Anthropic $65B Series H at $965B valuation

Anthropic's Series H caps a valuation curve with almost no precedent in private markets. The company sat at $61.5B in March 2025 ^[1], jumped to $183B at the September 2025 Series F ^[2], hit $380B at the February 2026 Series G ^[3], then 2.54x'd again to $965B three months later ^[4]. That is roughly a 16x repricing in fourteen months on the same underlying business. Fortune notes Series H rounds themselves are rare enough that the prior cohort is essentially Facebook, Lyft, Discord, and Slack ^[5], and that $965B sits at roughly the GDP of Switzerland, above the combined market cap of every U.S. airline, and above the entire U.S. defense budget ^[5]. The bull anchor is run-rate revenue: Anthropic crossed a $47B annualized pace in May 2026 ^[4], up from a $30B run-rate earlier in 2026 and roughly $10B in full-year 2025 revenue ^[11], with 1,000+ customers spending $1M+ a year and enterprise representing ~80% of the book ^[12]. The bear anchor is the speed itself — a 2.5x step-up in 90 days leaves very little margin for execution slippage, and Anthropic has already pushed its cash-flow-positive target out to 2028 as compute and training costs climbed ^[5].

The most underappreciated structural detail is that $15B of the $65B headline is previously committed hyperscaler capital, including a $5B tranche from Amazon announced in April ^[4]. Read alongside the partner list, Series H functions less like a traditional equity raise and more like a compute lock-in instrument. Anthropic concurrently disclosed agreements for 5 gigawatts of additional Amazon capacity, 5 gigawatts of next-generation Google/Broadcom TPU capacity, and GPU access through SpaceX's Colossus 1 and Colossus 2 superclusters ^[6]. Layered on top are strategic memory and chip relationships with Micron, Samsung, and SK hynix, which read as an attempt to lock in HBM and logic supply through the next training cycle ^[4]. In aggregate that is 10+ gigawatts of new compute contracted in a single announcement — comparable to the load of a mid-size U.S. state — and it reframes the round: investors are not just buying Claude's revenue multiple, they are co-financing the substrate Claude has to rent to keep growing. Time's coverage explicitly ties the raise to grid and energy knock-on effects ^[7], and that's the part that doesn't show up in the headline number.

For the first time, Anthropic is the most valuable private AI company. The $965B post-money sits above OpenAI's $852B March 2026 valuation ^[8], and Bloomberg's framing is blunt: Anthropic has eclipsed OpenAI ^[9]. The mechanical consequence is that OpenAI's next round, whenever it prices, now has to clear a tougher comp — crossover investors will benchmark to the Anthropic multiple, not the prior OpenAI mark. The strategic consequence is narrower and more interesting: enterprise CIOs who were running dual-vendor pilots now have a defensible reason to standardize on Claude, because the price tag certifies durability and the pre-IPO disclosures will give them more financial transparency than OpenAI currently provides. The order also rewires the capital stack. With GIC, Coatue, Capital Group, and D1 all co-leading, Series H pulls in the exact pool of investors that typically anchor IPO order books, which is consistent with reporting that this is the last private round before an October 23, 2026 listing ^[10]. The flip matters less as a status symbol than as a signal that the AI capital market now has two trillion-dollar gravity wells competing for the same compute, talent, and enterprise budgets.

Community pushback has converged on a sharp critique worth taking seriously. The most-cited comparison is Samsung: a roughly $1T conglomerate with hundreds of thousands of employees, owned factories, and decades of cash flow, sitting at a similar valuation to a ~5,000-person company that leases nearly all of its production infrastructure. The dot-com analogy is the second recurring beat — a 2.5x repricing in 90 days against unchanged fundamentals is the kind of pattern that historically precedes air pockets, and Fortune coverage flagged retail-investor risk into the IPO for exactly this reason ^[5]. Dragoneer's Marc Stad's own framing — that we are still in the earliest days of both development and commercialization — implicitly concedes that today's revenue does not justify today's price; the bet is on the slope of the curve ^[5]. What the leased-infra critique gets wrong, though, is treating compute leases as a weakness. The 10+ gigawatts of contracted capacity is the asset — it's just an asset that lives on Amazon, Google, and SpaceX balance sheets instead of Anthropic's. The real risk is not that Anthropic doesn't own the data centers; it's that the lease terms get repriced if model economics shift, and that is the variable retail buyers will have far less visibility into than the IPO valuation itself.