The Number on the Term Sheet
Somewhere between the last private round and the S-1 filing, a valuation stops being a negotiating position and becomes a public promise. For OpenAI and Anthropic, that number is reportedly approaching $1 trillion — a figure that requires a specific and demanding set of assumptions to justify, and that will be tested, quarter by quarter, by investors who did not sign up for a decade-long science project.
That tension is the subject of a Fortune analysis invoking Clayton Christensen's theory of good money and bad money — a framework that has aged better than most business-school concepts because it describes a structural problem, not a cultural one.
What Christensen Actually Said
Christensen's argument, developed across his work on disruptive innovation, is straightforward: early-stage companies need patient capital that tolerates experimentation and accepts that the first business model is rarely the right one. When that capital is replaced — or overwhelmed — by money that demands near-term growth, companies are forced to scale prematurely, often locking in a model before they've learned what actually works.
The 'bad' in bad money isn't moral. It's structural. Investors with short time horizons aren't villains; they're operating rationally within their own constraints. The problem is the mismatch between what the capital needs and what the business requires.
SpaceX as the Counterexample
SpaceX is frequently invoked in this context, and not without reason. Elon Musk's ability to retain control of the company's capital structure — keeping it private, managing dilution carefully, and avoiding the quarterly earnings cycle — gave SpaceX the runway to fail expensively and repeatedly before achieving the unit economics that now make it defensible. That's not a replicable template for every deep-tech company, but it illustrates what patient capital actually looks like in practice.
OpenAI and Anthropic are on a different trajectory. Both have raised at valuations that imply the market-sizing work is largely done, even as the fundamental questions about AI monetization — enterprise contract durability, consumer retention, inference cost curves — remain genuinely open.
What a Trillion-Dollar Valuation Requires
To justify a $1 trillion valuation using conventional revenue multiples, a company would need to demonstrate either current revenues in the tens of billions with strong margin expansion, or a credible path to that scale within a timeframe that public market investors will accept. Neither OpenAI nor Anthropic has disclosed financials that publicly anchor those projections.
That doesn't mean the valuation is wrong. It means the assumptions required to make it right are doing a great deal of work — and that public market investors, unlike the venture funds that set the last round price, will be marking those assumptions to market every 90 days.
The IPO as Inflection Point
Private valuations are, in a meaningful sense, hypothetical. They reflect what a small number of sophisticated investors agreed to pay for a specific tranche of preferred stock, with specific liquidation preferences and anti-dilution provisions. The IPO strips most of that away. Common shareholders get the upside and the downside, without the structural protections that made the private round price rational for the investors who set it.
That's the moment Christensen's framework becomes most relevant. The capital that arrives at IPO is, almost by definition, impatient. It has a benchmark, a redemption cycle, and a fiduciary obligation to mark positions. If OpenAI and Anthropic's business models are still being discovered — and there's a reasonable argument that they are — the question worth asking is whether the public market is the right environment for that discovery process.
The answer may well be yes. But it's worth noting that the question is being asked less often than the valuation is being celebrated.