December 23 – Microsoft (MSFT.O) is actively exploring ways to integrate both in-house and third-party AI models into its flagship product, Microsoft 365 Copilot, aiming to diversify beyond OpenAI’s technology and lower operational expenses, according to sources familiar with the matter. This move signals a shift from Microsoft’s heavy reliance on OpenAI’s GPT-4 model, which has been a cornerstone of the tech giant’s AI efforts.
Moving Beyond OpenAI
Microsoft’s desire to lessen its dependence on OpenAI—despite being a major backer of the startup—marks a notable change. When 365 Copilot was first introduced in March 2023, a key selling point was its use of GPT-4, one of OpenAI’s most advanced models. However, sources say Microsoft is now keen to adopt alternative options, citing concerns over cost and performance for its enterprise customers.
“OpenAI continues as our partner on cutting-edge models,” a Microsoft spokesperson said, referring to frontier AI technologies. Under Microsoft’s original agreement with OpenAI, the Redmond-based company can customize OpenAI’s models to meet various product requirements.
Cost Reduction and Efficiency
Microsoft has already begun training its own smaller models, including a system called Phi-4, while also customizing other open-weight models. The strategy behind using a broader range of models is to optimize 365 Copilot for faster response times and more efficient resource usage—improvements that could ultimately translate into cost savings for end users.
One insider noted that Satya Nadella, Microsoft’s Chief Executive Officer, is closely following these developments. The goal is to make running 365 Copilot more affordable for Microsoft and pass potential savings on to customers.
Expanding AI Model Choices
This push for diversification mirrors similar efforts within other Microsoft units. GitHub, acquired by Microsoft in 2018, introduced new AI models from Anthropic and Google in October as alternatives to OpenAI’s GPT-4 for its GitHub Copilot. Meanwhile, Microsoft’s consumer-facing chatbot, Bing Chat Copilot, also relies on a mix of proprietary and OpenAI models.
Adoption and Skepticism
Despite these new initiatives, Microsoft 365 Copilot is still in its early stages of proving real return on investment for enterprise users.
While Microsoft has yet to disclose specific sales figures or widespread adoption rates, industry experts have expressed concerns about its pricing and overall utility. According to August data from research firm Gartner, most of the 152 IT companies surveyed had not progressed beyond pilot tests of 365 Copilot.
Still, analysts from BNP Paribas Exane report seeing growing interest, predicting that more than 10 million paid users will adopt 365 Copilot this year. Microsoft also highlighted in a November blog post that 70% of Fortune 500 companies are already testing or using 365 Copilot, suggesting momentum is building for the AI-powered assistant.
As Microsoft continues to refine 365 Copilot and incorporate an ever-growing suite of AI models, the balance between cost, performance, and innovation will remain a focal point. By leveraging internal technologies like Phi-4 and third-party solutions beyond OpenAI, Microsoft aims to make its flagship AI product more versatile, cost-effective, and appealing to a broader range of enterprise customers—ultimately shaping the next chapter in AI-driven office productivity.