Microsoft’s Scott Guthrie says “competition is good” as Maia 300 looks to reduce reliance on Nvidia chips

Microsoft is betting that custom silicon can help it scale AI more efficiently, and its senior cloud and AI executive is framing the broader chip competition as healthy for the market. Speaking to Yahoo Finance in an interview, Scott Guthrie, Microsoft’s Executive Vice President of the Cloud and AI group, addressed whether Maia 300 will cut Microsoft’s reliance on Nvidia. His response was blunt: “Competition is good.”

Maia 300 is Microsoft’s newer generation of purpose-built AI chips designed for workloads connected to the company’s cloud AI stack. In the Yahoo Finance conversation, Guthrie tied the chip’s potential value to performance and efficiency, saying Maia 300 can deliver about 40% more tokens per megawatt and per rack. In plain terms, tokens are the basic text units that large language models generate, and megawatt and rack are measures of power and hardware footprint, so the metric is intended to capture how much useful output the data center can produce for a given energy and equipment constraint.

That argument matters in the AI industry because both power capacity and data center infrastructure have become limiting factors as model sizes and usage grow. While many deployments still rely on Nvidia GPUs, Microsoft has spent time building an ecosystem around its own chips to support training and inference needs. Guthrie’s comments suggest Microsoft sees custom silicon as a lever for scaling AI deployments without being boxed in by a single supplier’s product cadence or supply constraints.

Guthrie also positioned the chip conversation as more than a company-level procurement issue. By emphasizing that competition is good, he implied that multiple hardware options can push performance improvements across the sector, rather than locking customers into one supplier. For Microsoft, that stance aligns with its broader approach to reducing dependency risk by diversifying the underlying compute used for AI services.

Microsoft did not provide additional specifics in the posted interview beyond the efficiency claim and the question of Nvidia reliance, according to the published description of the conversation. The post did not spell out, at least in the available account, how quickly Maia 300 will be deployed in production, what portion of Microsoft’s overall AI compute Maia 300 could cover, or whether the company expects to migrate specific customer workloads in phases tied to performance or cost benchmarks.

For investors and customers watching the AI infrastructure arms race, the immediate takeaway is that Microsoft is publicly tying its chip roadmap to both output efficiency and supply diversification. The next indicates to watch are any Microsoft statements that quantify deployment timelines, data center ramp plans, and workload coverage, including whether Maia 300 is intended to complement or substitute for Nvidia-based systems in particular services. The company’s messaging in this interview suggests it will keep using internal chips as a way to control efficiency, while still benefiting from an industry-wide competitive environment.