Microsoft is Spending $2.5 Billion on a Company That Helps Businesses Make AI Actually Work

Microsoft announced Tuesday it’s spinning up a new operating unit called Microsoft Frontier Company, backed with an initial $2.5 billion investment. The goal: help enterprises like Unilever and Novo Nordisk figure out which AI tools actually work for their business — and then make them pay off.

Large companies have stopped relying on a single AI provider like OpenAI or Anthropic. Instead, they’re mixing and matching — pulling in open-source models, renting proprietary ones, fine-tuning everything against their own data. It’s more flexible, but it’s also expensive and slow to show returns.

Microsoft Frontier is designed to cut through that mess. The company will help clients integrate Microsoft’s own AI tools alongside third-party models, connect them to internal data, and deploy the result. Crucially, whatever gets built belongs entirely to the customer. None of it flows back to Microsoft.

The move mirrors what Palantir has been doing with Nvidia’s open-source models, and what AWS has pursued through a $1 billion embedded engineering unit. Microsoft is late to a trend its competitors helped define.

Judson Althoff, Microsoft’s president of commercial business, told Reuters that the company learned its lesson the hard way. “Three years ago when we built Copilot, we made a mistake — we only locked it to OpenAI’s models,” he said. “What enterprises really need is a model that amplifies their own business capabilities, with the flexibility to swap in the best model at any time and fine-tune it themselves.”

That sentiment mirrors a broader anxiety playing out in boardrooms. Patrick Moorhead, CEO of Moor Insights & Strategy, noted that companies worry prolonged reliance on OpenAI or Anthropic models could eventually give those labs technical knowledge that rivals their own — particularly in areas like code generation and legal work.

Microsoft already holds a stake in OpenAI and earlier this year added Anthropic’s models to Copilot in response to rising enterprise demand. But Althoff acknowledged that the landscape has shifted. DeepSeek from China, Google’s Gemini, and other models have closed the gap with OpenAI faster than anyone expected.

“For enterprise customers, the combination of their own data with a model that fits their needs matters far more than any single model,” Althoff said. “They need a flexible setup where they can swap AI models quickly.”

Microsoft Frontier launches at a time when enterprise AI spending is under intense scrutiny. Companies poured billions into experimentation over the past two years, but many are still searching for tangible returns. Microsoft is betting that a dedicated, vendor-neutral-sounding intermediary — one that keeps customer data walled off — is what it will take to unlock the next phase of adoption.