Apollo and Blackstone Close $35B SPV for Anthropic's Alphabet Chip Access

Apollo Global Management and Blackstone have closed a $35 billion debt package financing Anthropic's access to Google's custom Tensor Processing Units through a special purpose vehicle, with $22 billion arranged as senior bank debt. The structure marks the first time alternative asset managers have financed AI compute capacity as a standalone asset class rather than through direct equity positions in the operating company.

The SPV sits between Alphabet's chip manufacturing capacity and Anthropic's training requirements, effectively creating a leasing mechanism for compute infrastructure. Anthropic commits to multi-year chip access under fixed terms while the SPV services debt obligations through those contracted payments. Banks led the $22 billion loan tranche at rates understood to be 175-200 basis points over SOFR, with Apollo and Blackstone providing approximately $13 billion in subordinated debt across their private credit and infrastructure vehicles.

The structure solves three problems simultaneously. Anthropic secures guaranteed compute capacity without balance sheet expansion. Alphabet monetizes chip production capacity ahead of actual deployment, converting future hardware revenue into immediate capital. Apollo and Blackstone access AI infrastructure exposure without betting on any single model provider's commercial success, since the debt is collateralized by the chips themselves and Anthropic's binding access agreements.

This matters because it creates a precedent for financing AI compute as infrastructure rather than technology. If the structure performs as modeled, expect similar vehicles for other frontier labs' GPU and TPU requirements within six months. Microsoft and Amazon both operate chip design programs comparable to Google's TPU ecosystem. OpenAI, xAI, and Mistral all face identical compute financing constraints. The $35 billion deal size suggests allocators now view contracted AI compute capacity as worthy of infrastructure-grade leverage ratios, somewhere between data center debt and equipment finance.

The second-order effect lands in private credit allocations. Apollo and Blackstone are creating yield on an asset class that didn't exist eighteen months ago, with returns likely in the 9-11% range on the junior debt. That's 300-400 basis points above comparable data center mezzanine loans, pricing in both novelty risk and the possibility that Anthropic's chip utilization falls short of projections. But the contracts are take-or-pay, meaning Anthropic owes regardless of actual usage, and the chips carry residual value to other hyperscale buyers if the SPV needs to restructure.

Operators and allocators should monitor Anthropic's next funding round, expected in Q2 2025, for signals on whether the company views this as bridge financing or permanent capital structure. Watch for similar announcements from Microsoft regarding OpenAI's compute needs, likely structured through Azure infrastructure vehicles. The private credit arms of Ares, KKR, and Oaktree are understood to be developing comparable financing frameworks, with term sheets circulating among Tier 1 labs as of late March.

The $22 billion bank portion drew fifteen lenders, none holding more than $2.3 billion, suggesting broad syndication appetite despite the structure's novelty. That syndication depth is the tell. Banks are treating AI compute infrastructure as they would toll road concessions or LNG terminals—contracted cash flows against hard assets, regardless of the underlying technology's ultimate winners.