Microsoft and OpenAI have something of a symbiotic relationship, with the former providing billions in capital to a startup AI lab and in return getting early access to the cutting-edge models Microsoft’s is currently cooking up. suite of productivity software. The two companies are headed in diverging directions, however, and Reuters It’s now being reported that Microsoft is looking to add more models to its 365 Copilot product that aren’t built on OpenAI.

The reasoning, according to the report, is that Microsoft sees OpenAI’s cutting-edge GPT-4 model as too expensive and not fast enough to satisfy its business customers. Copilot 365 is an AI-powered assistant built into Microsoft’s suite of productivity applications including Word and PowerPoint. The tool should absorb all the data of a company and do many things, such as giving users the ability to quickly find information without having to hunt through different applications; quickly create a list of the most profitable business units of the company; or quickly summarize meetings and emails.

It is supposedly to do those things, but customers and insiders alike are still weary of Copilot 365, which costs an extra $30 per month per user on a team. Recently Business Insider storyMicrosoft employees who spoke anonymously called the tools “awful” and “gimmicky,” not working well 75% of the time. In front of the customer, Business Insider Cited a survey of 123 IT leaders published by management consultancy Gartner, which found that only four said Copilot provided significant value to their companies. It should be noted that other stories reported by companies that have found value in using large language models, such as through simplifying customer support.

Some customers are talking Business Insider specifically noted that 365 Copilot is very expensive.

OpenAI’s ChatGPT is a frontier, general model, meaning it is trained on wide swaths of data and can be more expensive and slower to run; so most models are offered in “lite” versions that do less intensive inference or “thinking.” Microsoft is training its own in-house, smaller models like the so-called Phi-4, and Reuters reported that sources who spoke to the outlet said that the company is looking to “customize other open weight models to make the 365 Copilot faster and more efficient.”

On the one hand, it makes sense that Microsoft would want to reduce its reliance on OpenAI. If the company is right and AI is the next revolution in computing generation, relying on an independent company for core technology is not a good idea.

Microsoft is plowing billions of dollars into OpenAI and will receive 75% of its profits until it breaks even on its investment, and yet still holds a large stake in the startup. The company will in effect get to hedge its bets—building its own in-house models while saving an OpenAI lottery ticket if it continues on its current path to heaven.

Despite being the front-runner today, some skeptics of OpenAI say that we may not yet know a real winner in the AI ​​race (should these technologies be as revolutionary as they say we believe). In the same way that there were many search engines that came online in the ’90s, they were quickly defeated when the late Google appeared. Microsoft would probably be wise to take the plunge.