'Just remember, you have to be ahead for all time to come': Microsoft CEO Satya Nadella reveals AI chip shopping spree - says it will still buy Nvidia, AMD chips, despite recent launch of in-house Maia 200 hardware
Microsoft’s Maia 200 chip is designed for inference-heavy AI workloadsThe company will continue buying Nvidia and AMD chips despite launching its own hardwareSupply constraints and high demand make advanced compute a scarce resource
Microsoft has begun deploying its first internally designed AI chip, Maia 200, inside selected data centers, a step in its long-running effort to control more of its infrastructure stack.
Despite this move, Microsoft’s CEO has made it clear that the company does not intend to walk away from third-party chipmakers.
Satya Nadella recently declared Nvidia and AMD will remain part of Microsoft’s procurement strategy, even as Maia 200 enters production use.
Microsoft's AI chip is designed to support, not eliminate, third-party options
“We have a great partnership with Nvidia, with AMD. They are innovating. We are innovating,” Nadella said.
“I think a lot of folks just talk about who’s ahead. Just remember, you have to be ahead for all time to come. Because we can vertically integrate doesn’t mean we just only vertically integrate.”
Maia 200 is an inference-focused processor that Microsoft describes as built specifically for running large AI models efficiently rather than training them from scratch.
The chip is intended to handle sustained workloads that depend heavily on memory bandwidth, fast RAM access, and rapid data movement between compute units and SSD-backed storage systems.
Microsoft has shared performance comparisons that claim advantages over rival in-house chips from other cloud providers, although independent validation remains limited.
According to Microsoft leadership, its Superintelligence team will receive first access to Maia 200 hardware.
This group, led by Mustafa Suleyman, develops Microsoft’s most advanced internal models.
While Maia 200 will also support OpenAI workloads running on Azure, internal demand for compute remains intense.
Suleyman has said publicly that even within Microsoft, access to the latest hardware is treated as a scarce resource. This scarcity explains why Microsoft continues to rely on external suppliers.
Training and running large-scale models require enormous compute density, persistent memory throughput, and reliable scaling across data centers.
No single chip design currently satisfies all these requirements under real-world conditions, and so as a result, Microsoft continues to diversify its hardware sources rather than betting entirely on a single architecture.
Supply limitations from Nvidia, rising costs, and long lead times have pushed companies toward internal chip development.
These efforts have not eliminated dependence on external vendors. Instead, they add another layer to an already complex hardware ecosystem.
AI tools running at scale expose weaknesses quickly, whether in memory handling, thermal limits, or interconnect bottlenecks.
Owning part of the hardware roadmap gives Microsoft more flexibility, but it does not remove the structural constraints affecting the entire industry.
In simple terms, the custom chip was designed to reduce pressure rather than redefine it, especially as demand for compute continues to grow faster than supply.
