Microsoft Is Quietly Building an OpenAI Competitor

For a company that has invested over $14 billion in OpenAI, and has enjoyed front-row access to its groundbreaking models, embedding them into its suite of apps, you’d think Microsoft would be perfectly content relying on it to continue to power its AI ambitions.

But behind the scenes, it’s been working on something that suggests otherwise.

According to multiple reports, Microsoft has developed a new series of large language models called MAI—short for Microsoft Artificial Intelligence, which have been tested against OpenAI and Anthropic’s models and proven competitive.

The company has even tested whether MAI could replace OpenAI’s models in Copilot, its AI-powered assistant, suggesting it’s serious about moving beyond OpenAI’s models.

How Microsoft’s $10 billion investment into OpenAI is a big bet on the future of AI
The Draft by Techloy brings you data-driven news and insights into technology across the world’s largest emerging markets from trusted sources. If you’re not already a paid subscriber, join today to gain full access to data-led startup news and VC deals, incisive charts, and deeper insights. Our lead story for

The tension between Microsoft and OpenAI has been building for months. According to The Information, OpenAI refused to share technical details about its o1 model, which reportedly frustrated Microsoft engineers. Meanwhile, the two companies have also recently renegotiated their partnership, allowing OpenAI to move some workloads away from Microsoft’s cloud, suggesting that Microsoft is preparing for a future where it doesn’t have to rely on OpenAI at all.

If Microsoft is able to integrate MAI into Copilot, it could move beyond OpenAI’s technology altogether.

But Microsoft isn’t betting on just one alternative. It has also reportedly tested models from Anthropic, Meta, DeepSeek, and xAI, hinting at a strategy where Copilot could become model-agnostic, rather than being tied exclusively to OpenAI’s technology.

It’s not Microsoft’s first AI project, either. The company has already developed Phi, a series of lightweight yet powerful AI models. The latest versions, Phi-4-mini and Phi-4-multimodal, launched in February, proving that Microsoft can train efficient AI models without relying on OpenAI’s vast infrastructure. If the company applies similar techniques to MAI, it could produce powerful AI models at a fraction of the cost.

For now, Microsoft insists that it remains committed to its partnership with OpenAI. But actions speak louder than words. If MAI continues to improve, and Microsoft finds success with its own models, it may not be long before OpenAI realizes that its biggest financial backer is also becoming one of its biggest competitors.

Microsoft launches Phi-3 Mini, its smallest AI model with big potential
Phi-3 Mini joins the growing market of lightweight AI models.