This is an essay submission to the Dwarkesh competition (https://www.dwarkesh.com/p/blog-prize), essay prompt below (max response 1000 words):
What’s the most plausible story where foundation model companies actually start making money? If you consider each individual model as a company, then its profits may be able to pay back the training cost. But of course, if you don’t train a bigger, more expensive model immediately, then you stop making money after 3 months. So when does the profit start? Maybe at some point scaling will plateau, but if progress at the frontier has slowed down, then the combination of distillation and low switching costs (cloud margins result from high switching costs) makes it really easy for open source to catch up to the labs, eating into their margins. So how do the labs actually start making money?
Omaha as judgment day for AGI
Successful technology companies — the ones that actually “make money” — must transition from venture-backed growth (“revolutionary” mode) to durable, compounding cash flow. Call this end state legible to Berkshire Hathaway, or simply put, Berkshire-legible.
Apple is a canonical case of a Berkshire-legible technology company. Under Tim Cook, Steve Jobs’ successor as CEO, Apple became an operational juggernaut and the iPhone captured the market vis-a-vis recurring consumer packaged goods. Berkshire Hathaway took a position in Apple in 2016 and made it one of their largest holdings in the portfolio.
The question is: which, if any, of the foundation labs will make the same transition?
The answer has already been given to us: last year, in Q4 2025, Berkshire Hathaway disclosed a new $4.3 billion position in Alphabet (the parent company of Google), the only new position that quarter and immediately a top-10 holding. How do the other labs fare?
The modern foundation lab was birthed by Demis Hassabis’s solve intelligence, then solve everything else mandate. Artificial general intelligence (AGI) became legible to capital via market size: the total addressable market (TAM) would be the cognitive-labor share of GDP. As of early 2026, Silicon Valley largely considers the AGI revolution a resounding success: OpenAI and Anthropic together gross over $50 billion annualized and investors have underwritten hundreds of billions of dollars expecting trillion dollar run rates. Despite these high valuations, private financing rounds still seem to be oversubscribed, and thus there’s a tension in the Valley: what happens when these labs have to report to New York (and, perhaps, Omaha)?
At the same time, the foundation lab is still in revolutionary mode because the TAM of intelligence is still unbounded. These labs may be volatile, but they are revolutionary hypergrowth machines. Technology investors are wondering if we are still early.
Berkshire Hathaway refuses to underwrite revolutionary businesses. Alphabet is not a revolutionary pure-play lab but rather a search-advertising monopoly that acquired the revolutionary DeepMind in 2014. It then self-funded DeepMind as an R&D subsidiary for nearly a decade, and now integrates its models across its distribution surface. Independent of DeepMind, Alphabet was already a Berkshire-shaped investment; DeepMind just provides a de-risked R&D endeavor for Alphabet to maintain an excellent product moat with good-enough AI, something they have successfully executed for two decades – while also capturing the upside of the AGI revolution.
OpenAI is a revolutionary business. Projected cumulative cash burn through 2030 is $665 billion against $25 billion in run-rate revenue. In other words, it is increasing dependence on external capital, contrary to Berkshire-legible companies that self-fund R&D. In addition, the November 2023 OpenAI board episode – in which Sam Altman was fired and subsequently restored a few days later amidst a huge Board turnover – is the exact “founder-as-god” dependency Berkshire Hathaway refuses to underwrite. Warren Buffett has allegedly cited fellow value investing guru Peter Lynch’s idiot-proofing maxim:
“The simpler it is, the better I like it. When somebody says, ‘Any idiot could run this joint,’ that’s a plus as far as I’m concerned, because sooner or later any idiot probably is going to be running it.”
Alphabet’s Sundar Pichai is clearly no idiot and he rather seems to reflect an operational temperament compatible with Tim Cook’s Apple era.
The implications for the other labs (and Big Tech players) can be analyzed through Berkshire-legibility. Anthropic, the pure-play lab that most rivals OpenAI, provides a strong case of approaching durable compounding cash flow with its rise in enterprise revenue, setting up for strong software margins like Salesforce/Oracle/SAP, which are arguably Berkshire-legible. But the Valley’s switching between OpenAI and Anthropic every few months creates anxiety, and until that oscillation is resolved, it’s too early for Berkshire.
Meta’s open-source angle is more revolutionary, not cleanly mapping to increased cash flow but more of a directional market bet. The current fear is that their bet represents another Metaverse – a revolutionary attempt by “founder-as-god” Mark Zuckerberg, but again one Berkshire feels no need to underwrite until less idiot-proof.
Microsoft itself has long been considered investable by Berkshire (though Buffett has historically recused from Microsoft investment due to potential conflicts of interest with Bill Gates). CEO Satya Nadella arguably deserves an MVP Award for revolutionizing Microsoft’s distribution moat and converting it into a durable compounding edge, while also now owning a part of the upside of the AGI revolution with their OpenAI stake. They are likely the closest to Alphabet in Berkshire legibility, but their status as a foundation lab is shaky.
Both NVIDIA and Elon Musk’s empire may provide counterexamples to the validity of Berkshire-legibility. They are large institutions with heavy public market investment, and they have also strongly benefitted – both in revenue and in valuation – from the AGI revolution.
NVIDIA is intriguing because it’s a semiconductor chip company that is benefitting infrastructurally (often likened to a “railroad” if the foundation labs are the “trains”). Even if Berkshire concedes that Jensen Huang, CEO of NVIDIA, is less “founder-as-god” than sometimes perceived, Berkshire Hathaway famously refuses to invest in businesses it doesn’t understand and semiconductors are not Omaha’s bread-and-butter (they did buy TSMC – an NVIDIA competitor – in 2022, only to offload it a year later citing geopolitical risk).
Elon Musk’s xAI (and broader empire) would also fail the Berkshire Hathaway test due to founder dependency, but the synergistic dealmaking Musk can generate along his portfolio (xAI, SpaceX, Tesla, etc.) means investment may be more akin to that in a holding company with de-risked R&D than a direct revolutionary business.
NVIDIA and Elon Musk’s empire may not be Berkshire-legible now, but their investors have benefitted over the past few years and won’t mind the superior returns.
“What’s the most plausible story where foundation model companies actually start making money?” is a question of story, perception, and ultimately anxiety. Warren Buffet’s Berkshire Hathaway refuses to underwrite anxiety and they have cast their vote in the AGI race: Sundar Pichai’s Alphabet. Their story is simple: idiot-proof, durable, compounding cash-flow business with self-funded R&D.
(998 words)
Epilogue
The revolution has not ended, and perhaps it has just begun. This is the layer that Silicon Valley is currently underwriting heavily: artificial superintelligence labs such as Mira Murati’s Thinking Machines and Ilya Sutskever’s Safe Superintelligence, both OpenAI “spinouts” which raised at billion-dollar-plus valuations despite being pre-product and pre-revenue.
The $50 billion run rate combined between Anthropic and OpenAI has made the trillion dollar run rate legible to Silicon Valley. The next revolutionary bet is the $10 trillion dollar rate, where the foundation labs become more akin to sovereign entities rivaling the GDP of the US and China. This requires superintelligence, sometimes modeled as country’s participants working together – but arguably now through an exponential takeoff in AI capability and interaction. OpenAI’s Sam Altman and Anthropic’s Dario Amodei have an enviable task of pleasing larger, public market investors like their Big Tech counterparts, while also fending off revolutionary attempts from their kin.