Microsoft outlines it wants to reduce Nvidia’s cut of its AI spend as Maia chip moves closer to Copilot

Microsoft is making a strategic bet that it can keep a larger portion of the dollars flowing through its AI stack, according to a market report published this week by TheStreet and syndicated by Yahoo Finance. The report argues that Microsoft’s own Maia chip could change which companies capture profits as demand grows for AI inference, the phase where models run to produce answers and outputs for customers.

The central claim is not that Microsoft is abandoning Nvidia outright, but that it wants more of the “AI billions” spent to run AI services to accrue to Microsoft instead of Nvidia. In the framing of the report, the Maia chip is positioned as a tool that could shift costs and bargaining leverage in Microsoft’s broader cloud and AI supply chain.

The report ties the implications directly to Copilot, Microsoft’s branded suite of AI features that is embedded across products and used by businesses and developers. The idea is that if Copilot-related workloads increasingly route through Microsoft-designed hardware for inference, Microsoft would be better positioned to capture margin internally, rather than paying as much to hardware partners for the compute needed to serve user requests.

Beyond the high-level thesis, the market report does not lay out specific customer contracts, percentage-of-workload targets, or timelines for any replacement of Nvidia hardware. It also does not provide quantified numbers such as projected unit volumes, spend shifts, or revenue impacts tied to Maia, leaving the magnitude of the economic change unclear from the published account.

Still, the strategic logic aligns with a broader pattern in the AI industry: as AI deployment moves from training to serving, companies seek control over the inference layer, where steady, recurring usage can translate into predictable demand. Inference is the portion of AI computing most closely tied to end-user experiences, and it can be supported by custom silicon that is optimized for the specific model families and runtime needs of a platform.

For Microsoft, the stakes are tied to both cost management and competitive positioning. Copilot is a flagship offering, and the company’s ability to deliver responses at scale is closely linked to the compute economics of the systems behind it. The report’s emphasis on “who profits” suggests Microsoft is focused on the hardware and performance trade-offs that determine whether AI services become more or less expensive to run over time.

What remains uncertain is how quickly Maia-based inference could spread across the environments that serve Copilot and other AI workloads, and whether Nvidia remains a major supplier for other layers of the stack. The published report also does not disclose any explicit procurement changes or detailed partner strategy, so readers are left with a directional view rather than a measurable plan.

Investors and customers will likely watch for future indicates around Maia adoption, including any disclosed mix of hardware in Microsoft’s cloud AI infrastructure, changes in cloud AI pricing, or additional engineering milestones that indicate which inference workloads are moving onto Microsoft-designed chips. If Microsoft provides more specificity, it could clarify how much leverage the Maia strategy actually creates in the competitive AI compute market.