For more than a year, Microsoft has been signaling that its long honeymoon with external frontier models was coming to an end. The company has been steadily promoting its own SMLs across Azure, Copilot, and internal developer tooling. Now reports say the shift is accelerating inside Microsoft’s flagship apps, where the company is beginning to replace OpenAI and Anthropic models with its own in-house systems.
If you have been following our coverage, this moment has felt inevitable. Microsoft has been laying the groundwork through a mix of technical previews, internal documentation, and subtle product updates that all point toward one goal. The company wants to control more of its AI stack and reduce the cost and complexity of relying on external partners.
The latest reporting describes Microsoft using its own models to handle large volumes of AI prompts inside apps like Excel and Outlook. These are not experimental deployments. They are production workloads in two of the most widely used applications on the planet. That alone tells you how confident Microsoft has become in the performance and reliability of its internal models.
The broader context matters here. Microsoft has spent the past two years building out the infrastructure required to support this kind of independence. The Maia and Cobalt chips, the expanded global datacenter footprint, and the push for more efficient model architectures all point to a company preparing to run more of its AI on its own silicon and designs. The strategy is not subtle. It is a long-term play to reduce dependency on OpenAI and Anthropic while still keeping those partnerships alive for higher-end or specialized workloads.
Bloomberg’s reporting highlights the cost angle, which has always been the quiet part of the story. Running frontier models at scale is expensive. Running someone else’s frontier models at scale is even more expensive. Microsoft has been looking for ways to bring those costs down without sacrificing capability. The rise of its SMLs is the clearest sign of how it plans to do that.
There is also a strategic dimension. Microsoft has spent years building Copilot into the center of its product ecosystem. That ecosystem is stronger when Microsoft controls the models that power it. Internal models give the company more flexibility, more control over tuning, and fewer external constraints. They also reduce the risk of being caught in the crossfire of another company’s roadmap or regulatory challenges.
None of this means Microsoft is walking away from OpenAI or Anthropic. The partnerships remain valuable, and the company still relies on frontier models for certain experiences. But the balance is shifting. Microsoft is moving from a world where external models were the default to a world where internal models are the baseline and external models are the premium option.
For anyone watching the industry, this is one of the most important transitions happening right now. It signals a future where major tech companies run more of their AI on their own terms. It also shows how quickly the AI landscape is changing. What was once a dependency is becoming an option.
And for Microsoft, it is a reminder that the company has been preparing for this moment far longer than most people realized.
