Anthropic is in talks with Microsoft to use its Maia 200 AI chip, CNBC reports. No deal has been signed yet, but conversations are underway.
Why This Matters
Microsoft unveiled the Maia 200 in January — a custom AI chip that CEO Satya Nadella says delivers over 30 percent better tokens-per-dollar than the best silicon currently in Azure’s fleet. So far, Microsoft has only used the chip internally. Anthropic would be one of the first external customers.
For Microsoft, this would be a prestige win. In the race for custom AI silicon, they’re trailing behind Amazon (Trainium) and Google (TPUs). Landing Anthropic — one of the hottest AI companies on the planet — would change the narrative.
Anthropic’s Compute Strategy Keeps Getting Broader
What strikes me most is how Anthropic isn’t betting on a single provider. They’re building an entire network.
The current lineup:
- SpaceX/xAI: $1.25 billion per month for Colossus data center capacity, through May 2029
- AWS: 10-year deal worth over $100 billion using Trainium chips
- Google Cloud: $200 billion compute commitment with a Broadcom partnership
- Microsoft: Now in talks for Maia 200 chips
Combined, that’s compute deals in the hundreds of billions. No other AI company — not even OpenAI — has this level of infrastructure diversification.
The Context: Demand Outpaces Supply
Dario Amodei said it himself: 80x growth in Q1 explains the compute difficulties. Claude is being adopted faster than Anthropic can stand up servers. In that context, bringing every available chip supplier on board makes perfect sense.
My Take
Anthropic’s multi-cloud strategy is smart. They’re avoiding dependency on any single provider and can choose the best hardware for each workload. It also gives them enormous negotiating leverage — when you’re talking to everyone, nobody can dictate terms. The only question is whether demand for Claude keeps growing at this pace, because the bills are monumental.
Sources: