Broadcom, Apollo, and Blackstone Launch $35B Platform to Build 20 Gigawatts of AI Compute for Anthropic and OpenAI
The new AI XPV Platform starts with a $35 billion debt tranche led by Apollo and targets more than 20 gigawatts of Broadcom XPU capacity through 2028. First up: over 1 gigawatt for Anthropic at Fluidstack sites starting mid-2026.
Broadcom, Apollo, and Blackstone have established the AI XPV Platform, a financing vehicle designed to deploy more than 20 gigawatts of AI compute capacity through 2028 — built on Broadcom’s custom XPU accelerators and networking gear for frontier labs including Anthropic and OpenAI.
The opening move is a $35 billion capital tranche led by Apollo in partnership with Blackstone and a group of global banks. The first deployment funds more than 1 gigawatt of compute for Anthropic, running in data center sites operated by cloud provider Fluidstack, with hardware landing from mid-2026.
Why this structure matters
This isn’t a venture round and it isn’t a hyperscaler writing checks from its balance sheet. It’s a chip-backed special purpose vehicle: the debt is secured against the Broadcom silicon itself, packaged for credit investors who want AI exposure without buying AI equity. Axios reports the Apollo-led deal as the clearest signal yet that AI compute is becoming its own asset class, financed like power plants and aircraft fleets rather than like startups.
The numbers explain the shift. Morgan Stanley projects AI-related debt issuance will approach $570 billion in 2026 alone — more than double earlier forecasts. Equity markets can’t absorb that. Private credit can, and Apollo and Blackstone are the two biggest players in private credit on the planet.
The Broadcom angle
For Broadcom, the platform locks in years of demand for its custom XPUs — the application-specific accelerators it co-designs with AI labs as an alternative to NVIDIA GPUs. Anthropic already runs significant workloads on Google’s Broadcom-built TPUs; this deal extends that custom-silicon strategy to dedicated capacity. Twenty gigawatts is staggering scale: roughly the output of twenty nuclear reactors, all feeding training and inference for a handful of frontier labs.
That both Anthropic and OpenAI are named as target customers is the detail worth sitting with. Fierce rivals at the model layer are now sharing financing infrastructure at the silicon layer, because nobody — not even companies with nine-figure monthly revenue — can fund this buildout alone.
What to watch
The platform’s first gigawatt for Anthropic starts deploying in mid-2026, which means the capacity crunch frontier labs keep citing won’t ease before 2027. And if chip-collateralized debt becomes the standard playbook, the next question is what happens to those collateral valuations when accelerator generations turn over every 18 months. Private credit is making a duration bet on silicon. We’re about to find out if that bet prices correctly.
Sources
Related reading
- AI Policy 1,178 AI Employees Ask Governments to Build a Kill Switch for Frontier AI — OpenAI and Anthropic Back It
- Big Tech Nvidia Weighs $250 Billion Guarantee So OpenAI Can Lease a 10-Gigawatt Ohio Campus
- AI Policy Amodei, Altman, and Hassabis Take the G7 Stage to Push U.S.-Led Global AI Governance