Apollo Seeks Partners for $36B Debt Deal to Buy AI Chips for Anthropic

Apollo and Blackstone are reportedly जुटing additional investors for a roughly $36 billion debt financing package to help Anthropic expand AI infrastructure. The proceeds would fund purchases of Google custom TPUs that Anthropic will lease, underscoring continued large-scale capital spending in AI compute. The deal is supportive for AI infrastructure providers and private credit markets, though the article is still based on reports rather than a completed transaction.

Analysis

This is a financing story that matters less for the debt itself than for who gets embedded in the AI capex stack. If the market accepts long-duration, asset-backed-ish financing on compute infrastructure, it effectively lowers the hurdle rate for frontier-model scale-up and shifts bargaining power away from pure equity funding toward capital providers with physical collateral control. That is mildly constructive for hyperscale-adjacent infrastructure providers over the next 6-18 months, because the next wave of AI spend is likely to be funded through hybrid structures rather than straight corporate cash flow.

For Google, the strategic read-through is more important than the near-term financial contribution. Custom chips are only valuable if they become the default utility layer for third-party model training and inference; this arrangement signals demand validation and could improve utilization visibility for TPU capacity, which is the key driver of incremental economics. Second-order, it pressures the CUDA/NVIDIA ecosystem by proving that large buyers will accept non-NVIDIA architectures when financing and supply are bundled, though this is a multi-quarter adoption battle rather than an immediate share shift.

Blackstone’s angle is that private credit/structured finance is moving deeper into AI infrastructure, which can expand fee pools but also introduces concentration risk: if compute lease rates compress or chip generations turn over faster than expected, the collateral value can decay quickly. The biggest tail risk is technological obsolescence over 12-24 months—if the TPU generation financed today is leapfrogged, the asset looks less like infrastructure and more like stranded inventory. Near term, the catalyst is whether additional lenders join at tight spreads; if they do, it validates a repeatable funding template for AI capex and extends the trade across the sector.

Key Decisions for Investors

• Long GOOGL vs. semis basket for 3-6 months: express via long GOOGL / short equal-weighted AI hardware proxy on the thesis that TPU monetization and utilization visibility improve without requiring full-margin capex spend; stop if market starts pricing TPU as a true NVIDIA substitute.
• Add BX on pullbacks over 1-3 months: private credit and structured finance fees should compound if this becomes the template for AI infrastructure financing; risk/reward improves if additional tranches are syndicated, but size modestly because headline optics can overstate economics.
• Shortdated call spread on GOOGL into the next AI-infrastructure funding catalyst: limited upside participation if the market re-rates custom silicon optionality, with defined risk if investors dismiss the financing as non-recurring.
• Avoid chasing pure-play AI hardware names that rely on scarcity premium over the next 1-2 quarters: this structure is a reminder that buyers will seek financing and supply diversification, which can cap multiple expansion for the most crowded beneficiaries.
• Monitor credit spreads on AI-linked private debt and any TPU lease-rate disclosures over the next 6-12 months; if spreads tighten while utilization stays high, it supports adding to infra/credit exposure, but widening spreads would be an early warning of asset-value impairment.