The Reckoning: OpenAI, Anthropic, and the Great AI Valuation Divergence
The speculation began in earnest on the evening of 20 May 2026. Across financial Twitter and a handful of Polymarket markets, a question that had been circulating in venture circles for 18 months crystallized into a binary bet: would Anthropic, the safety-focused AI laboratory co-founded by former OpenAI researchers, be worth more than its former parent by the end of June? The market put the odds at 83 percent in Anthropic's favour. By the morning of 21 May, OpenAI had confirmed the other half of the story. According to a source with direct knowledge of the matter, the company is aiming for a speedy initial public offering — one that could come within days of a filing.
The two developments are not coincidental. They represent the collision of two fundamentally different theories of AI development, each now arriving at its respective moment of truth simultaneously. OpenAI has chosen the path of maximum commercial velocity: rapid deployment, enterprise monetization, and now the transparency — and accountability — of public markets. Anthropic has bet on a slower build: rigorous alignment research, constrained but sustainable growth, and revenue projections that, per its own investor communications, will exceed $10.9 billion in the second quarter of 2026 with the company finally turning profitable for the first time.
The IPO Gambit
OpenAI's ambition to go public has been telegraphed for months, but the confirmation that a filing could arrive within days marks a qualitative shift. The company, which began life as a nonprofit research laboratory before restructuring around a commercial arm, has raised more than $40 billion since 2019 and was last valued at approximately $157 billion in a secondary transaction. An IPO would test whether public market investors — subject to quarterly earnings scrutiny and the unpredictable moods of retail traders — will value a company whose most transformative capabilities remain contested, whose profitability is uncertain, and whose governance structure, even after significant reform, remains unusual.
The timing is aggressive. Markets have grown cautious about AI valuations following a prolonged period in which revenue multiples for AI-adjacent companies bore little relationship to current earnings. OpenAI's filing will arrive while the company continues to burn cash at a significant rate — a fact that its advocates argue is the correct posture for a firm investing at the frontier of a technology that may reshape every major industry. Critics will note that the same argument has been made by every capital-intensive tech platform before it, and that public shareholders, unlike venture investors, cannot afford the time horizons required.
The mathematics conjecture that OpenAI announced on 20 May complicates the picture in ways that may cut either direction. The company's reasoning model claimed to have disproved a geometry conjecture unsolved since 1946 — a result that, if verified, would represent a genuine scientific contribution and a compelling demonstration of frontier capability. The verification came from mathematicians who had previously exposed an embarrassing OpenAI claim about another mathematical result, lending the announcement an unusual degree of credibility. It is the kind of landmark that public market prospectuses are built from. Whether it translates into the revenue trajectory that equity analysts will demand is another question.
Anthropic's Counterargument
Anthropic's position, by contrast, reads like a deliberate rebuttal to the OpenAI model. The company has consistently argued that sustainable commercialization and frontier research are not in tension — that the path to transformative AI runs through alignment, not around it. Its investors were told in communications circulating as of 20 May that the company will more than double quarterly revenue to approximately $10.9 billion in the second quarter of 2026 and record its first profitable quarter. Those are not the numbers of a research laboratory. They are the numbers of a company that has found product-market fit.
The Polymarket odds reflect a growing consensus that Anthropic's trajectory is sustainable in ways that OpenAI's may not be. The 83 percent probability assigned to Anthropic outperforming OpenAI on valuation by June 30 is a statement about investor expectations for the next six weeks — a compressed timeframe that suggests the market sees something structural rather than speculative in Anthropic's position. Whether that reflects genuine confidence in Anthropic's business or growing scepticism about OpenAI's governance and cost structure — or both — is impossible to disentangle from the data alone.
Anthropic's relationship with Amazon and Google, its primary cloud and investment partners, provides a structural buffer that OpenAI lacks. The hyperscaler partnerships give Anthropic distribution, computational infrastructure, and strategic cover in a market where compute availability is a meaningful competitive moat. OpenAI's dependence on Microsoft, its largest investor and cloud provider, creates a different set of incentives — ones that have occasionally surfaced as tension between the two companies, particularly around the pace of commercialization and the terms of intellectual property sharing.
The Valuation Problem
The core difficulty facing both companies is that AI valuation methodology remains primitive relative to the technology it purports to price. Traditional metrics — price-to-earnings ratios, revenue multiples, user growth curves — were designed for software companies with predictable cost structures and established monetization channels. AI companies of this generation do not fit the template. Their revenue may be growing rapidly, but their costs — primarily compute and talent — are growing at rates that can outpace revenue by an order of magnitude. Their products are improving in ways that may, over time, compress the competitive moat they currently enjoy. And their most valuable outputs, in some cases, are themselves difficult to define with the precision that public market analysts require.
The Polymarket market assigning a 12 percent probability to OpenAI announcing AGI before 2027 is instructive here. It suggests that a meaningful faction of speculative capital believes the next 18 months could produce a step-change in AI capability significant enough to warrant the label — a belief that, if it materializes, would dramatically restructure the valuation calculus for any company positioned near the frontier. If it does not materialize, the same capital faces a reckoning with the gap between hype and commercial reality that has undone previous generations of transformative technology bets.
What is striking is the degree to which the valuation question has been externalized — placed into markets like Polymarket where anyone can take a position — rather than resolved through the conventional channels of analyst coverage and institutional due diligence. This is partly a function of the opacity of private AI companies and partly a function of the speed at which the landscape is shifting. But it also reflects something deeper: the difficulty that even sophisticated investors face in pricing organizations whose outputs are simultaneously commercial products, scientific instruments, and potential civilizational infrastructure.
What Comes Next
The next sixty days will provide a significant data point. OpenAI's IPO filing — assuming it arrives as reported — will force the company to submit its financials to public scrutiny for the first time. The S-1 will reveal the scale of its losses, the composition of its revenue, and the degree to which its enterprise contracts are sticky or vulnerable to competitive pressure from Anthropic, Google DeepMind, Meta AI, and a cohort of well-funded open-source alternatives. The reaction of institutional investors to that document will tell us something real about the market's appetite for AI exposure at current valuations.
Anthropic's first profitable quarter will arrive in the same window. The milestone carries symbolic and practical weight: it suggests that the safety-first model can produce a sustainable business without the perpetual fundraising cycle that has defined frontier AI to date. Whether that sustainability is a feature or a constraint — whether Anthropic's profitability reflects prudent growth or an inability to outspend competitors on compute — will be debated in the weeks that follow.
The conjecture result, meanwhile, sits in a category of its own. It is a reminder that beneath the corporate maneuvering and market speculation, something genuinely novel is being built — and that the institutions trying to commercialize it are, in some cases, still figuring out what they have created. The mathematicians who verified the proof were careful to note that the result is narrow, that it applies to a specific domain, and that it does not represent the kind of general reasoning that AGI rhetoric implies. That caution is worth noting. It is also, perhaps, the most honest assessment available of where the frontier actually sits.
The AI industry's two dominant models are converging on their respective tests at the same moment. What those tests reveal about the compatibility of safety and velocity, of scientific ambition and commercial scale, will shape the industry's next chapter — and perhaps its next decade.
This article was structured around two parallel reporting tracks: OpenAI's IPO preparations and Anthropic's investor communications. Wire coverage of the math conjecture was cross-referenced against the technical assessment published alongside it.
Wire provenance
This editorial synthesis draws on the following public wire/social posts:
- http://reut.rs/4dZZZY7
- https://x.com/unusual_whales/status/1912790839200473089
- https://x.com/unusual_whales/status/1912718980401176585