Microsoft and OpenAI have deep financial ties together, with Microsoft investing over $13 billion in the artificial intelligence startup led by Sam Altman. Despite Nadella even offering to bring Altman aboard when he was temporarily ousted as CEO of OpenAI in November 2023, the partnership shows fragility in an increasingly competitive AI market.
Reuters first reported that Microsoft is considering developing new proprietary and adding third-party AI models that could be used in future Microsoft 365 Copilot products. This starkly contrasts with the existing approach, which exclusively uses OpenAI models.
It’s unclear how the implementation of multiple models or providers would be incorporated into the Copilot suite–or if it marks a departure from OpenAI support in its flagship software. However, the message is clear: reduce reliance on OpenAI.
The news arrives as Microsoft is under increasing pressure to meet market and user expectations. Slow performance has plagued user perception and experience of Copilot.
It is also increasingly expensive to scale and serve generative AI to its enterprise customers, which utilize costly graphical processing unit (GPU) hardware from providers such as NVIDIA.
According to the Financial Times, Microsoft purchased 485,000 NVIDIA Hopper GPU chips in 2024, more than double the amount purchased by Meta at 224,000. Microsoft has invested over $31 billion in capital expenditures on server infrastructure this year alone.
According to Reuters, Microsoft is permitted to customize OpenAI models to its needs with the existing agreement and is working to deploy smaller, open-weight models to incorporate into Copilot to improve performance and cost.
“We incorporate various models from OpenAI and Microsoft depending on the product and experience,” Microsoft said. OpenAI declined to comment.
Microsoft 365 Copilot criticized for cost and performance
Microsoft’s change in direction isn’t hard to predict.
In November 2023, Microsoft launched its flagship 365 Copilot offering (initially called Microsoft Copilot for 365), which was highly anticipated to mark a paradigm shift in generative AI adoption across the enterprise. Copilot includes multiple models by OpenAI, most notably GPT-4.
Instead, the launch had several missteps, most notably in cost and performance over time. It also included rebrands, such as shedding “Bing” titles.
Microsoft initially had several requirements for enterprise and small businesses to access 365 Copilot, including a 300-user license count minimum. Customers would require an existing E3 or E5 Microsoft 365 subscription to add on the Copilot capabilities. Existing E3 or E5 Office 365 license holders wouldn’t be eligible to add Copilot.
However, license and cost flexibility forced Microsoft to pivot, offering Copilot as an add-on to either Microsoft 365 or Office 365’s office productivity licensing model. Today, new or existing Microsoft 365 users can add Copilot for $30 per user per month.
“Copilot is more like Clippy 2.0”
License cost and entitlement aside, performance and user experience remain a common complaint of resource-intensive generative AI workloads.
Salesforce CEO Marc Benioff infamously declared Copilot “disappointing” and “slow” and compared it to Clippy.
“When you look at how Copilot has been delivered to customers, it’s disappointing,” Benioff wrote on X in October 2024. “It just doesn’t work, and it doesn’t deliver any level of accuracy.”
“Copilot is more like Clippy 2.0”, he doubled down.
Is generative AI still in the enterprise pilot stage?
Enterprise acceptance and widespread adoption of generative AI is still in its infancy.
Reuters reports that a Gartner study performed in October 2024 found that a “vast majority” of Microsoft 365 Copilot initiatives across 152 organizations had not progressed past the pilot stage.
Microsoft does not report specific sales data on the number of Copilot licenses sold but claims that 70% of Fortune 500 companies are using 365 Copilot.
