On January 2, 2026, analyst Benedict Evans described what he sees Sam Altman doing: "desperately swapping paper for assets and distribution and product and people before the music stops." The same day, the Financial Times reported that SpaceX, OpenAI, and Anthropic could all go public this year—and those three IPOs alone would outstrip the total proceeds from roughly 200 US IPOs in 2025. One analyst names the play. One newspaper quantifies it. Together they describe the largest paper-for-assets conversion in tech history.
The Conversion
Evans' claim is not abstract. The evidence is specific and recent:
Paper → Cash: SoftBank completed a $22.5 billion investment in OpenAI on December 31—New Year's Eve—finalizing its $40 billion commitment and taking an ~11% stake. OpenAI is seeking another $100 billion at an $830 billion valuation. Amazon is in talks to invest $10 billion at a $500 billion valuation, with OpenAI committing to use AWS Trainium chips.
Paper → IP: Disney's Sora deal is entirely in OpenAI stock warrants rather than a cash licensing fee. Disney gets the option to buy more shares beyond its $1 billion stake. As one observer noted: "Those are IOUs." OpenAI gets Disney's characters. Disney gets paper.
Paper → Distribution: OpenAI has sold 700,000+ ChatGPT licenses to 35 US public universities. Students and faculty used it 14 million times in September alone, beating Copilot. Once a generation of students builds workflows around ChatGPT, switching costs compound.
Paper → Talent: The Wall Street Journal reported on December 31 that OpenAI's stock-based compensation averaged $1.5 million per employee in 2025—roughly 7x what Google paid pre-IPO and 34x the average pre-IPO peer. OpenAI recently eliminated the six-month vesting cliff so new hires start earning equity immediately. If the paper holds value, this locks in the best AI researchers in the world. If it doesn't, OpenAI got their best years at a discount.
Paper → Infrastructure: Oracle is building data centers for OpenAI through the Stargate project. But Oracle pushed back completion dates from 2027 to 2028 due to labor and material shortages. Its stock fell 30% in Q4—the steepest quarterly drop since 2001—as the market priced in the distance between announced capacity and physical reality. Cash, IP, distribution, and talent can all be acquired with paper. Data centers cannot.
The Music
What could stop the music? Evans offered a specific historical parallel in the same interview: he compared Nvidia to Sun Microsystems.
Sun Microsystems built the servers that powered the dot-com boom. When the bubble burst, Sun's revenue collapsed from $18.3 billion to $11.1 billion in two years. The company never recovered. It was acquired by Oracle in 2010 for $7.4 billion—a fraction of its peak $200 billion market cap.
Evans isn't predicting a crash. He's observing a pattern: when the fundamental technology is real but the valuations have outrun the revenue, the smart play is to convert paper into things that retain value regardless. Data centers. Distribution. Brand. Talent. IP rights.
The same day, Meta's chief AI scientist Yann LeCun admitted to the Financial Times that Llama 4's "results were fudged a little bit"—the team used different models for different benchmarks to get better scores. AI valuations rest on capability benchmarks. If the benchmarks are gamed, the valuations are built on performance that doesn't exist. The paper behind the paper is unreliable.
The Three-IPO Market
Consider what this means. The entire 2025 IPO market—roughly 200 companies—would be outstripped by three. Not a sector. Not a wave. Three companies.
SpaceX. OpenAI. Anthropic.
One builds rockets. Two build language models. Together they represent such a concentration of private market value that their public offerings would constitute the majority of all new equity issuance in a year.
An IPO is the ultimate paper-to-cash conversion. Private valuations—negotiated between a company and a handful of investors—become public market prices, tested by millions of participants. The question is whether the conversion happens at current valuations or at lower ones. The urgency Evans describes suggests the participants themselves aren't sure.
The Brookfield Signal
On the same day Evans spoke and the FT reported the IPO pipeline, The Information reported that Brookfield—one of the world's largest alternative asset managers—is starting a cloud company called Radiant and a new $10 billion AI fund. Brookfield has said it plans to acquire up to $100 billion in land, data centers, and power assets for AI.
Brookfield manages $1 trillion in assets. It builds toll roads, pipelines, ports, and power plants. It does not chase bubbles. It builds the infrastructure that survives them.
The significance is in the contrast. OpenAI is converting paper into assets. Brookfield is converting assets into AI exposure. They're meeting in the middle—and the direction of each flow tells you something about what each participant believes.
OpenAI wants real things: cash, data centers, distribution, talent. Brookfield wants exposure to the AI economy: cloud customers, compute demand, power contracts. One is moving from abstract to concrete. The other is moving from concrete to abstract. The question is which direction carries more risk.
The Gap
Evans' word "desperately" doesn't mean Altman is panicking. It means the pace of conversion—$40 billion from SoftBank, Disney's IP for warrants, $100 billion raise, 700K university licenses, $1.5M/employee in stock comp—is faster than the pace of revenue.
OpenAI reportedly beat its 2025 revenue goal of $13 billion. That's real money. But the $830 billion valuation implies the market expects $80-100 billion in revenue within a few years. The distance between those numbers is the space in which the music plays.
If OpenAI reaches $80 billion, the conversion was smart business—locking in partnerships, distribution, and talent at prices that will look cheap in retrospect. If it doesn't, the conversion was still the right play: turning an unsustainable valuation into durable assets before the correction. Either way, converting paper into concrete is the rational move. That's what Evans observed. That's what January 2 showed.