Microsoft AI chief wants out of OpenAI’s shadow

The awkward interrogation on video

Last fall, a high-level video call between Microsoft and OpenAI executives turned distinctly tense.

According to a report from The Information, Mustafa Suleyman, head of Microsoft’s in-house AI division, pressed OpenAI staffers to explain the inner workings of their latest reasoning model, o1.

This wasn’t just intellectual curiosity. It was a clear sign of Microsoft’s behind-the-scenes push to break free from the startup it has backed with more than $13 billion since 2019.

Suleyman, a co-founder of DeepMind and Inflection AI who joined Microsoft in March 2024 to lead Microsoft AI, has spent the past year pushing for independence. The message is simple: Microsoft cannot afford to tie its entire AI roadmap to a single outside partner.

Replacing the irreplaceable

Two big factors are driving Microsoft’s restlessness: mounting bills and fragile governance.

OpenAI’s frontier models, including GPT-4 running inside Copilot, are punishingly expensive to host on Azure. With The New York Times reporting an estimated $5 billion loss for OpenAI in 2024, Microsoft has strong financial incentives to lean on smaller, cheaper in-house models.

Then came the boardroom chaos of late 2023, when Sam Altman was briefly pushed out. That episode served as a wake-up call for Microsoft leadership about the risks of leaning on an unpredictable startup.

Suleyman’s team has already begun testing third-party alternatives from xAI, DeepSeek, and Meta, while actively swapping out OpenAI models in Copilot where feasible. Back in December 2024, Reuters reported that Microsoft had started plugging its own Phi-4 model into Microsoft 365 Copilot to shave latency and costs. Chatter on X also suggests Microsoft aims to build reasoning models that directly compete with OpenAI.

Still, The Information reports that Suleyman is making slow progress after unspecified setbacks. OpenAI’s technology is deeply woven into Windows, Office 365, and Bing. Even though a January contract tweak allowed OpenAI to work with cloud providers like Oracle, Microsoft retains exclusive rights to OpenAI models for its products until 2030. Unwinding that kind of marriage is easier said than done.

Why it matters

We think Microsoft’s hedging strategy makes total sense on paper, but pulling it off without degrading product quality will be brutal.

For enterprise customers and everyday Copilot users, Microsoft swapping out OpenAI engines behind the scenes could bring faster responses and lower subscription costs over time. The real risk is quality: if Microsoft replaces GPT-4 or o1 with in-house or open-source models that fall short on complex logic, users will notice immediately.

For the wider tech industry, this report shows that the most celebrated partnership in generative AI is increasingly transactional. Big tech will gladly write multi-billion-dollar checks to get an early edge, but no cloud titan wants to remain a glorified compute reseller forever.