Microsoft Shifts AI Workloads To In House Models To Cut Costs
Microsoft is moving to cut the cost of powering its AI products by leaning more heavily on its own in-house models, as the company continues to expand features across Copilot and other services.
The shift was reported as part of a broader effort inside Microsoft to reduce spending tied to third-party AI systems. In recent months, Microsoft has been building and deploying its own models while also working with external providers, including OpenAI and Anthropic, according to the reporting.
The cost-cutting push is tied to Microsoft’s strategy of increasing self-reliance in the AI stack that underpins its consumer and enterprise offerings. The company has made AI a central part of its product roadmap, with Copilot positioned across Microsoft 365, Windows, and other services. As usage scales, the expense of running large models becomes a major operational factor.
This development matters because Microsoft’s approach influences how AI is delivered to millions of users and how quickly new AI capabilities can be rolled out. A heavier reliance on Microsoft-built models could change the economics of features such as summarization, chat, and content generation by giving the company more control over pricing, performance tuning, and deployment choices.
It also signals a competitive dynamic among major tech companies as they work to secure the key inputs needed for AI at scale. Alongside model development, the industry has been investing in infrastructure and specialized hardware. Separately, sources told The Business Standard that China’s DeepSeek is developing its own AI chip, highlighting how companies are seeking more control over both software and the compute resources required to run it.
For Microsoft, using more in-house models could affect how it allocates workloads across providers and how it negotiates future partnerships. The reports frame the effort as part of internal “streamlining” tied to Copilot, with an emphasis on reducing reliance on outside models where Microsoft’s own technology can be substituted.
What happens next will be reflected in how Microsoft updates its AI product architecture and how it describes model usage across its services. Any changes would likely emerge through product releases, platform documentation, or developer-facing updates, as well as through the company’s ongoing collaborations and procurement decisions.
As the cost of running AI becomes a defining pressure for the industry, Microsoft’s move underscores that control over models—and the bills that come with them—is now a core part of the AI race.
