Since Anthropic filed its confidential S-1 with the SEC earlier this week, the conversation has moved past the headline numbers — and the real test is just getting started.

The Number That Decides Everything

Harrison Rolfes, analyst at PitchBook, told CNBC it plainly: the figure that determines whether Anthropic's IPO succeeds or collapses isn't the $965 billion valuation. It isn't the $47 billion annualized revenue run rate. It's gross margin — the percentage of revenue left after paying the staggering cost of running frontier AI infrastructure.

"No one outside Anthropic has ever seen it," Rolfes said, "and it will either validate or collapse the entire narrative the private markets have been pricing for three years."

Private investors have pumped this company to nearly a trillion-dollar valuation without ever seeing how much profit is actually left after the bills are paid. The public markets won't be that patient.

Gil Luria, head of technology research at D.A. Davidson, told CNBC that Anthropic's growth trajectory is genuinely impressive. But growth means nothing if you're burning faster than you're earning.

The Investor Overlap Nobody Talked About

According to a Wired analysis of PitchBook data, approximately 90 venture capital firms have placed bets on BOTH Anthropic and OpenAI simultaneously. Sequoia Capital, Greylock, Founders Fund, Redpoint Ventures — they're in both camps. OpenAI shares about 42 percent of its investor base with Anthropic. About a third of Anthropic's investors also hold OpenAI stakes.

Tom Nicholas, a Harvard Business School professor and author of VC: An American History, told Wired this is "unusual" in venture capital.

The smart money isn't convinced either company wins. They're hedging. Both directions. At the same time.

How does that square with the breathless "AI race" narrative mainstream media keeps selling? It doesn't. These aren't fans in opposing stadium sections — they're the same people writing checks to both teams.

Anthropic Calls for a Slowdown — While Racing to Go Public

This week Anthropic also published a blog post calling for a global slowdown — possibly even a temporary pause — of AI development. The stated rationale: AI systems could soon be capable of building their own successors, which Anthropic's safety research institute (established in March) says could arrive "sooner than most institutions are prepared for."

Anthropic is not calling for this to happen now. It's proposing the framework: multiple frontier labs, in multiple countries, agreeing to stop under the same conditions, with verification mechanisms to ensure nobody cheats.

That's a legitimate concern worth taking seriously. But the timing is impossible to ignore.

The same week Anthropic files for an IPO and touts a $47 billion revenue run rate, it's also warning the world its technology might be dangerous enough to require a global pause. Critics quoted by Engadget aren't letting that slide — they're calling it a marketing play. The argument: warning about your own tech makes you look like the responsible adult in the room, which is a competitive advantage when you're trying to win enterprise contracts and government favor.

Anthropic's limited release of its cybersecurity AI model has fueled this critique. The company says it restricted access due to the model's ability to rapidly identify software vulnerabilities. Skeptics say it's a hype strategy dressed up as caution.

Both things can be true at once. The safety concerns can be genuine AND conveniently timed. Anthropic doesn't get a pass on the contradiction simply because it acknowledges the risk.

What the IPO Actually Tests

Anthropic's Daniela Amodei — who leads the business side while brother Dario leads the technical — reportedly told TechCrunch she's shrugging off doubts about AI's returns ahead of the IPO. That's the right public posture. But shrugging doesn't make the gross margin question disappear.

The S-1, once public, will force disclosure of exactly how expensive it is to run Claude at scale. Compute costs for frontier AI are enormous. If Anthropic is spending 70 or 80 cents of every revenue dollar on infrastructure, that $965 billion valuation is going to face a very uncomfortable conversation in the roadshow.

OpenAI is reportedly still working toward its own IPO. Both listings represent a major test of tech-era valuations and will be closely watched by investors and analysts alike.

What Investors Need to Know

The AI boom has minted paper trillions in private markets where nobody had to show their work. The gross margin line in Anthropic's S-1 will be the first time any pure-play frontier AI company has had to show theirs — in public, under SEC scrutiny, with real money on the line.

Regular investors considering buying in should understand one thing: the insiders already hedged. The VC firms who built these valuations have bets on both sides. If you're buying only Anthropic — or only OpenAI — you're taking more concentrated risk than the people who actually built these companies.

That's not a reason to stay out. It's a reason to wait for the gross margin number.