Anthropic Negotiates with Microsoft on Maia 200 Chips for AI Inference

Anthropic is negotiating with Microsoft about using Maia 200 chips for inference of its AI models. This is a telling signal about growing computational power shortages and the willingness of leading AI startups to leverage any available alternative resources.

What is Maia 200

Microsoft introduced Maia 200 chips in January 2026 as a specialized processor for inference—for running already-trained language models in production. Unlike GPUs for training, inference processors are optimized for speed and energy efficiency of computations, allowing companies to significantly reduce the operational costs of large AI services. Maia 200 is positioned as a more accessible and energy-efficient alternative to expensive GPUs like NVIDIA's H100.

This makes it particularly attractive for companies that critically need to scale services without catastrophic increases in infrastructure costs. Notably, Microsoft itself has not yet deployed Maia 200 in its own Azure cloud platform, despite nearly a year having passed since the official presentation. This creates an interesting situation: external buyers like Anthropic could potentially gain access to these chips and test them in real production conditions before Microsoft itself begins full-scale deployment in its own data centers.

Why Computational Power Is in Acute Shortage

Demand for computational resources for AI is growing exponentially. Each new version of large language models requires ever-greater power for both training and inference. Startups like Anthropic, OpenAI, Mistral, and others compete on technology quality and in fierce battles for access to computing hardware. NVIDIA de facto dominates the AI GPU market, but manufacturing capacity lags behind demand. Wait times for significant GPU purchases can be measured in months. Microsoft is attempting to diversify its supply sources through investments in NVIDIA and the development of its own processors. The negotiations between Anthropic and Microsoft are logical: both sides are interested in practical experimentation with alternative solutions.

• Explosive demand for inference for ChatGPT-like commercial services
• Limited supply of high-performance GPUs from NVIDIA
• Long development cycles for competitors' custom chips
• Need to combine different types of hardware for flexibility

What Both Sides Stand to Gain

For Anthropic, access to Maia 200 means potentially lower operational costs and strategic independence from NVIDIA in inference computing. For Microsoft, it's an opportunity to prove to the market that its inference processor is genuinely production-ready and attractive to serious AI industry players. If Anthropic starts using Maia 200 in real production workloads, this will attract other potential customers and validate the approach. However, success depends on three critical factors: performance-to-cost ratio, competitive pricing, and supply reliability. Microsoft must guarantee stable output, or Anthropic will quickly revert to more proven NVIDIA GPUs.

What This Means

Diversification of computational power sources is becoming a strategic priority for AI companies. Successful deployment of Maia 200 at Anthropic could accelerate the development of alternative chips, reducing monopolistic pressure and opening new pathways for infrastructure optimization.