Apollo and Blackstone Close a $35 Billion AI Compute Deal — The Largest Private-Credit Transaction in History

Apollo Global Management and Blackstone have closed a $35 billion private-credit transaction — the largest of its kind on record — to finance Anthropic's compute expansion. The capital is being deployed through a special-purpose vehicle called AI XPV, which will purchase Google's custom TPU hardware and lease it back to Anthropic and other model developers under multi-year agreements.

How the Structure Works

AI XPV is a special-purpose vehicle financed by Apollo and Blackstone, with Broadcom handling the chip procurement and Google providing the underlying TPU design and a long-term offtake guarantee for non-Anthropic capacity. The capital structure layers senior debt against a residual equity claim held by the sponsors. Anthropic is the anchor customer, but the SPV is designed to onboard additional model developers over its life.

「This is private credit growing up. The transactions used to be a few hundred million for a single data center. This is a sector-scale deal,」 said a credit analyst at a major asset manager.

Why It Matters

Three things are happening at once. First, the cost of frontier-model training has grown faster than any single company's ability to fund it from operating cash flow, even with the backing of large cloud providers. Second, the private-credit market has matured to the point where it can absorb transactions of this scale, particularly when collateralized by hardware with a liquid secondary market. Third, hyperscalers have realized that they can monetize their custom-silicon investments more efficiently by selling capacity through structures like AI XPV than by running it themselves.

What the Other AI Labs Are Doing

The Anthropic-Apollo-Blackstone deal is the largest, but several similar structures are reportedly in market. OpenAI has explored variations, with discussions involving sovereign-wealth co-investors. Mistral has signaled interest in a smaller, Euro-denominated version with European private-credit sponsors. And xAI is rumored to be in early talks about a deal that would monetize its Memphis data-center build-out.

The Risks

Private-credit transactions of this size carry new risks. The TPU hardware has a useful life that depends on continued software optimization from Google. The lease economics assume that frontier-model demand continues to grow at recent rates. And the residual equity is exposed to a generation of accelerators becoming obsolete faster than expected. None of these risks are unmanageable, but they are large enough that the rating agencies have paid close attention.

If the structure works, it becomes a template for the next several years of AI infrastructure finance. If it doesn't, the AI boom will have produced its first major private-credit workout. Both possibilities are worth watching closely.