Amazon committed $33 billion to Anthropic in a transaction that functions less as venture capital and more as infrastructure forward-contracting. The deal locks Anthropic to Amazon Web Services compute for the next decade, with minimum annual spend thresholds that Bloomberg Intelligence estimates at $2.8 billion beginning in 2027. The investment follows Amazon's broader $200 billion AI infrastructure buildout, now the largest capital commitment in hyperscaler history.

The structure inverts traditional equity logic. Amazon receives equity at a $220 billion post-money valuation while Anthropic commits to multi-year GPU lease minimums that exceed the investment's net present value by 18-24 percent, according to three term sheet reviewers who requested anonymity. The arrangement effectively converts Amazon from investor to landlord, with Anthropic's compute dependency serving as both collateral and distribution lock. Amazon gains first access to Claude model weights for internal tooling, a provision that positions AWS customers as the primary integration surface for Anthropic's enterprise roadmap. The compute commitment covers H200 and next-generation Trainium instances, with 72 percent of the contracted capacity tied to Amazon's proprietary silicon rather than NVIDIA hardware.

The deal clarifies what strategic investment now means in AI infrastructure markets. Hyperscalers no longer fund model companies to gain exposure—they fund them to secure captive demand that justifies unprecedented capital expenditure. Amazon's $200 billion AI infrastructure budget requires utilization rates above 68 percent to meet internal return thresholds, per analysis from Bernstein Research. Anthropic's contracted minimums provide 1.4 percent of that utilization floor, but the signaling value matters more than the dollar contribution. The transaction establishes proof-of-concept for compute-backed equity, a financing structure that transforms venture rounds into capacity reservation systems. Microsoft's earlier OpenAI arrangement followed similar physics but lacked explicit minimum-spend provisions; Amazon's Anthropic terms codify the landlord model without ambiguity.

The dependency cuts both directions. Anthropic gains guaranteed access to frontier compute at a moment when GPU lead times stretch past 11 months for non-contracted buyers, but it sacrifices optionality. The company cannot shift workloads to Google Cloud or Microsoft Azure without breaching its AWS minimums, and those minimums escalate 12 percent annually through 2034. The equity valuation—$220 billion for a company with $4.2 billion in annualized revenue—prices in growth that requires multimodal dominance and enterprise penetration that Anthropic has yet to demonstrate outside developer communities. The compute lock also exposes Anthropic to Amazon's proprietary silicon roadmap; if Trainium underperforms NVIDIA's next architecture by more than 15 percent on key benchmarks, Anthropic faces margin compression it cannot escape contractually.

Allocators and operators should track three immediate follow-ons. First, Google's response positioning around DeepMind and Gemini enterprise distribution, expected before June earnings. Second, contracted utilization disclosures in Amazon's Q2 filing, due late July, which will clarify how much of the $200 billion infrastructure spend now carries revenue backstops. Third, Anthropic's next funding round, likely in Q4 2026, where valuation will test whether the compute dependency trades at a discount to unencumbered competitors. The Trainium deployment schedule also warrants attention; if Amazon can prove proprietary silicon performance parity by early 2027, the compute-backed equity model becomes replicable across the hyperscaler landscape.

The transaction establishes infrastructure dependency as the new venture currency. Amazon just bought $33 billion of influence that pays rent.