Anthropic confirmed on April 28 the close of a $5 billion Series F at an $80 billion post-money valuation, a level that more than doubles the $40 billion mark set in its August 2025 secondary tender. The round, reported first by The Information and confirmed in regulatory filings reviewed by Reuters and Bloomberg, was led by Lightspeed Venture Partners with a $750 million primary commitment, alongside expanded checks from Fidelity, GIC, and the Qatar Investment Authority. Amazon, already Anthropic’s largest strategic backer, contributed an additional $1.25 billion, taking its cumulative commitment to roughly $9.25 billion since 2023.
The numbers tell a familiar story for the frontier-model cohort: the cash is for compute, not headcount. Anthropic’s spend on training and inference capacity at AWS Trainium and Google TPU clusters has more than tripled year over year, and CEO Dario Amodei told staff in a memo seen by Bloomberg that the company is now planning around a “five-year horizon where compute is the binding constraint, not talent.”
Compute is now the line item that matters
Anthropic disclosed alongside the close that its forward-12-month revenue run rate crossed $5.4 billion as of March 31, up from $1.2 billion at the same point a year earlier. Roughly 70% comes from API usage, with the balance split between Claude Enterprise seats and a fast-growing channel of model-licensed-deployment deals — including the Palantir-led federal agreement disclosed in last week’s Q1 update and a previously unreported partnership with SAP for embedded Claude Code inside S/4HANA developer toolchains.
Even at that scale, the gross-margin picture remains tight. Bernstein analyst Mark Moerdler estimates Anthropic’s blended gross margin at 38–42% — well below the 70%-plus enjoyed by classic SaaS — because frontier-model inference still runs at a measurable per-token cost. “What this round buys is the right to keep training,” Moerdler wrote in an April 28 client note. “It does not yet buy structural margin.”
The valuation places Anthropic at roughly 15x forward revenue, slightly below OpenAI’s last-trade implied multiple of around 22x from its February tender, and meaningfully above Mistral and Cohere on the same basis. The round was upsized from an initial $3.5 billion target after demand from sovereign-wealth participants — a pattern that mirrors OpenAI’s 2024 SoftBank-anchored vehicle and the Stargate UAE consortium that broke ground in Abu Dhabi over the weekend.
Use of proceeds: Claude 4.7, agentic infrastructure, and a US footprint
Anthropic confirmed three concrete uses of the new capital. First, a multi-year compute commitment to Amazon Web Services for Project Rainier — the all-Trainium2/3 cluster being built in Indiana and slated to bring roughly 1.6 million accelerators online by mid-2027. Second, a parallel TPU-v7 reservation with Google Cloud, sized at “low-eleven-figure dollars” per a person familiar with the deal cited by The Information. Third, a formal expansion of the Anthropic API into agentic primitives — chained tool-use, persistent memory, and computer-use — packaged for enterprise availability ahead of the planned Claude 4.7 release window.
The fundraising also disclosed governance changes: a new long-term benefit trust seat for a designated safety-research representative, and a five-year lockup on Amazon’s secondary disposition rights — both of which shipped in response to questions raised by the Federal Trade Commission’s December 2025 inquiry into AI-cloud tying arrangements.
What it signals for the rest of the field
Anthropic’s mark sets a fresh anchor for the broader generative-AI capital stack. xAI’s reported $50 billion target and Mistral’s expected European Series D — both still in syndication, per the Financial Times — are now negotiated against a higher comparable. And on the buyer side, the round sharpens an awkward question for hyperscaler customers: every dollar booked into Project Rainier is a dollar Amazon directly underwrites, even as Anthropic remains AWS’s flagship third-party model. As Goldman Sachs’s Kash Rangan put it on a Tuesday client call, “It is increasingly hard to draw a clean line between what is a customer, what is a supplier, and what is an investee in the frontier-model economy.”
Discussion
Sign in to join the discussion.
No comments yet. Be the first to share your thoughts.