When the CFO of the world’s most valuable chipmaker declares that frontier AI labs will become the largest technology companies in history, we are not hearing a prediction. We are hearing a prayer. It is a prayer offered by a high priest of the compute altar, whose congregation—OpenAI, Anthropic, DeepMind—must scale their GPU cathedrals to prove their faith. But in the valley of bear markets and broken promises, I have learned to read such prophecies not for their vision, but for their subtext. Nvidia’s forecast is less about the future of AI and more about the present tense of its own balance sheet. The question we should be asking is not whether AI labs will be the largest companies, but whether the infrastructure of dependence they are building will leave them—and us—structurally unable to govern the very systems they create.
The context here is straightforward, yet often obscured by the glow of benchmark scores. Nvidia’s CFO, Colette Kress, has suggested that frontier AI labs could reach unprecedented scale, a statement that aligns perfectly with the company’s $3 trillion market capitalization. The logic is elegant: if AI labs become giant companies, they will buy more GPUs. But this circular reasoning ignores the uncomfortable physics of the AI economy. According to Epoch AI estimates, the stock of high-quality text data will be exhausted between 2026 and 2028. We are approaching a data wall, and the scaling laws that have driven progress since GPT-3 are hitting a hard ceiling. The industry’s answer has been to pivot toward synthetic data and test-time compute, but these are workarounds, not breakthroughs. The linear logic of "more compute equals more capability equals more revenue" is beginning to show cracks.
From my seat in Taipei, watching the cross-currents of Asian capital and Western innovation, I see a more profound issue than a mere data shortage. The real bottleneck is not compute; it is the architecture of trust. The scaling laws that Nvidia’s prophecy depends on are not just technical—they are economic and ethical. The cost structure of AI inference is fundamentally different from the marginal cost structure of traditional software. A GPT-4 level model costs between $0.03 and $0.06 per thousand tokens for input, and long-context scenarios push that higher. Traditional software scales with near-zero marginal cost, but AI’s unit economics are brutal. This is not a recipe for the high-margin, asset-light business model that defines today’s tech giants. The "maximum tech company" designation assumes a profitability profile that frontier AI labs, with their staggering compute burn rates, may never achieve.
In my 2022 burnout retreat in Yilan, I spent months journaling about the human need for trust in digital systems. That reflection crystallized into a belief that guides my work today: Trust is the only protocol that cannot be coded. Nvidia’s prophecy treats AI labs as if they are building on a foundation of pure mathematics, but they are actually building on a foundation of human vulnerability. Consider the regulatory landscape: the EU AI Act, China’s generative AI regulations, and the U.S. executive order on AI all impose compliance burdens that act as a tax on scale. The more these labs grow, the more they become targets for regulation, litigation, and ethical scrutiny. The copyright lawsuits against OpenAI are not a sideshow—they are a fundamental challenge to the data provenance that underpins every model. We built not for the peak, but for the valley, and the valley is where the legal and ethical challenges live.
Now, let me offer a contrarian angle that the market is ignoring. The prophecy assumes that frontier AI labs will emerge as the sole winners of the AI race. But the competitive landscape is far more complex. Microsoft, Google, and Amazon have all made deep investments in AI while retaining their core businesses. These incumbents possess distribution channels, user bases, and cash flows that AI labs simply do not have. The more likely outcome is not a coup, but a symbiosis—a world where AI labs supply the technology and incumbents supply the market access. Nvidia’s CFO is essentially betting on a future where the tail wags the dog, but the dog is still very much in charge. The relationship between OpenAI and Microsoft is a case in point: OpenAI provides the models, but Microsoft provides the enterprise distribution. This is not the stuff of a "largest tech company" narrative; it is the stuff of a dependent supplier relationship.
We don’t need more users; we need more stewards. The obsession with scale, with being the "largest," is a fundamentally centralized way of thinking. It flies in the face of the decentralized ethos that first drew many of us into this space. If we take Nvidia’s prophecy at face value, we are accepting a future where a handful of unaccountable labs control the most powerful technology ever created. That is not progress; it is a new form of feudalism. The real question is not whether these labs will be the largest companies, but whether they can be trusted to steward the technology they are building. The prophecy of scale is a distraction from the more urgent work of building governance frameworks that ensure these technologies serve humanity’s collective good. The future is not written in GPU orders. It is written in the values we choose to encode—or fail to encode—in the systems we build. And that is a choice no chipmaker can make for us.
