Building the Other AI Giant: Anthropic’s Gigawatt Bet on Google Silicon
Anthropic’s latest compute commitments — 1GW+ of leases and up to a million Google TPUs — show how big the second horse in the AI race has become.
TL;DR — Anthropic is securing its first direct U.S. data-center leases — more than 1 gigawatt — with Google financial backing and up to 1 million Google TPUs, part of a pledged $50B U.S. compute build-out as it heads toward a possible IPO.
For all the attention on OpenAI, the AI build-out is a two-horse race. Anthropic’s latest compute commitments show just how big the second horse has become.
The deal
Anthropic has signed letters of intent for direct U.S. data-center leases totaling more than 1 gigawatt of compute, with Google providing financial backing for the lease payments (reported at up to $40 billion) and supplying up to 1 million Google TPUs. It is consistent with Anthropic’s pledge to invest $50 billion in U.S. computing infrastructure, with the vast majority sited domestically.
The expansion comes as Anthropic moves toward the public markets, having confidentially filed for an IPO at a reported ~$965 billion valuation, with an annualized revenue run-rate said to exceed $50 billion.
| Anthropic | OpenAI | |
|---|---|---|
| Primary silicon | Google TPU + Amazon Trainium | Nvidia + custom "Jalapeño" (Broadcom) |
| Lead cloud backer | Microsoft / Oracle | |
| Headline compute pledge | $50B U.S. infrastructure, 1GW+ leases | "Stargate" + 10GW custom racks |
What they said
"Anthropic’s choice to significantly expand its usage of TPUs reflects the strong price-performance and efficiency its teams have seen with TPUs for several years." — Thomas Kurian, CEO, Google Cloud
Why it matters
- TPUs are now a genuine Nvidia alternative at scale. A commitment of up to 1 million accelerators validates Google’s silicon for frontier training and inference.
- Capital is the moat. Google backstopping lease payments shows that financing — not just chips — increasingly decides who can build frontier AI.
- Two ecosystems are hardening. The Anthropic–Google and OpenAI–Nvidia/Broadcom camps are pulling the industry into competing stacks.
FAQ
How much compute is Anthropic securing?
Anthropic has letters of intent for direct U.S. data-center leases totaling more than 1 gigawatt, with up to 1 million Google TPUs and Google financial backing — part of a pledged $50 billion U.S. compute build-out.
Why is Anthropic using Google TPUs instead of Nvidia GPUs?
Google Cloud CEO Thomas Kurian said Anthropic cited the strong price-performance and efficiency it has seen from TPUs over several years. Anthropic uses a mix of Google TPUs and Amazon Trainium, distinguishing its stack from OpenAI’s Nvidia-centric approach.
Sources
- DataCenterDynamics — Google and Anthropic confirm ~1GW deal, up to 1 million TPUs
- Fortune — Anthropic confidentially files for IPO
Image: “Data Center” by Carl Lender, CC BY 2.0, via Wikimedia Commons.
← Back to all posts