Apollo Shops $36 Billion Debt Deal to Buy Google Chips for Anthropic

Apollo Global Management and Blackstone are arranging roughly $36 billion of debt financing to help Anthropic build AI infrastructure by buying Google-custom TPU chips that Anthropic will lease. Broadcom is backstopping payments on the largest portions of the transaction, reducing execution risk. The deal underscores continued large-scale private credit funding for AI capex and could support sentiment across AI infrastructure and private markets.

Analysis

This is less a one-off financing than the emergence of a quasi-utility model for frontier AI compute: capital is being pulled into a structure where debt funds hardware, hardware is leased, and the balance sheet risk is effectively warehoused by financial sponsors and upstream suppliers. That matters because it lowers the near-term capex burden for model developers, which should extend the runway for training/inference scaling even if equity markets stay selective on AI monetization. The second-order winner is anyone with scarce, defensible compute IP and financing capacity; the loser is the commodity end of the AI stack, where generic GPU exposure becomes relatively less differentiated.

For AVGO, the signal is stronger than simple TPU volume: sponsor-backed leasing implies multi-year utilization visibility and a higher probability of repeat orders, which improves the economics of custom silicon design/support relationships. For GOOGL, this is an underappreciated form of capital-light monetization of internal infrastructure and deepens TPU ecosystem lock-in without needing to carry all the end-customer credit risk directly. The credit market angle is that this transaction could become a template for off-balance-sheet AI infrastructure finance, which would likely compress spreads for similar asset-backed/contract-backed deals over the next 6-18 months.

The main risk is not demand; it is execution and residual value. If TPU lease economics fail to keep pace with rapid model architecture shifts, the financing could become a stranded-asset problem in 2-3 years, especially if utilization slips or next-gen chips obsolete the installed base faster than expected. Near term, the catalyst is deal syndication: successful placement should catalyze a read-through to more private credit/Apollo-led AI infra financings, while a poor syndication process would signal market fatigue and tighter terms across the sector.

Key Decisions for Investors

• Long AVGO on any post-news consolidation for a 3-6 month horizon; the market may still be underpricing the durability of custom-AI silicon demand. Risk/reward improves if the stock trades down on perceived supply concentration, since the actual implication is recurring financed demand.
• Add to GOOGL selectively versus hyperscaler peers over the next 1-3 months; this is a hidden monetization channel for TPU capex with limited incremental balance-sheet strain. Use any AI hardware skepticism to build exposure, with downside capped by diversified ad/cloud cash flows.
• Initiate a relative-value trade: long AVGO / short a basket of generic AI hardware beneficiaries over 1-2 quarters. The trade benefits if custom silicon and financed compute procurement prove more resilient than broad-based hardware enthusiasm.
• Watch APOS and BX for follow-on announcements over the next 3-9 months; if syndication broadens, both names gain fee-earning and structuring optionality. Consider tactical longs only on weakness, since the market may already be extrapolating a lot of transaction velocity.
• Avoid chasing high-beta AI credit until pricing is visible; the embedded risk is lease residual value and refinancing, which can surface only after 12-24 months. Prefer structures with supplier backstops or hard asset support over unsecured AI-related private credit.