Microsoft is weighing whether to release some of its internally built AI models with downloadable weights, a potentially huge shift as Chinese systems win over U.S. users with lower prices, fast performance and fewer locked doors. The company has not announced a release or named which MAI models could be involved, so for now this is a strategic option—not a product launch.
According to Nikkei Asia, Microsoft is considering offering some internally developed models with open weights. The report identifies Mustafa Suleyman, Microsoft’s AI chief, as the executive leading the MAI family, and says the idea is being weighed as Chinese AI models gain popularity among U.S. users.
Microsoft Is Talking Like An Open-Weight Company
Open-weight models are AI systems whose trained parameters can be downloaded, inspected, modified and run on a user’s own infrastructure. In its open-weight policy paper, Microsoft argues that the technology can expand competition, reduce vendor lock-in and make advanced AI more affordable; the page listed more than 230 signatories as of Thursday.
That stance is striking because Microsoft has also spent years building a business around cloud-hosted AI access through Azure and Copilot. In practical terms, opening some MAI weights could give developers more control, lower inference bills and reduce dependence on any single model provider—an especially useful hedge for a company whose AI stack has long involved outside partners.
MAI Is Already Being Used As A Cost-Cutting Hedge
Microsoft is not starting from scratch. In last Thursday's update, Microsoft AI said MAI models were powering workloads in GitHub Copilot and Excel, with its Excel model performing on par with GPT-5.6 for common tasks while costing less to run.
That helps explain why open weights are suddenly on the table. A model that can be customized and run on a company’s own infrastructure can be cheaper and easier to swap than a closed system billed token by token, while Microsoft still gets to sell the cloud, tools and enterprise controls around it.
Chinese Models Are Turning Price Into A Weapon
The pressure is no longer theoretical. The Associated Press reported that San Francisco-based Mozilla CTO Raffi Krikorian switched to Moonshot’s Kimi K3 for many day-to-day tasks after finding it “snappier” than Anthropic’s more expensive Fable, while Coinbase and others have also tested Chinese models to cut costs.
Data cited by The Washington Post put Chinese models at nearly half of U.S. traffic by tokens on OpenRouter in the last week of June, up from 16% at the start of the year. Microsoft Azure is already among the cloud platforms offering access to Chinese open-weight models, which makes the company’s position particularly awkward: it can profit from the open ecosystem even as it tries to build a U.S.-made alternative.
The Catch Is Security And Control
Microsoft’s own open-weight argument concedes the downside: once weights are released, they can move beyond the original developer’s control, and modified versions can be difficult to trace or reverse. The company says policymakers should distinguish legitimate model distillation from unlawful extraction, rather than impose broad restrictions.
That debate is colliding with accusations around Chinese models. AP reported that the Trump administration accused Moonshot of using covert methods to build Kimi K3 from Anthropic’s Fable, while Beijing rejected the allegation as groundless. The dispute could make any future Microsoft release a test case for whether Washington wants open weights treated as a competitive advantage, a security risk or both.
For now, the headline is less Microsoft open-sources Copilot than Microsoft may need to open the toolbox. Nikkei’s report does not identify a timetable or specific MAI model for release, and Microsoft has not publicly confirmed that a launch is imminent—but the company’s lobbying, cost-cutting and model-building efforts suggest the old closed-versus-open divide is getting harder to maintain.
