Microsoft has been a significant supporter of OpenAI but now they are ready to challenge them. In a “strategic” twist, Microsoft is now going solo by developing proprietary AI models, even as it maintains its partnership with the ChatGPT-maker.
Microsoft Developing Its AI Models
The Information was one to break the news. Microsoft is developing in-house AI models that they may sell to developers and use for their own AI needs.
The AI division inside the tech giant is led by Mustafa Suleyman. He and his team have already trained a bunch of models. They are currently being called MAI (Microsoft Artificial Intelligence). MAI models are much larger than their earlier Phi models and are focused on the reasoning capabilities, just like OpenAI’s o1.
Note that they are being developed with Copilot in mind. Copilot is a popular AI assistant that helps users write documents, manage spreadsheets, or even help in coding, and relies heavily on OpenAI models. So, there’s still a chance that these new models will be more general-purpose in the beginning. But we still don’t have much detail about them.
The development process for the MAI models was nearly a year long, with many roadblocks including technical difficulties and changes in the team members. Still, they managed to create something that met industry standards. The report states that MAI had a great benchmark score, quite similar to OpenAI and Anthropic.
The goal is to replace OpenAI’s models from Copilot with these new MAI models soon. They will also release them as an API for developers by the end of 2025.
Why are they Moving Away from OpenAI?
Microsoft has been a major backer of OpenAI since 2019, where both have benefitted from this partnership from the early days. In this time, OpenAI grew from a small AI startup into the most popular tech on the internet with ChatGPT. This innovation caught the world’s attention, showcasing the potential of AI in everyday applications. Till now, they had invested more than $14 Billion into this venture.
The move is to reduce Microsoft’s dependence on OpenAI. This will reduce the substantial costs associated with licensing OpenAI’s technology as well as diversify underlying technologies for their products.
The company is also evaluating xAI, Meta, and DeepSeek models in Copilot as an alternative to OpenAI.
There were rumors of an internal clash between the two partners, and this decision might be a concern. Even in 2024, they added OpenAI in their rivals list. However, Microsoft said this is not anything unusual.
“As we’ve said previously, we are using a mix of models, which includes continuing our deep partnership with OpenAI, along with models from Microsoft AI and open source models,” a Microsoft spokesperson said.
By investing in in-house AI development, they seek to maintain a competitive edge in the rapidly evolving tech landscape.
The other reason can be regulation. The UK’s Competition and Markets Authority (CMA) recently evaluated Microsoft’s substantial investment in OpenAI to determine if it warranted an antitrust investigation. After thorough assessment, the CMA concluded that Microsoft’s influence over OpenAI did not amount to de facto control, thereby deciding against a formal probe. But similar problems may reappear in the future.
Just a few weeks ago, Salesforce CEO Marc Benioff predicted Microsoft won’t use OpenAI’s technology across its tech stack because it’s too expensive and isn’t fast enough. He also said that Mustafa Suleyman (the man behind these new models) and Sam Altman (CEO of OpenAI) are not best friends. So, something is brewing!
Takeaways
Such developments highlight the fact that Microsoft also wants to advance their own AI technology.
Furthermore, Microsoft’s upcoming 50th-anniversary event is expected to showcase advancements in its Copilot AI companion, reflecting the company’s ongoing commitment to AI innovation.
