Anthropic Is in Talks to Use Microsoft's Custom Maia 200 AI Chips

By Hector Herrera | May 22, 2026

Anthropic is in preliminary talks to adopt Microsoft's custom Maia 200 AI chips, according to CNBC, a move that would add a third major chip supplier to the AI lab's already diversified compute stack. No deal has been signed, and the talks are early — but if completed, it would mark a significant expansion of Microsoft's Maia program beyond its own Azure infrastructure.

Why This Matters

Anthropic is one of the most compute-hungry companies in AI. It currently pays SpaceX roughly $1.25 billion per month for GPU access, according to prior reporting — a figure that underscores just how aggressively the company is scaling its model training and inference. Adding Microsoft's Maia 200 chips to its supply chain would further reduce Anthropic's dependence on Nvidia's H100 and H200 GPUs, which remain in tight supply and command premium pricing.

What We Know

• The chip: Microsoft's Maia 200 is a custom AI accelerator designed in-house for Azure workloads. Microsoft built Maia specifically to run large language models (LLMs — the type of AI that powers Anthropic's Claude) at lower cost than third-party GPUs.
• The talks: Preliminary, per CNBC. No timeline, no confirmed deal terms.
• Anthropic's current compute suppliers: AWS via Trainium chips, Google via TPUs, and SpaceX for GPU compute. The company has no single compute dependency — by design.
• The cost pressure: At $1.25 billion per month in GPU spend, Anthropic has enormous incentive to find cheaper or more available alternatives. Custom silicon from hyperscalers typically comes with lower per-unit cost in exchange for architectural lock-in.

The Bigger Picture: AI Labs Are Diversifying Away from Nvidia

Anthropic isn't alone. Meta, Google, Amazon, and Microsoft have all built custom AI chips in the past three years. The pattern is consistent: train initially on Nvidia GPUs, then migrate routine inference workloads to cheaper custom silicon as scale grows.

What makes Anthropic's situation different is the breadth of its existing supplier relationships. Most AI companies have one or two primary compute sources. Anthropic is actively building a portfolio of four or more — a strategy that trades architectural optimization for supply resilience and negotiating leverage.

If Microsoft's Maia 200 enters that mix, it would also deepen a commercial relationship that already includes Anthropic's models being available through Azure AI. A chip deal on top of a distribution deal would tie the two companies closer together at the infrastructure layer.

What to Watch

Whether this becomes a signed agreement — and whether Anthropic discloses which workloads, if any, run on Maia versus its other compute options. Custom chip adoption often starts with narrow inference tasks before expanding to training runs; watch for any Azure-specific Claude deployment announcements as a signal that a deal closed.

Sources: CNBC