Microsoft is advancing the development of its own and third-party AI models for 365 Copilot to reduce its reliance on OpenAI and reduce operating costs. This emerges from information provided by anonymous sourcesReuterspresent.
Background: OpenAI and Microsoft's cooperation
Since the launch of 365 Copilot in March 2023, OpenAI's GPT-4 has been a core part of Microsoft's AI catalog. The US company, a major investor in OpenAI, has so far benefited from close collaboration. But according to the sources, Microsoft plans to diversify to address costs and performance concerns, particularly for enterprise customers.
A Microsoft spokesperson confirms this:
»We integrate different models from OpenAI and Microsoft, depending on the product and experience.«
OpenAI itself did not want to comment on the reports.
New approaches for 365 Copilot
In addition to further developing internal models, including Phi-4, Microsoft is working on adapting open models. The goal is to increase 365 Copilot's efficiency and potentially pass on cost savings to customers. An insider has stated that senior Microsoft executives such as CEO Satya Nadella are closely following the progress.
This strategy is consistent with similar adjustments in other Microsoft areas. Since October, GitHub has been using models from Anthropic and Google as alternatives to GPT-4.
365 Copilot under pressure
Despite these aspirations, 365 Copilot is under pressure to prove its value to companies. Sales figures for licenses have not yet been disclosed, and aGartner surveyhas revealed that many companies have not made progress with their 365 Copilot pilots. Still, analysts at BNP Paribas Exane report accelerating adoption, predicting more than 10 million paying users by 2024 (viaReuters).
SeaMicrosoft's blog entryBy November 2024, 70 percent of Fortune 500 companies are already using 365 Copilot. However, it remains to be seen whether the current investments will pay off.
Do you use Microsoft 365 at work? If so, what experiences have you had with AI functions? Can you understand the criticism that OpenAI's models are too slow and too expensive? Feel free to write it in the comments below!
