Apollo, Blackstone Close $35B AI Chip Loan for Anthropic's Alphabet Infrastructure Access

Apollo Global Management and Blackstone finalized a $35 billion debt package financing Anthropic's access to Alphabet's custom AI chips, the largest AI infrastructure financing on record. The arrangement uses a first-of-its-kind off-the-floor SPV structure that treats chip access as a financed asset rather than an operating expense or equity dilution. Anthropic pays down the facility over time as it consumes Alphabet's TPU capacity, insulating its cap table while securing long-term silicon supply.

The structure separates infrastructure procurement from balance-sheet risk. Anthropic receives guaranteed access to Alphabet's tensor processing units without upfront capital expenditure or vendor lock-in clauses that would typically accompany such scale. Apollo and Blackstone hold the debt against future revenue tied to AI model deployment, effectively treating compute capacity as collateral. The banks priced the facility at SOFR plus 275 basis points, reflecting confidence in Anthropic's commercial trajectory and the strategic value of silicon supply chains. Alphabet benefits by monetizing excess TPU inventory without direct customer financing, keeping the transaction off its own books.

This matters because it opens a new asset class for infrastructure debt. If custom silicon access can be securitized and financed outside traditional vendor credit lines, hyperscalers and AI labs gain leverage against chip manufacturers. The SPV model also signals that credit markets now view AI compute pipelines as predictable cash-flow generators, not speculative bets. Alphabet's willingness to participate—albeit indirectly—suggests it sees more margin in volume TPU sales than in equity stakes or revenue-sharing arrangements with model developers. Apollo and Blackstone are effectively underwriting the thesis that inference workloads at scale are bankable, which has direct implications for how Nvidia, AMD, and new entrants price long-term capacity contracts.

Operators should watch for copycat structures in the next six to nine months, particularly among Tier 2 AI labs that lack Anthropic's fundraising muscle but need multi-year chip commitments. If this SPV model works, expect similar deals for inference-specific hardware, edge deployment infrastructure, and data center build-outs tied to model hosting. Alphabet's Q2 earnings call in late April will reveal whether it books this as a one-time inventory reduction or the start of a new TPU-as-a-service revenue line. Apollo's infrastructure fund disclosures in July should show whether the debt is held at par or if early refinancing talk has started.

The real tell will be whether Amazon or Microsoft offer competing off-balance-sheet chip financing to their own AI customers by year-end, turning cloud vendor relationships into structured credit plays.