Microsoft Diversifying 365 Copilot Away From OpenAI

Matthew Broersma,
Linkedin
A Microsoft data centre. Datacentre

Microsoft reportedly adding internal and third-party AI models to enterprise 365 Copilot offering as it seeks to diversify, cut costs

Getting your Trinity Audio player ready...

Microsoft has been working to bring internal and third-party models into its enterprise AI offering, 365 Copilot, in order to diversify away from its current OpenAI technology and cut costs, Reuters reported.

The company has already made similar moves with its other AI offerings, including GitHub Copilot, an product aimed at developers, and the consumer-oriented version of Copilot.

But the strategy is a significant departure for Microsoft 365 Copilot, which previously touted its early access to OpenAI models.

As well as diversifying away from dependence on a single supplier, Microsoft is concerned about cost and speed for enterprise users, the Reuters report said, citing unnamed sources.

An icon for Microsoft's 365 Copilot enterprise AI offering. Image credit: Microsoft

Internal models

In addition to training its own smaller models such as the latest Phi-4, Microsoft is customising other open-weight models to make 365 Copilot faster and more efficient, the report said.

The company is reportedly trying to make 365 Copilot less expensive to operate, potentially allowing it to pass on cost savings to customers.

Microsoft’s top leaders, including chief executive Satya Nadella, are tracking the effort closely, according to the report.

Microsoft said in a statement that OpenAI continues to be the company’s partner on frontier, or cutting-edge models.

“We incorporate various models from OpenAI and Microsoft depending on the product and experience,” Microsoft said.

Costs

Microsoft is OpenAI’s biggest single backer, with a funding deal worth up to $13 billion (£10bn) announced in January of last year.

Most recently OpenAI raised $6.6bn in a funding round valuing the start-up at some $150bn.

The company’s AI models, while the best-known in the industry, are also expensive to operate.

This is particularly the case with OpenAI’s current “reasoning” models, such as o3 and o3-mini, which it began testing last week.

Enterprise AI adoption

Such models run through a self-critique operation before delivering a response to a user, making them more reliable, especially for complex reasoning, science, mathematics and programming tasks.

GitHub Copilot has added models from Anthropic and Google alongside OpenAI, while the consumer Copilot offering uses in-house and OpenAI models.

Microsoft is currently working on pricing and utility concerns with 365 Copilot.

A Gartner survey of 152 information technology companies published in August found the vast majority of them had not progressed their 365 Copilot initiatives beyond the pilot stage.

Microsoft said in November that 70 percent of Fortune 500 companies are using 365 Copilot.