Microsoft wants its own frontier AI models on par with OpenAI and Anthropic by 2027

Microsoft wants to build its own large, frontier-class AI models as early as 2027. In effect, the company is trying to develop an in-house alternative to the most powerful systems from OpenAI and Anthropic — not as a supporting tool, but as a full technological foundation for the next phase of the AI race. That sends an important signal to the entire market.

Until now, Microsoft has been seen primarily as one of the leading infrastructure and product players in AI: it aggressively builds artificial intelligence into its services and grows an ecosystem around it. Now the focus is also shifting toward creating its own foundation models that can compete at the top end of the market. This isn't about small, task-specific tools — it's about large models that set the bar for quality, breadth of tasks, and degree of autonomy.

According to the report, Microsoft's goal is to have an in-house alternative to the strongest models from OpenAI and Anthropic by next year, meaning 2027. For the company, this isn't simply a matter of prestige. Owning a model of this class gives it control over a key layer of the AI stack: architecture, development priorities, update cadence, and how the model is integrated into its own products.

The less it depends on an outside foundation model provider, the more flexibly it can shape strategy and the faster it can make decisions. Another important factor is the fight for control over the economics of AI. Large models require enormous compute resources, expensive infrastructure, and long research cycles.

If Microsoft manages to bring its models up to frontier level, it will be able not only to use them in its own services but also to offer customers a more tightly integrated bundle of cloud, developer tools, and the model itself. That matters especially in the enterprise market: businesses need not just strong AI but also a clear vendor that takes responsibility for several layers of the technology platform at once. The timeline is telling, too.

Targeting 2027 implies a demanding pace: at this level, it isn't enough to assemble a team and ship a demo. It takes mature research methods, reliable access to compute, a well-tuned training pipeline, and the ability to improve the model quickly after the first launches. In other words, Microsoft doesn't just want to catch up with a trend — it wants to join the companies that define what the next wave of generative AI will look like.

It also matters that an in-house alternative doesn't necessarily mean immediately ending partnerships with outside labs. It's more about rebalancing power. A company with its own strong model gains more leverage in negotiations, more room to experiment, and fewer strategic risks.

That's especially valuable at a time when the market is moving very fast and access to the best models is becoming critical for both consumer services and enterprise software. Against the backdrop of generative AI's rapid growth, the move looks logical. OpenAI and Anthropic are currently seen as the benchmark for the most powerful models, and Microsoft's attempt to build a comparable system in-house shows that partnerships and integrations are no longer enough.

Market leaders want to own not just the distribution channel, the interface, or the cloud, but the core of the technology itself. That makes the business more resilient: if outside partners change their priorities, pricing, or access terms, the company still has its own strategic asset. For users and companies building products on top of AI, this could mean more choice and fiercer competition among foundation model providers.

And for Microsoft itself, success with this project by 2027 would show that the era of depending on other companies' frontier models is ending: the largest tech platforms want to build their own intelligence in-house.