Microsoft will reportedly unveil a suite of homegrown AI models next week at its Build developer conference in San Francisco, with a coding-focused model designed to revive GitHub Copilot at the centre of the announcement. The Information broke the story, Reuters confirmed the lineup, and the share price jumped roughly 3 percent on the news. The headline reads like a routine product preview. The subtext is rather sharper: the most lucrative software partnership of the AI era is being quietly converted into a competition.
Microsoft has spent the last three years powering Copilot, Microsoft 365 and Azure services with models from OpenAI, Anthropic and Google. According to the same reporting, the Build lineup will also include in-house systems for transcription, reasoning, speech and image generation. That is the full vertical stack Microsoft has historically rented from partners, now mostly built in-house. The shift was made possible by an April renegotiation of the OpenAI deal, which lifted restrictions that had prevented Microsoft's AI organisation, run by AI chief Mustafa Suleyman, from training top-tier models of its own.
The real motivator, though, is Anthropic. GitHub Copilot was the original star of AI-assisted coding, but Claude Code has overtaken it inside the developer community. Microsoft itself has been a customer: thousands of employees have used Claude Code internally for months, and multiple reports now say the company plans to phase out that internal use by the end of June and push staff onto a Copilot-based command-line tool. Microsoft is, in effect, banning its own engineers from the rival product so that its own product has a chance to catch up.
The numbers around the move are easy to miss because they are dwarfed by Microsoft's overall scale. AI revenue has reached an annual run rate of $37 billion, up 123 percent. Azure grew 40 percent in the most recent quarter. Capital expenditure is on track to exceed $40 billion a quarter, with $190 billion planned for calendar 2026. Investor sentiment, despite all this, has soured: the stock is down about 16 percent year to date. The market is pricing in the cost of training and running models that the company used to license from OpenAI. Building those models in-house lets Microsoft control the input cost over time, which matters precisely because token bills have become the variable that public investors can no longer ignore.
Two questions hang over the announcement. The first is whether a coding model assembled in months can compete with one Anthropic has spent two years sharpening on real developer feedback. Claude Opus 4.8 scored 69.2 percent on SWE-Bench Pro this week, with OpenAI's GPT-5.5 at 58.6 percent. Microsoft has not previewed numbers, only intent. The second question is what happens to its OpenAI relationship. The two companies signed a non-binding deal in September allowing OpenAI to pursue a for-profit conversion and eventual IPO. Microsoft retains hosting rights, but the exclusivity that defined the partnership is gone, and Reuters has reported that Microsoft is now studying acquisitions to bring AI researchers in-house ahead of a target to ship a frontier model next year.
Build will be the moment Microsoft has to put performance behind the messaging. A new coding model is the most visible test, because GitHub Copilot is the AI product Microsoft sells most directly to the people who decide which tool to use. If the in-house model lands close to Claude Code on real developer work, the story of the next twelve months becomes one of platform owners taking their own roofs back. If it does not, Microsoft will have publicly committed to a tool many of its own engineers have already rejected.