can connect the intelligence to their operating system.

Subscribe now

Apple delaying their intelligence offering is bullish

This tweet got 6.9 million views and counting:

The entire internet exploded with people claiming that Apple’s end is near because they’re “falling so far behind on AI”. On March 4th, I published this rebuttal on X:

My take on Apple delaying "New Siri" until 2027: It's actually bullish.

It indicates that Apple realizes what the "killer app" is. I've been saying publicly since early 2023 that the killer app for LLMs was something integrated into the OS, capable of communicating with all apps as if it is a "new GUI component", and which has zero latency.

Something which can have a human-equivalent conversation with you about your email, do your shopping, change settings on the phone, and even help you make a graphic for a presentation you are working on...

Piecing it together, I've predicted that this would only be possible in the late 2027/2028 (for mobile devices) timeframe due not only to software features that need to be figured out (e.g. dynamic caching), but also because hardware needs to get upgraded (especially more memory) to make this a seamless, "Apple quality" experience.

Apple went so hard after Facebook over "privacy" because at the time - Facebook had become the number one determiner of what apps got downloaded on the app store. Apple had less influence than Facebook did about how people used their own phone. This terrified them and they had to put an end to it. This is the true reason they invented their "privacy" campaign.

It was the only way to go hard after Facebook without triggering anti-trust investigations. LLMs have the opportunity to do the exact same thing that Facebook was doing - and Apple probably (almost certainly) knows this. However, Apple also knows it can sweep in with its own LLM that will be a 10X better user experience than anything else whenever it wants to b/c their version will be connected to the OS and all of the apps - something no other LLM will be able to do.

If you've used LLMs as much as I have, then you can probably already intuit what an insane leap it would be to have one of them just watching your screen at all times, listening to what you say at all times, etc. It would eliminate the need for uploading documents, prompting, having different chats, etc...It would just automatically be caching all of the recent activity at all times - ready to jump in wherever necessary.

This is where things are going, Google has already announced a screen share feature. Apple is just doing what Apple always does - waiting until it can launch the "killer app" that changes everything.

Ben Thompson is one of my favorite people and definitely the world’s best tech analyst. Everyone should subscribe to his newsletter. The below comes from his post on march 5th (one day after I posted my rebuttal) - he basically agrees with my take :)

To summarize, Ben believes the risk to Apple of moving too quickly around intelligence outweighs the potential benefits. This fits nicely with my commentary above. The killer app isn’t technically feasible to build yet - but when it is Apple will immediately be able to own it - no matter what other companies have been building in the mean time.

Google is in a similar position with its 4 billion Android devices, as is Microsoft with it’s 1.6 billion windows PCs.

Share

DEFLATION IS COMING

In high school (a charter school for math + science) I remember executives from Fortune 500 companies coming to my school to ask some of our young tech geniuses for help on their corporate websites. That’s how new things were in the early days of the internet. And boy did they cost a lot…

Were you to try and launch an ecommerce company back when Amazon was getting started your budget would have looked something like the following:

1. Planning and Expertise: $10k+

   1. You needed to register trademarks, incorporate, write an operating agreement, draft ancillary docs (e.g. terms of service, return policies, etc)

2. Technology and Software: $15,000+

   1. Upfront hardware costs for servers: $5,000-$20,000

   2. Operating system (Windows NT) $500-$1,000

   3. Web Servers (Microsoft IIS) $500-$2,000

   4. Oracle database server license ($5,000-$20,000)

   5. Payment Setup ($500-$2,000)

3. Website Development: $25,000+

   1. A site with product catalog, shopping cart, checkout would cost $20,000 minimum

   2. $2,000-$10,000 to create marketing content

Total costs of $50k + $50k per year ongoing…

Ongoing operational costs were probably $50k+ per year just for technical expertise. Now compare that to where we are today. I’m going to use real figures b/c I just so happened to have started a new ecommerce company recently.

1. Planning and Expertise: ~$500

   1. You can get an operating agreement drafted by LegalZoom (or at this point an LLM) for $100; you can teach yourself to file trademarks with an LLM for $350; ancillary docs are available for free online

2. Technology and Software: $0

   1. $300 for a Shopify template

   2. $10ish for Google Workspace account per user per month

3. Website Development: $1,500

   1. We had someone help us out for no charge, but she’s very talented and if we would have paid her by the hour it probably would have been around this

Ongoing operational costs are just the Workspace account, and our subscriptions to Shopify and Canva ($45.00 per month).

Total costs: $2,000 + $540 per year ongoing…

There were also soft costs in the 1990s that don’t exist today. For example, the number of people who had the expertise to set up an Oracle database was extremely limited. If something broke it would have been more expensive and time consuming to fix. The only way to guarantee very little downtime would have been to have someone on staff full time. Today Shopify delivers 99.99%→99.999% uptime (4-5 nines).

Another example is legal expertise. Back in the 1990s if you needed a new wholesale agreement drafted it would have cost thousands of dollars. Today you can find countless examples online or just use ChatGPT to draft one.

Collectively, this represents approximately 97.5% cost deflation over the past 30 years.

I personally know multiple people running multi-million dollar ecommerce companies with a handful of people.

This deflation has been caused by advances in both software and hardware (Wordpress and Shopify // AWS, Azure, GCP).

Incredibly, we’re still deflating! Consider the following: what will happen when AI can write the copy, make the creative, a/b test the ad campaign, etc…?

Did you know this is already happening with the latest versions of Meta’s marketing tools? The primary driver of deflation today is the plummeting costs of generating tokens (a reasonable analogy for the cost of using AI):

Since ChatGPT launched, the cost of “creating” AI has fallen by 99.6% - meanwhile it has added 3-4 standard deviations of intelligence. Combine the two and you have a ~500 to 1,000 X improvement in cost per unit of intelligence.

What happens when artificial intelligence gets a body?

Let’s look at the costs to build a house today…

1. Raw Materials: 40-50% of total cost

   1. Lumber / Concrete / Roofing / Windows / Doors / Finishes

2. Labor: 30-40%

   1. Carpenters, electricians, plumbers, masons, roofers, GCs

3. Other Costs: 20-30%

   1. Permits, architect fees, land prep, equipment rental, builder profit

For a $300k house this could break down to:

1. Raw Material: $150,000

2. Labor: $90,000

3. Other Costs: $60,000

What would happen if just half of the labor costs were removed via automation? We’d see a 15% decrease in build costs. Shelter makes up 35-36% of the CPI. If construction costs were to drop 15% inflation would be gone forever. And, don’t forget that labor makes up a material portion of the costs of the raw materials (both from harvesting + storage + transport).

The same thing that has happened to LLMs and generative AI is going to happen to robotics (humanoid + other). For reference, here is Nvidia’s revenue over the past 8 years:

And here is what is expected to happen between now and 2030 (the red line starts where the first chart ends):

The total addressable market for machines with human dexterity and human intelligence will be effectively unlimited.

If it’s done on a computer - it will be automated…

White collar work will probably end up coming under attack even more quickly (and arguably already has according to revenue per employee metrics). This chart shows the growth of administrators vs. doctors from 1970 to 2015.

What percent of these jobs can be automated? The answer is obviously most of them - or at the least automation will enable one person to do 10X the work they were doing before.

This year we’ve gotten two different announcements about tech giants releasing products that enable a person to “screen share” with an AI. I already use AI to do many things I never would have been able to do before - like set up web domains (I’m not a technical person), migrate enormous QuickBooks files with 75k lines, and set up complex cap table conversion formulas in Excel to get around the recursive reference issue that always shows up when a million different notes with different terms are being converted at the same time and language is ambiguous on order…

In each of the above examples - the way I went about it was to take a screen shot of my computer and then ask the AI what to do. In many cases the instructions were complex enough that I started telling it: “Provide me step-by-step explicit instructions, one instruction at a time, and wait for me to ask to go on to the next step”.

The AI was able to look at the image and tell me exactly what to click on, what to type where, and how to test if it was working along the way (so that we didn’t get to the end of a process and miss something that broke everything we had done).

The next step is screen sharing. The step after that is giving the AI access to the operating system so that it can just do the work for you. It can click on the files, write the excel formulas, etc…How long will it be before we start seeing demos of this primitive form of agency? I believe they’ll be here by the end of the year…

Eventually, one person will be monitoring AIs that are doing the work of 10 or 100 people.

While I am not excited about the disruption this will cause to many people’s lives - I could not possibly be any more bullish on the future of semiconductors and companies that own operating systems, cloud, and distribution.

All tech companies could aptly be considered as providers of “Semiconductors as a service”. This is as true of Meta as it is of Microsoft or Apple.

Only the operating system providers can deliver the AI which is connected to everything and which will eventually replace humans en-masse. If not for technical reasons - then because the owners of the operating system will use their position along with some fake narrative about “privacy and security” to make sure that nothing else works as good as their own products.

Share

From inside:

“If you’ve been telling yourself this isn’t coming, I really think you need to question that. It’s not web3. It’s not vaporware. A lot of what we’re talking about is already here, right now.

I think we are on the cusp of an era in human history that is unlike any of the eras we have experienced before. And we’re not prepared in part because it’s not clear what it would mean to prepare. We don’t know what this will look like, what it will feel like. We don’t know how labor markets will respond. We don’t know which country is going to get there first. We don’t know what it will mean for war. We don’t know what it will mean for peace.

And while there is so much else going on in the world to cover, I do think there’s a good chance that, when we look back on this era in human history, A.I. will have been the thing that matters.”

What is AGI?

I’m almost giddy with excitement. Every day word is spreading further about the potential of AIG. I explained over two years ago why LLMs represented the unlock of a sufficient level of intelligence to create new trillion dollar industries:

Alas, even trillion dollar industries are not the same as AGI.

The reason ChatGPT wasn’t considered AGI then and why nothing we have is considered AGI today (by most people) has less to do with intelligence than agency. AI systems like ChatGPT 4.5, Claude 3.7, and Grok are already better than humans at many things. But until they get “plugged in” to GUIs, software APIs, and have their memory upgraded to be effectively unlimited - they will continue not feeling like AGI.

Every thought leader in the AI space - including a person named Ben Buchanan (NOT me! the guy Ezra interviewed) who was an AI advisor to the Biden administration - laments they are not satisfied with the term artificial general intelligence (in most cases because it is ill-defined). If more people read 01Core then we wouldn’t have this issue, because my definition is ideal:

1. Digital AGI: An AI that would be indistinguishable from a human working remotely

2. Physical AGI: An intelligence that - if embedded into a humanoid form - could do whatever an average human could do.

See how easy that was?

Most people have no idea what’s already here! Much less what is coming…

From Ben (if you see this name for the rest of this article - just remember I’m talking about the other guy!):

Software developers regularly report that they have cut the time it takes to do a multitude of tasks by 50-90%. Deep Research can be used to carry out complex research assignments and put together reports that are hundreds of pages long. Countless examples abound of it being used for analytical work. Grok (from xAI owned by Elon Musk), DeepSeek (from China) and of course Google have all released similar products.

I subscribe to three different LLMs (ChatGPT, Claude, and Grok through Twitter). I have saved myself ballpark $15k of hard costs (money I would have needed to pay to someone else - mostly subject matter experts like Quickbooks gurus or lawyers) from these subscriptions over the past twelve months, and probably more than that in terms of time savings. And don’t forget this post from yesterday:

As Ezra points out in the article - systems today are as bad as they will ever be.

There are three switches that will flip soon - by the end of this decade - and as they turn on chaos will ensue (but only briefly).

1. AI will get agency - the ability to take actions and perform any task a human with computer access could do. For example, instead of asking ChatGPT “How do I set up a website” or “How many calories are on this plate of food?” - you will simply instruct it to: “Set up a website, find some products on Alibaba that I can import at a good price, then list them on the website and start selling them. Also, set up some social media accounts and begin marketing campaigns for those products.”

2. AI will get a body - The “Synths” as I call them - short for Synthetic Humanoid - are coming. I’ve written about this extensively. Until recently there was no justification for pouring a hundred billion dollars into R&D to figure out how to create a functional “skeleton” to put a brain inside of. Because there was no brain. Now that there is the incentive to own the business of manufacturing the last thing humans will ever need to make is incomprehensibly large. Humanoid robots will end up being a $10T+ industry - the largest in history by far. These things will start mass production this decade.

3. We’ll get a “zero to 1 billion” moment - The reason most people don’t know about AGI is because to date the impact on most people’s daily routine has been minimal. This will change once Apple and Google release zero latency LLMs with operating system access. Suddenly - overnight - the entire world - everyone with a smartphone - will have access to something that feels indistinguishable from having a “person in your pocket”. It won’t be long after that people realize how limited the need for humans will be in the imminent future.

The zero to 1 billion moment is what will kick of the greatest asset bubble in history. Analysts will start forecasting the profit margins of companies by assuming they employ zero workers. People will start valuing asteroid mining companies as if they’ll succeed at harvesting all of the precious minerals in Saturn’s rings. Real estate prices will catch a bid because everyone will start putting 2 and 2 together - and realize that a cure for aging is right around the corner - so the demand for prime real estate will increase rapidly while the supply usually provided by elders dying off will start evaporating.

Meanwhile, inequality will be skyrocketing because capital will have embarked on its final campaign to wipe out the value of labor once and for all…

The first new technology paradigm over which the government has little control

From the article:

Companies including OpenAI, Microsoft, and Google have all made overt attempts to encourage government to regulate AI. The cynical perspective on this is that they realize how indefensible the LLM technology is from a competitive standpoint and want to use the regulatory system to create a moat. This is probably true, but it’s also true that most people who believe AGI is coming think that society needs to figure out how to prepare for the chaos that will ensue thereafter (e.g. from mass job losses).

The examples Ben provides (nuclear, space, aviation) are fascinating to contemplate. There are two primary differences between everything that came before and AGI:

1. The rate of advancement far outpaces the speed of societal evolution and government capacity to respond

2. The internet ensures that anyone determined will be able to access AGI

One of the key insights I feel like I’ve had that doesn’t seem to be widespread - is the realization that only the United States - among ALL nations on Earth - has a corporate shadow government whose collective power dwarfs that of the federal and state bodies and the judiciary. I’ll never stop tooting my own horn about this. I predicted almost 7 years ago that the United States would become a “Big Technocracy”. I gave a talk about this to a Rotary group in Atlanta. If you look over the regulatory landscape in the US you will notice that virtually all impactful regulations might as well have been drafted by the C-suite of the companies who are regulated. This is not true in Europe, it’s not true in China. This sets up a dynamic that will likely exacerbate existing trends in the diverging paths of major economic zones.

The EU will react to AGI by over-regulating, making it difficult for companies to adapt. They will limit the ability of companies to employ humanoid robots. The net result will be the complete de-industrialization of the European Union. Japan, Korea and China - on the other hand - will automate as quickly as possible - because their citizens will trust that the government knows best and will end up providing for them. China in particular will avoid chaos caused by automation by being able to rapidly adjust the distribution of resources in response.

In the United States - particularly in red states - companies will be left to do as they see fit. Manufacturing will take off as labor costs plummet. Europe’s loss will be to the gain of Asia and the United States. There will be a lag between the surge in corporate profit margins and when society at large realizes we’re going to need to implement a new social welfare system. Thankfully, the US will be able to afford it because advanced technology acts like a tax charged (mostly) by the US (and likely soon China) on the rest of the world.

Blue states within the US will be inclined to add additional labor protection laws - and this will cause an exodus of business owners, entrepreneurs and the most ambitious members of society to red states.

Implications for national security

There are two primary attack vectors in the cyber world:

1. Vulnerabilities in tech

2. Vulnerabilities in people

AI as it currently exists already magnifies these vulnerabilities by multiple orders of magnitude. First, regarding the vulnerabilities in tech.

AI makes it easier to write code. AI can already - today - be asked to try and hack into some system. It will take a stab at producing the code - then a human will execute the code - then the human tells the AI whether it worked or not and gives the AI another chance if it didn’t.

By the end of this year - if it isn’t already happening - AI will be able to simply keep attacking systems on its own, relentlessly, indefinitely - with the only thing holding it back being compute power.

Vulnerabilities in people come primarily through “social hacks”. A hacker sets up an email address that looks like it belongs to an executive at the same company as the victim works for. They send them an email asking them to click a link to approve something - and that link infects the computer and gives access to the hacker.

AI makes this process easier by automating the process of writing the emails, identifying targets, etc. Again - right now this activity must be overseen by a human - but what will things look like when you have 1 billion AI agents constantly trying to impersonate people to get access to critical systems?

Another “social hack” that has only been made possible within the past 12 months is using AI to generate a real-time deep fake and mimic the voice of someone else. See this article from CNN:

My current view is that this problem will actually be solved with technology pretty quickly - because the ramifications of not solving it are enormous and may border on MAD (this is a surprisingly helpful heuristic for predicting the future).

But - you can imagine how being the first to AGI could lead to one country quickly gaining a compounding advantage vs. another.

This brings up an interesting aside. What happens to the demand for memory and compute if the technological solution to increased hacking ends up being having a billion agents constantly trying to hack and fix vulnerabilities? One way people defend against cyberattacks today is by employing people to attack their own systems to discover where they’re vulnerable. If you end up in a situation where billions of “good agents” are trying to constantly find exploits before billions of “bad agents” - it seems reasonable to me that this could end up using a material portion of total compute globally.

This post has hit my target length for these quicker “Bytes” and I need to get started on my other work. Let me know if you enjoy this format of responding to articles, there’s a bunch of other fascinating stuff to address in Ezra’s piece, including the ramifications of “free” satellites and the increased importance of Space; DeepSeek, China and export controls; why drug development will accelerate + a cure for aging - and more! I’ll get to it in the next few days.

So - I’ve added a new section to the page called XBytes. I am regularly bookmarking things on Twitter (now called X) that I find interesting. In this section I intend to just share the things that I have found over the prior week, along with brief commentary.

I will know whether the experiment is successful based on how many people subscribe/unsubscribe as a result. If you don’t like the new format and feel it’s spammy please let me know ASAP before I lose too many followers!

Now on with the first post…

Google is dying. Also, Google does 5 trillion searches annually?

Hat tip to @SupBagholder for the scoop - one of my favorite follows on X.

The death of Google has been greatly exaggerated. I’ll admit that I have probably dropped my own usage of Google Search by ~95% (seriously). In its place I use ChatGPT, Claude, and now Grok. However - everyone on X / Substack needs to remember that we are the few. Most people don’t know what an LLM is and have never tried ChatGPT (those are the facts). While ChatGPT now has hundreds of millions of users, Google has 7 products with 2 billion+ users:

1. Search

2. Android

3. Chrome

4. Gmail

5. YouTube

6. Play Store

7. Maps

What’s more, as I have described in countless posts on X and on 01Core - the killer LLM product is one that is connected directly to the operating system. One that can share screens, write emails, change settings, etc. ONLY the companies that own operating systems can deliver this.

Apple kneecapped Facebook after it realized that Facebook influenced more than 50% of app store downloads. Its fear of Facebook’s dominance of its own app store is what led it to create the “privacy” marketing campaign. As it turned out this ended up being a good thing for Facebook - but it also gave Apple what they wanted - which was dominant influence over which apps people download.

Today, Google is seeing more searches and more time spent searching because of LLMs. This means even if ChatGPT is taking share - the pie is growing fast enough to compensate. Search was up 13% YoY in the most recent quarter. Meanwhile, this is what Google’s EPS growth looks like:

Not exactly a portrait of a dying company is it?

The real question is whether ChatGPT might end up in a position like Facebook did in the early 2020s where it is having more influence over how people discover information than Google even on an Android device…

It’s possible, but both Apple and Google are in a position to “kneecap” ChatGPT just like Apple initially did to Meta. How?

They’ll release an LLM that is connected to the OS, and if need be - block ChatGPT or other LLMs from directly accessing apps on “privacy” grounds or some such theory. Would I start using Google as my go-to LLM if it were connected to everything on my phone? Obviously - so would everyone else.

What’s more? You might be shocked to hear that Google only gets 55% of revenue from Search. Cloud is the fastest growing segment, and YouTube is IMHO the most under monetized asset on Earth only after Android…

Google also has the lowest cost of delivering cloud-based AI (compared to Amazon and Microsoft).

Google trading for a multiple 15-20% lower than the S&P 500? Ridiculous.

The homebuilder / construction complex

This always has been and always will be a tech-focused newsletter. I’m pathologically optimistic about the future of technology so it’s naturally where I spend most of my time. Still, I may veer off topic from time to time. One of my favorite stock screens is to look at companies who have grown earnings at a higher clip over the past 5/10 years than their forward PE multiple. For example, a company that has grown earnings at 20%+ over the past five years, but which trades for a forward PE < 20. Or, a company that has grown earnings faster than 10% over the past 10 years, but trades for a forward PE < 10. Interestingly, there are ONLY TWO groups of companies who regularly appear on this list. There are of course other companies on it - but only two groups of businesses have almost every single constituent in the group appear. Those two groups are semiconductor capital equipment companies (Lam Research, Applied Materials, KLA Corp) and the construction complex (homebuilders like Pulte and DR Horton, and suppliers like Owens Corning and Carlisle).

See below the multiples of the homebuilders back in August of last year - BEFORE the recent 20-30% correction:

What’s more - the companies in this complex have disgusting historical financials. Here’s Carlisle companies actual / forward estimates and P&L over the past 8 years:

Note below that gross profit has almost doubled, meanwhile EPS has increased from $3.53 to $18, with estimates around $23 for next year. Also note the share repurchases:

What drives semiconductor capital equipment makers and the homebuilder complex to be the ONLY TWO groups of companies that have this weird situation where trailing EPS growth is > forward multiples? I have a theory…

They’re both extremely cyclical and they both have extremely powerful tailwinds.

Semicap tailwinds are obvious - semiconductors are the underling magic that drives/benefits most from the exponential growth of technology.

The homebuilder complex has a few tailwinds:

1. Consolidation

2. Difficulty of acquiring good lots at scale (smaller builders can’t compete)

3. Under supply of housing stock

4. Higher costs (contractors and raw material) wreck smaller builders

5. Blue-state to Red-state migration

6. Less rate sensitivity b/c can buy-down rates (again, small guys can’t)

Combine the above with homebuilders’ new asset light business model (optioning vs. owning lots) - and the complex as a basket looks more like a group of highly cyclical manufacturing companies than it does a bunch of real-estate speculators. Suppliers like Owens Corning and Carlisle have - literally - decades ahead of them to consolidate the market. Check out the stock charts of Waste Management and Republic Services if you want to see how beneficial it is to have an infinite amount of opportunities to roll up a market by purchasing at one EBITDA multiple, then using your scale to make the effective acquisition multiple 30%+ less…

We’re sitting on 7% mortgage rates and homebuilder margins are doing just fine. There’s always the possibility that the bottom falls out of the market - this might happen because rates drop (ironically) and a flood of existing homes comes to the market - or it might happen because we actually get a recession and stock market crash which impacts housing via the net-worth effect. However, what is obvious is that long-term earnings will go on to make new highs, and in the mean time you have a basket of companies that are more shareholder friendly than any others - by far - who gush cash and will use it to re-purchase shares throughout the down-cycle.

AI Cracks superbug problem in two days that took scientists years - an example of Acceleration

Another fun guy to follow on X is @Dr_Singularity. Here’s one of the more interesting use cases of AI I’ve run across recently

From the article:

“A complex problem that took microbiologists a decade to get to the bottom of has been solved in just two days by a new artificial intelligence (AI) tool.

Professor José R Penadés and his team at Imperial College London had spent years working out and proving why some superbugs are immune to antibiotics.

He gave "co-scientist" - a tool made by Google - a short prompt asking it about the core problem he had been investigating and it reached the same conclusion in 48 hours.

He told the BBC of his shock when he found what it had done, given his research was not published so could not have been found by the AI system in the public domain.

"I was shopping with somebody, I said, 'please leave me alone for an hour, I need to digest this thing,'" he told the Today programme, on BBC Radio Four.

"I wrote an email to Google to say, 'you have access to my computer, is that right?'", he added.

The tech giant confirmed it had not.

The full decade spent by the scientists also includes the time it took to prove the research, which itself was multiple years.

But they say, had they had the hypothesis at the start of the project, it would have saved years of work.”

This is an example of the type of acceleration we’re about to start seeing everywhere. I explained what this will feel like in a post on X (apologies if it’s a repeat for you):

If you think it is going to take humanity 50 years to figure something out (a cure for cancer, AGI, etc) - we might be just 11 years away. Before I walk through the math here are a few points worth noting:

1. Some types of compute efficiency (alternatively: the cost of intelligence) are improving by 50%+ per year.

2. Efficiency in analyzing biotech lab samples has improved by around 40% per year over the past 20 years (we can now test millions of samples with the same amount of human resources that were once capable of handling thousands).

3. However many people you think are working on a problem - the number is likely to be at least 10X higher. Why? Because most people working on compute efficiency, AI, robotics, etc are at least tangentially working on that problem. For example, no one would have thought about Nvidia as a "natural language processing company" a few years ago, but we now know that their GPUs are what unlocked ChatGPT. A similar argument will soon be made regarding their Chips' impact on biotech (as Jensen so regularly reminds us).

This is the beauty of general purpose technologies and the MEGA trend of everything turning into an engineering problem.

Now for the math. Let's be a tad more conservative than the above numbers imply might be possible and assume that we're accelerating the rate at which we're solving a problem by 30% per year. Put differently:

From 2024 to 2025 we make one year worth of progress.

From 2025 to 2026 we make 1.3 years worth of progress (30% more).

From 2026 to 2027 we make 1.69 years worth of progress (30% more than from 2025 to 2026).

The numbers blow up pretty quickly:

By year 7 we will have made more than 17 years worth of progress, and by year 10 we will make 10 years worth of progress in a single year, and will have made 42+ years worth of cumulative progress.

Most people need to dramatically shrink the timelines over which they expect the seemingly impossible to turn into the imminently probable.

Please remember, if you like this shorter form content and style please let me know - if you don’t like it it’s even more important to let me know!

Subscribe now

Share