Microsoft has begun routing some workloads in Excel and Outlook to its own internally developed AI models, replacing selected OpenAI and Anthropic models as part of a broader push to reduce enterprise AI inference costs. The company is now processing tens of thousands of prompts each week through these in-house models, though they still represent a small fraction of overall usage. A Microsoft spokesperson declined to comment on the change, which was first reported by Bloomberg.
From frontier models to deployment economics
At the Build conference in June, Microsoft AI CEO Mustafa Suleyman introduced seven new MAI models for reasoning, coding, transcription, and other tasks. He said the company wanted to "reduce, and ultimately eliminate, spending on Anthropic models." The MAI-Code-1 model, for example, delivers coding performance comparable to Anthropic's earlier Opus 4.6 model at a lower operating cost, Suleyman said.
The move into production, reported by Bloomberg, shows Microsoft executing on a strategy CEO Satya Nadella has outlined publicly: long-term AI leadership will depend on infrastructure, deployment capabilities, and ecosystem efficiency as much as model capability. Every Copilot interaction consumes computing resources, and as enterprise adoption grows, small per-request savings can compound into significant operational gains.
A portfolio of models for different workloads
The shift reflects an architectural pattern spreading across enterprise AI platforms. Instead of relying on a single foundation model, vendors assemble portfolios optimized for different tasks. Complex reasoning may still require the most capable frontier models from OpenAI or Anthropic, but routine activities like email assistance, summarization, or document generation can often be handled by smaller, less expensive models without a noticeable difference for users.
This multi-model routing approach matches workloads with the optimal combination of performance, latency, and cost. For Microsoft, the immediate goal is not to abandon partnerships with OpenAI or Anthropic, but to manage the economics of serving millions of enterprise users at scale. The architectural shift is also being explored by AI leaders looking to apply similar principles across their own operations, a topic covered in AI for Executives & Strategy.
Why this matters for Executives and Strategy
For business leaders, the development signals that the enterprise AI market is entering a new phase. The competitive differentiator is no longer just access to the most advanced model, but the ability to deliver AI capabilities at sustainable cost. Microsoft's internal model deployment shows that even the largest cloud providers are prioritizing operational efficiency over raw capability in their own product suites. As AI adoption scales across the enterprise, the economics of inference will directly impact margins, pricing, and the pace of deployment.
Your membership also unlocks:
