OpenAI's revenue is $20B below prior reports, AI stocks sink
A revenue clarification rattled the AI trade on Thursday, a leaderboard startup nearly doubled its valuation, and Anthropic picked a side in critical-infrastructure defense. Here's the afternoon in brief.
OpenAI told investors its annualized revenue was approaching $50 billion at the end of September — far below the roughly $70 billion that had circulated from investor documents, and the market moved accordingly. The Financial Times broke the story Thursday, and the afternoon session saw Nvidia give up about 3%, Oracle drop 6%, and CoreWeave slide 8%, with AMD, Broadcom, Intel and Super Micro also in the red, according to CNBC. Part of the gap is definitional rather than demand-driven: CNBC reported that the earlier $68 billion figure included gross revenue from OpenAI's partners — money OpenAI itself does not book — while the $70 billion framing took hold in media coverage. Growth is still there; a person familiar with the investor materials told CNBC the deck showed 77% run-rate growth in the third quarter and 107% growth on the enterprise side. The take: for a lab that filed confidentially for an IPO with a 2027 target and is weighing a $1.2 trillion private round, a methodology haircut of this size lands exactly when underwriters are doing their homework — the numbers weren't fake, but the leak-and-refute cycle around OpenAI's finances is becoming its own risk factor. We dug into OpenAI's burn math last month — Deep Dive — OpenAI's cash gap shrank while its compute bill grew.
Arena, the company behind the AI industry's most-argued-over model leaderboard, raised $200 million at a $3.1 billion valuation — nearly double the $1.7 billion it carried in January. The Series B was co-led by Lightspeed Venture Partners and Khosla Ventures, with Salesforce Ventures, a16z, Felicis, Dell Technologies Capital and others joining, TechCrunch reports. The money is following usage: Arena says its platforms now serve more than 350 million sessions and 62 million votes, and that annualized revenue passed $100 million in June. Alongside the round, it launched an Alignment Index — scoring more than 20 frontier models on unauthorized action, false attribution and deceptive completion across 90,000 real-world agent sessions — with OpenAI's models topping the preliminary standings and Claude Opus 5.5 in sixth place. Why it matters: benchmarking stopped being a side project somewhere in the last year, and Arena is now valued like the gatekeeper it has become.
Anthropic launched the Anthropic Cyber Mission, headlined by a Critical Infrastructure Defense Program that puts frontier Claude models, on-site engineers and threat research in the hands of the firms that guard power grids, water systems and transport networks. Eleven founding partners signed on, among them CrowdStrike, Palo Alto Networks, Accenture, Deloitte, Booz Allen, Dragos and Rockwell Automation; a companion OSS Scanner service will run periodic scans of open-source projects with Anthropic's strongest models, free of charge, with reports sent without human review — the company says it expects a true-positive rate above 90%. Anthropic's own framing is a two-sided forecast: within two years AI should favor defense, but today the cost of finding vulnerabilities has already fallen while verification and fixing remain slow. The program also widens the split between Anthropic's commercial cyber push and its standing with Washington — we covered that tension in Deep Dive — Anthropic's guardrails cost it the Pentagon, court or not.
What to watch: whether OpenAI moves to standardize how it reports run-rate revenue before the IPO roadshow — and whether any rival lab tries to charge for what Arena gives away.
Is annualized run-rate still the right number to price frontier labs on, or should every lab be forced onto GAAP before it can talk valuation? Tell us in the comments.
