Anthropic tells investors it will be profitable for a second quarter
The week AI's biggest labs spent asking for permission to slow down ended with one of them telling shareholders the money is very good — and the number attached to that claim reframes the entire pacing debate.
Anthropic has told shareholders its adjusted operating income will be positive for a second straight quarter, with gross margins above 80%, according to the Financial Times, which sourced the report to multiple people with knowledge of the matter. The margin figure comes with the caveat that does most of the work: it is struck before revenue shared with distribution partners (Amazon chief among them) and before the cost of training new models. This is the clearest public number yet on what frontier inference actually earns — a roughly four-in-five gross margin on token sales — while the training bill, the genuinely ruinous line item, sits deliberately outside the frame. That distinction matters more than the profit headline: it means the company's path to durable profitability runs through inference volume and cache economics, not through the model launches that generate the news. The timing is not incidental either. Reuters notes Anthropic is heading toward an IPO that reporting has valued at up to $2 trillion, and a lab publicly asking Congress to let it decelerate while privately prepping investors for consecutive profitable quarters hands its skeptics the simplest possible question. We costed the inference side of that trade this morning — Tokens halved in price. The bill to make them didn't.
Z.ai's own stock fell more than 10% in Hong Kong after the company confirmed a plan to raise $5 billion or more through equity and debt, its second major raise in two months. The Wall Street Journal reports roughly $2 billion of it comes via a share placement, on top of the $4 billion round the GLM maker closed in July; Reuters detailed the structure last week — the placement alongside some $3 billion of zero-coupon convertible bonds. The sell reaction is the story: investors just watched China's most financeable frontier lab pay for compute by handing out paper at a discount, and the same day the country's other listed model name, MiniMax, slid with it. Chinese financial media tied the double drop to both the dilution and the overseas "deceleration" signals from Silicon Valley's slowdown week — two of the four Chinese frontier labs are public, and they are the only ones whose appetite for capital gets marked to market daily.
What to watch: whether an Anthropic S-1 ever has to reconcile the 80% gross-margin claim with the training spend it currently excludes — the IPO prospectus is where that arithmetic gets audited.
Does an 80% inference margin change your read on whether the AI business model works — or just move the goalposts to training costs? Tell us in the comments.
