OpenAI Blinks on the IPO. Anthropic Sets the Price.

Sam Altman said the quiet part out loud. Going public in 2026, he told investors this week, is "ill-advised." OpenAI will raise $1.2 to $1.5 trillion in private markets instead. Anthropic, meanwhile, is targeting an October IPO at roughly $2 trillion.

The order of these two events matters more than either one alone.

For the past five years, frontier AI valuations have been set by private transactions: a funding round here, a secondary market there, each number reflecting what a handful of sophisticated investors were willing to pay at a specific moment. None of those numbers created a permanent benchmark. They were data points, not anchors.

Anthropic's IPO changes that. The moment it prices, every AI company acquires a public-market comparator for the first time: application-layer software, inference infrastructure, foundation-model startups, AI-native fintechs. Not a private round negotiated behind closed doors. A price set by markets, visible to everyone, subject to daily revision, and used as a baseline by every analyst writing a comp sheet from October onward.

OpenAI chose not to create that benchmark. Whether Altman's concern about timing is principled or tactical is a separate question. What matters structurally is that by staying private, OpenAI ceded to Anthropic the right to define what frontier AI infrastructure is worth in public markets. That is an unusual gift.

The math runs in two directions

A $2 trillion public-market valuation for Anthropic anchors AI infrastructure high. That benefits anyone who holds private AI infrastructure at a discount to that anchor — which describes a large portion of venture portfolios built over the past three years.

But the math also runs the other way. A high public-market anchor for foundation models raises the implicit bar for application-layer AI companies to justify their valuations. If the model itself is worth $2T, the application running on top of it needs a clear theory for why it commands a premium over the underlying infrastructure. The theory: proprietary data, distribution, regulatory moats. That becomes the question every application-layer company must answer in pitch meetings starting in November.

What Brazil reads from this

Brazilian AI companies have operated in a peculiar valuation vacuum. The most active AI deployment in the country runs through banking infrastructure, principally Itaú, BTG, and Nubank, where AI is embedded in credit decisioning, fraud detection, and customer service but never priced as a standalone asset. Standalone Brazilian AI-native companies, of which there are fewer than most observers assume, have priced off European and US private-round comps with wide uncertainty bands.

Anthropic's IPO narrows those bands. It does not resolve them. A Brazilian AI-native company is not Anthropic, and multiple contraction from frontier to application layer is real. But it gives investors a public-market starting point they have never had before.

The specific number that matters most is not the $2T headline. It is the revenue multiple implied by that number at the moment Anthropic files its S-1. That multiple, once visible, becomes the ceiling from which Brazilian AI application companies are discounted. That is how comp sheets work. That is why the order of events — Anthropic first, OpenAI later — matters.

Private capital's position

OpenAI's decision to stay private is, from a private capital perspective, structurally bullish. It means one of the two most-watched AI companies will continue to raise at private valuations for at least another year, keeping deal flow and secondary transactions in private markets rather than routing them to public investors.

The combined effect creates a market where the two largest AI valuations in the world are simultaneously anchoring both the public and private sides of the same asset class: Anthropic public at roughly $2T, OpenAI private at $1.2 to $1.5T. That is not a normal market structure. It benefits investors with access to both.

For investors whose exposure is concentrated in Brazil and LatAm, the practical implication is portfolio calibration. AI infrastructure exposure through application-layer companies with genuine moats: data ownership, regulatory access, distribution. That exposure is now benchmarked against a public-market number. Whether that benchmark proves durable is a separate question. That it now exists changes the analysis.