Microsoft shares rise as reports say the company is rethinking Copilot’s third-party AI model usage

Microsoft edged higher in Tuesday trading, gaining about 1.75% as the market digested new reporting about how the company is building its AI offerings. The move came amid a broader push by large software companies to control the cost of running AI workloads, especially for products that require frequent, high-volume model calls.

The catalyst was a market-news report that framed Microsoft’s AI strategy as increasingly centered on in-house model development. The piece said Microsoft is starting to “ditch” pricey third-party AI models in favor of its own, with the stated intent of putting pressure on Copilot-related costs.

Copilot, Microsoft’s umbrella brand for AI-assisted features inside products like Microsoft 365 and developer tools, is often positioned as one of the company’s key routes to monetizing generative AI. But generative AI can be expensive to serve, because each query may require additional compute, licensing, or inference capacity depending on the model provider and deployment approach.

By tying the reported model shift directly to Copilot economics, the update suggested that margins and pricing power could become easier to defend if Microsoft reduces external model fees and deployment costs. The report did not, however, provide specific numbers on cost savings, model names, or timelines, leaving most operational questions unanswered.

Microsoft has continued to market its AI capabilities across cloud services, developer tooling, and enterprise applications, but the Tuesday report focused less on product headlines and more on supply-chain choices for the underlying AI. In this context, the stock reaction likely reflected investor sensitivity to cost per AI interaction rather than only demand growth.

What remains unclear is the scope of any change. The reporting, as presented in the article circulating via Yahoo Finance, did not detail which Copilot experiences or regions are involved, whether the shift is partial or complete, or how performance and safety evaluations are being handled when swapping model sources.