Microsoft built Maia 300 to stop paying Nvidia for AI chips

Microsoft has announced the Maia 300, its latest in-house AI chip, which the company says will be the primary AI inference system across its cloud services. This marks the third generation of Microsoft’s Maia chip line and signals a direct move to reduce its reliance on Nvidia’s GPUs for AI workloads.

Why it matters

For years, Nvidia has dominated the AI chip market, making its hardware a critical resource for anyone running large-scale AI in the cloud. Microsoft’s Maia 300 is designed to handle the latest reasoning models and agentic AI tasks, according to The Verge. By rolling out its own chip for AI inference, Microsoft is looking to control costs and improve efficiency across its Azure cloud platform.

Microsoft’s move echoes a broader trend among major cloud providers: building proprietary hardware to gain more control over their supply chain and performance. This shift could help Microsoft avoid GPU shortages and pricing swings that have complicated cloud AI planning for many companies. For background on this trend, see our earlier article here.

Microsoft’s Maia 300: What’s new and what’s changing

The Maia 300 is Microsoft’s third-generation in-house chip, intended to run AI inference workloads at scale. Microsoft says this chip will be used for its own products and services, making it a core part of the company’s AI infrastructure. According to The Verge, Maia 300 improves the cost and efficiency of running AI models in the cloud and is better equipped to handle advanced reasoning and agentic use cases than its predecessors.

Microsoft has not published public benchmarks in the initial announcement, so for now customers will need to rely on the company’s own claims. However, the decision to position Maia 300 as the “primary AI inference system” is a clear signal that Microsoft is betting on its own silicon to power the next stage of its AI offerings. This could affect which AI models run fastest or most affordably on Azure, depending on how well Maia 300 performs in real-world scenarios.

The move also means Microsoft will have more flexibility in scaling up its AI services. With Nvidia GPUs in high demand and often in short supply, Maia 300 gives Microsoft another way to meet customer demand during peak periods. Over time, this could also put pricing pressure on the broader GPU market if other cloud providers follow suit.

What this means if you run a business

If your company uses Microsoft’s cloud AI services, you may see lower or more predictable pricing for AI inference workloads as Maia 300 is deployed. Microsoft’s ability to bypass some of the supply constraints tied to Nvidia hardware could mean faster onboarding and fewer delays when spinning up new AI-powered features or pilots.

For companies building or deploying custom AI models on Azure, it is worth tracking which services and regions are running on Maia 300 versus Nvidia GPUs. Some specific model types or workloads may perform differently, depending on how Maia 300 is optimized. Early adopters will want to monitor documentation and performance guidance from Microsoft as more details are released.

The bigger picture is that Microsoft’s investment in custom silicon could shift the economics of AI in the cloud. If Maia 300 delivers on its efficiency and cost claims, business users may find it easier to budget for large-scale AI projects or experiment with new agentic automation tools. For more on how Microsoft is integrating its own AI chips across its cloud, see our report here.

For now, the Maia 300 rollout is a reminder that cloud AI infrastructure is changing quickly, and that the hardware running your models could have a direct impact on both your costs and your ability to deliver new products on time.