Apollo Global Management is marketing a $36 billion debt financing to fund the acquisition of Google-manufactured AI compute chips destined for Anthropic, according to people familiar with the matter. The deal represents the largest single debt raise tied to AI infrastructure this cycle and marks private credit's entry into hardware ownership at scale.

The financing will back the purchase of tensor processing units from Google, which Anthropic will lease under multi-year contracts to power its Claude large language model training and inference workloads. Apollo is structuring the debt as asset-backed, using the lease receivables and the underlying chip collateral as security. Pricing has not been disclosed, but comparable infrastructure-backed private credit in AI data centers has cleared 800 to 950 basis points over SOFR in recent months. The debt is being shopped to a mix of insurance balance sheets, business development companies, and sovereign wealth funds with appetite for illiquid, yield-bearing infrastructure exposure.

The transaction reveals two inflection points. First, AI model builders are shifting from outright capital expenditure to lease-and-consume structures, freeing balance sheets while locking in multi-year compute guarantees. Anthropic's forward commitment here likely spans three to five years, giving Apollo visibility into cash flows that resemble telecom tower leases more than traditional tech vendor paper. Second, private credit allocators are no longer treating AI as a software venture bet—they are underwriting it as infrastructure, with hard assets, contracted cash flows, and residual value in secondary chip markets. Apollo's willingness to finance $36 billion against chips, not equity, suggests the firm sees the hardware as durable collateral even if Anthropic's model leadership fades.

The structure also compresses risk for Anthropic. Rather than raise dilutive equity or sell forward compute contracts at a discount, the company secures chips at locked-in pricing and converts the obligation into an operating lease. If Apollo holds the debt to maturity, Anthropic avoids refinancing risk and gains flexibility to scale or wind down workloads without stranded asset exposure. For Apollo, the deal layers into its broader infrastructure credit strategy, which has deployed $92 billion across asset-backed lending since 2021, including data centers, fiber networks, and renewable energy assets.

Operators should monitor whether other hyperscale AI labs—OpenAI, xAI, Mistral—pursue similar sale-leaseback or forward-lease structures in the next six to nine months. If the model becomes common, it will pull forward demand for GPUs and TPUs, tightening supply and pushing up residual values, which in turn improves collateral coverage for lenders. Watch also whether Apollo syndicates portions of the debt or holds it on balance sheet; retention would signal confidence in Anthropic's ability to meet lease obligations even if inference pricing compresses.

Google benefits twice: it monetizes chip production at scale and locks Anthropic into its ecosystem, reducing the risk that the model builder shifts to Nvidia or custom silicon mid-contract.