Anthropic is reportedly planning to purchase Microsoft's self-developed AI chip "Maia 200"! Breaking Nvidia's dominance in computing power, MSFT's stock price rises 2%.

Microsoft Challenges Nvidia’s Dominance in Computing Power! According to The Information today (21st), AI giant Anthropic is in early negotiations with Microsoft to lease its self-developed Maia 200 chips to run its Claude model. This move caused Microsoft’s stock price to rise by 2% on Thursday. If this deal is finalized, it will mark a key turning point for Microsoft from “buying chips” to “selling computing power,” significantly reducing cloud costs and providing customers with a high-cost-performance alternative outside of Nvidia.
(Background: Trump Q1 Stock Purchases Expose $750 Million Sweeping Nvidia and Apple, Selling Microsoft, Amazon, Meta)
(Additional context: OpenAI and Microsoft agree on a $38 billion profit-sharing cap, saving hundreds of billions and gaining IPO freedom)

Table of Contents

Toggle

• Focusing on “Inference Optimization,” Maia 200 Becomes an Alternative to Nvidia
• Anthropic’s Computing Power Anxiety: Not Putting All Eggs in One Basket
• Microsoft’s Ambition: From Buyer to Top “Arms Dealer”

Microsoft (NASDAQ: MSFT) is about to make a strategic breakthrough in AI hardware infrastructure.

According to the latest report from tech media The Information today (May 21, 2026), the startup giant Anthropic, known for developing the famous AI model Claude, is currently in early negotiations with Microsoft to lease computing resources powered by Microsoft’s self-developed AI server chips (mainly the Maia series). Inspired by this major news, Microsoft’s stock price surged by 2% during Thursday’s trading.

Focusing on “Inference Optimization,” Maia 200 Becomes an Alternative to Nvidia

The report states that the core focus of the negotiations is Microsoft’s second-generation AI accelerator released in January this year — Maia 200.

Built on TSMC’s 3-nanometer process, this chip differs from Nvidia’s all-purpose GPUs by focusing entirely on “Inference Optimization”. This means that while it’s not suitable for training large models from scratch, it can deliver faster processing speeds, lower energy consumption, and highly competitive costs when running existing Claude models.

Microsoft has already applied Maia chips in internal services like Copilot. If it successfully secures Anthropic as a major external client, it will be a significant milestone for Microsoft’s self-developed chips going outside the company.

Anthropic’s Computing Power Anxiety: Not Putting All Eggs in One Basket

In fact, Anthropic has long been a major user of Microsoft Azure cloud services. In a deal at the end of 2025, Anthropic committed to purchasing up to $30 billion in Azure computing power (still mainly based on Nvidia’s Grace Blackwell architecture).

However, with the explosive growth in demand for Claude models, Anthropic is facing severe computing power shortages. By introducing Microsoft’s Maia chips, Anthropic aims to achieve the following strategic goals:

• Diversify supply chain risks: Implement a “multi-cloud + multi-silicon” strategy to reduce dependence on the dominant chip maker Nvidia.
• Lower operational costs: Use Maia’s high cost-performance ratio to handle the massive inference tasks daily.
• Customize future hardware: Leasing existing chips opens the door for collaboration, with Anthropic potentially participating in feedback on Microsoft’s next-generation chip designs to tailor hardware to its specific needs.

Microsoft’s Ambition: From Buyer to Top “Arms Dealer”

This negotiation reflects the current AI industry’s “computing power equals power” competitive landscape. Nvidia’s high hardware prices are continuously eroding cloud providers’ profit margins.

For Microsoft, if it can successfully persuade Anthropic to adopt Maia 200, it will not only demonstrate the strength of its self-developed chips to the market but also turn the tide in the cloud wars against Google (TPU) and Amazon (Trainium). This marks Microsoft’s transformation from a customer simply “buying Nvidia chips” to a top-tier infrastructure arms dealer capable of exporting its own AI silicon chips externally.