Microsoft’s New AI Strategy: Copilot Without the Frontier Price Tag
Microsoft’s current AI strategy in Microsoft 365 AI is to keep Copilot visible and useful for users while quietly replacing expensive third-party models like ChatGPT and Claude with internal AI models that can answer the same prompts at lower cost, especially in high-volume apps such as Excel and Outlook. To cut Microsoft Copilot costs, the company is pushing to have prompts inside Microsoft 365 answered by its own MAI systems rather than by OpenAI and Anthropic models. Although Copilot launched as a showcase for external frontier models, Microsoft has begun funneling at least some Office prompts away from those partners and toward its proprietary infrastructure. In other words, Copilot’s brand stays; the brains behind it are being swapped out where Microsoft thinks users will not notice.
Why Microsoft Is Done Subsidising Anthropic and Other AI Labs
The driving motive here is financial, not ideological. Every Copilot response carries an inference cost: the compute spent each time a model processes a prompt and generates an answer inside Microsoft 365 AI. At enterprise scale, those per-prompt costs stack up across email, spreadsheets, documents, meetings, and automation workflows. To cut costs, Microsoft is pushing more of those prompts to internal AI models instead of paying outside providers. Mustafa Suleyman, Microsoft’s CEO of AI, has been blunt about the pressure these third-party bills create. He said the company pays "a lot of money" to Anthropic and that the goal is to "reduce and ultimately eliminate" that cost. This comes after Microsoft reduced workers’ access to tools like Claude Code when costs skyrocketed, a clear sign that open-ended, uncapped access to external models is no longer sustainable.
Excel and Outlook Become the Testbed for Internal AI Models
Microsoft is reportedly beginning to route selected AI prompts in Excel and Outlook to its own Microsoft AI (MAI) models, a limited but significant cost-cutting move inside Microsoft 365. Historically, Copilot in these apps relied on models from OpenAI and Anthropic, but Microsoft has started directing at least some of that traffic to its own systems. This change still covers a small share of overall prompts, with MAI handling tens of thousands of Excel and Outlook queries per week rather than dominating usage. The test is focused on predictable enterprise workflows: spreadsheet explanations tied to a file, email drafts anchored to a specific mailbox or thread, and similar bounded tasks where context is narrow and repeatable. Microsoft’s bet is that these routine prompts do not need the most expensive frontier systems, and that tuned internal models can deliver comparable quality while keeping Microsoft Copilot costs under control.
What Changes (and Doesn’t) for Everyday Copilot Users
For everyday Microsoft 365 customers, the key question is simple: will Copilot stay useful as Microsoft swaps in cheaper internal AI models? The model badge is secondary; what matters is whether spreadsheet explanations, email drafts, meeting summaries, and transcriptions remain accurate and fast. A lower inference bill helps Microsoft, but it helps customers only if common Copilot responses keep their quality. A cheaper model that drafts weaker emails, misreads spreadsheet context, or produces less reliable summaries would turn a back-end cost saving into a front-end product problem. That is why this shift is best understood as a workload-routing test, not a full platform reset. Microsoft appears to be asking a narrow question first: which repetitive Office tasks can be handled by smaller in-house models without reducing the quality users expect from Copilot? Until the answer is proven, frontier models like Claude remain part of the mix as a ChatGPT alternative.
From Dependency to Self-Sufficiency: The Long-Term Copilot Play
Zooming out, the Excel and Outlook experiments signal a deeper strategic transition for Microsoft: from dependency on third-party AI providers to a proprietary model stack that it controls end-to-end. Microsoft’s official announcement of seven in-house MAI models spans image, voice, transcription, coding, and reasoning, giving the company a menu of specialised systems to route tasks to rather than sending everything to a single frontier model. These MAI models are already available in Copilot for Business and Enterprise users, with MAI-Code-1-Flash positioned as a coding assistant competing with Claude Code and OpenAI’s Codex. Microsoft wants more control over the models behind its products, where it can tune systems to specific Office tasks and avoid paying another provider for every request. If internal models can handle routine workloads across Excel, Outlook, Teams and beyond at lower operational expense, external partners will shift from default engines to premium options used only when their quality edge is clear.
