Anthropic at $2 Trillion: Right About AI, Wrong About the Price?
The technology is real. The question is who captures the value — and Buffett answered this one about cars and planes a century ago.
Anthropic is heading for a public listing at a reported $2 trillion.
I want to start with what I actually believe, because it matters.
This technology is real, and it will unlock an enormous amount of economic value. Anthropic went from $386m of revenue in 2024 to $4.59bn in 2025 — twelve-fold growth. Q2 2026 revenue hit $11.5bn, with two consecutive profitable quarters and 300,000+ business customers.
Anyone calling AI “all hype” isn’t paying attention.
So the interesting question isn’t whether AI creates value. It’s who captures it — and whether the price leaves anything for you.
The most under-discussed fact in this debate is Anthropic itself
Anthropic was founded in 2021. Within roughly four years it went from nothing to overtaking OpenAI — the company that invented this market, with the multi-year head start, the Microsoft partnership, and the ChatGPT brand.
On enterprise LLM API spend, Anthropic now leads with about 40% share. OpenAI has slipped to the high-20s.
Impressive. But read it again as an investor:
If a newcomer can reach parity with the leader in five years, the leader did not have a moat.
That is what a moat is supposed to mean — protection against exactly this. We have instead watched model leadership change hands repeatedly between OpenAI, Anthropic, Google, DeepSeek and xAI, with benchmark leads lasting months, not years.
The cost structure says the same thing. Anthropic spent $7.33bn on compute in 2025 — 1.6× its entire revenue — and has committed roughly $518bn to infrastructure over the next decade, about 80% reportedly non-cancelable.
That is usually described as a moat. I’d call it the opposite: the price of staying in the game, which every serious competitor is also paying. Capital intensity that everyone must match doesn’t protect returns — it raises the stakes and converts flexible costs into fixed ones owed even if pricing collapses.
Buffett’s lesson, not mine
This framing belongs to Warren Buffett, who has made the argument for decades — most memorably at Sun Valley in 1999, at the height of the dot-com bubble.
His automobile example: roughly 2,000 car companies were founded in America. Around 780 are documented between 1895 and 1969; the count peaked at 206 manufacturers in 1908, fell to 24 by 1929 and just 8 by 1940.
And the early leaders? Duryea, Olds, Studebaker, Packard, Pierce-Arrow. Almost none are what you’d have wanted to own. Ford, GM and Chrysler weren’t the obvious early champions either.
His aviation version, from the 2007 Berkshire Hathaway letter, is blunter:
““The worst sort of business is one that grows rapidly, requires significant capital to engender the growth, and then earns little or no money. Think airlines.”
““If a farsighted capitalist had been present at Kitty Hawk, he would have done his successors a huge favor by shooting Orville down.”
Airlines went from ~7 carriers in 1930 to 31 by 1950 and back to ~7 majors today. The industry transformed the world — and its investors in aggregate lost money from its birth at Kitty Hawk. Pan Am, TWA, Eastern, Braniff: household names, all gone.
Buffett’s point is not that aviation was a bad idea. It’s that the industry and its investors are different questions. A transformative technology attracts enormous capital, and that capital competing for the same prize is what erodes the returns.
Now the arithmetic
Against 2025 revenue of $4.59bn, $2 trillion is 435× sales. Against the current run-rate — $65bn in July 2026 — roughly 31×. Against Anthropic’s own 2028 projection of $190–200bn, about 10×.
For context: NVIDIA trades at ~18–21× trailing sales. OpenAI — valued at $852bn in March, now targeting $1.4 trillion on at least $30bn of fresh capital — is in a similar range. Even Cisco at the absolute top of the dot-com bubble was only mid-teens to low-twenties. And Cisco was profitable, dominant, and entirely real. It just got bought at a price that assumed the good times would never normalise.
Where I actually land
The bull case isn’t stupid. If you genuinely believe Anthropic reaches $200bn of revenue by 2028 at software-like margins, $2T is about a 10× multiple on 2028 sales — defensible.
But that means you’re underwriting something very specific: that this company holds its position through a shakeout that hasn’t happened yet, converts $518bn of fixed commitments into profit, and defends margins against rivals who have repeatedly closed the gap in months.
The technology will work. Anthropic is an exceptional company. Neither of those settles the valuation.
“A transformative industry and a good investment are different things. The car industry transformed the world and destroyed most of its early investors’ capital. It is entirely possible to be right about AI and wrong about the price.
Not a prediction. A description of the risk in a 435× trailing multiple — and a reminder that the company which ends up owning an industry is often not the one that was first, or biggest, at the start.
If you disagree on the moat point, I’d genuinely like to hear why — that’s the crux of it.
Sources: Anthropic’s confidential S-1 as reported by Reuters; Berkshire Hathaway 2007 letter; Buffett’s 1999 Sun Valley remarks; Federal Reserve Bank of Kansas City research on US automobile entry/exit (1895–1969); Reuters/Bloomberg on OpenAI’s $30bn round (29 Sep 2026); companiesmarketcap.com for market data.
