Anthropic secures ~$36B private debt to lease Google TPU AI chips

This debt package is a milestone in how the AI race is being financed. A near‑trillion‑dollar lab is no longer funding compute primarily from equity; it is securitizing future TPU resale value and long‑term lease payments through a structure that looks more like aircraft finance than startup funding. That tells us two things: first, the capital demands of frontier training runs have blown past what even mega‑rounds can comfortably cover; second, Wall Street is now confident enough in AI’s cash‑flow prospects to underwrite tens of billions against very specialized hardware.

For the race to AGI, the implication is straightforward: access to compute at this scale is becoming a function of balance‑sheet engineering as much as model quality. Anthropic is effectively locking in a multi‑year TPU pipeline that will let it train and serve ever larger Claude and Mythos‑class models, while also deepening its dependence on Google’s silicon and Broadcom’s guarantees. OpenAI, Google and other labs will be under pressure to match this capacity, likely via similar private‑credit structures or government‑backed initiatives.

The systemic risk is that we’re building an AI infrastructure bubble on top of opaque, hardware‑backed credit. If model revenues or utilization disappoint, the losses won’t just hit the labs—they’ll ripple through private credit funds, pensions and insurers that bought into these deals. In that sense, this transaction doesn’t just accelerate Anthropic’s path toward AGI‑class systems; it also entwines frontier AI with the broader financial system in a way that will be hard to unwind.