The AI in Your Excel Just Changed — and Microsoft Didn’t Ask You. Here’s Why It Matters.

When you ask Copilot to summarize a spreadsheet or draft a reply in Outlook, you probably don't think about which AI model answers. This week revealed that the answer has quietly changed for tens of thousands of prompts a week — and the shift, small as it still is, exposes the endgame of how power in the AI industry actually concentrates. gafam.ai reads what it means, especially for a Europe that runs its working day on Microsoft.

What Microsoft Is Doing — Stated Precisely

According to a Bloomberg report on July 7, corroborated by PYMNTS, Yahoo Finance and others, Microsoft has begun replacing OpenAI and Anthropic models with its own internally built MAI models inside Excel and Outlook, in an effort to reduce AI costs. Tens of thousands of AI prompts in those two widely used applications are now completed each week with MAI models — a scale of in-house usage that had not previously been reported. Microsoft declined to comment.

Two caveats matter, and gafam.ai will not skip them. First, this is not a clean break from the frontier labs. Microsoft maintains a multi-model strategy: routine, high-volume tasks shift to MAI, while frontier-grade tasks can still route to OpenAI or Anthropic — one outlet that initially framed it as a full replacement issued a correction to that effect. Second, MAI still accounts for a small share of Microsoft's overall AI usage. This is a beginning, not a completed migration, and the specific scale figure rests on Bloomberg's single sourcing.
But the direction is not ambiguous, because Microsoft's own AI chief has stated it plainly.

The Quote That Removes Any Doubt

At Build 2026 in June, Microsoft AI CEO Mustafa Suleyman was blunt about the goal: "We pay a lot of money to Anthropic — so our goal is to reduce and ultimately eliminate that cost." Not reduce — eliminate.

That framing is the whole story. Microsoft was reportedly on track to spend around $500 million a year on Anthropic's models, on top of its $13 billion-plus in OpenAI. It announced seven new MAI models at Build 2026, including one it says matches the coding ability of Anthropic's still-popular Opus 4.6 at lower cost, and it is extending MAI into GitHub Copilot and Teams. The clock is ticking on the discounted access Microsoft enjoys through its OpenAI partnership, and Suleyman's team is building so that Microsoft is never left paying whatever the leading labs decide to charge.

The Demotion — The Real Structural Story

Strip away the specifics and here is the dynamic. Microsoft is not terminating its partnerships with OpenAI or Anthropic. It is demoting them — from the default infrastructure underneath its products to a premium option reserved for the hardest tasks.

This is the endgame of what analysts have called the subsidized AGI economy. OpenAI and Anthropic grew enormous on the back of hyperscaler distribution — Microsoft's Office install base, Azure's reach, Amazon's Bedrock, Google's cloud. That distribution carried a hidden dependency. The distributors accumulated the usage data, the infrastructure, and — as of Build 2026 — their own competitive models. Now they can route the volume wherever they choose. The frontier labs keep the headline partnerships and the prestige frontier tasks. They lose the enormous, repetitive, commodity inference load that actually pays the bills. Revenue erosion from below, not from the top.
It connects directly to the token-cost reckoning gafam.ai covered last week: when Tesla, Uber and Meta started rationing AI spend, that was customers responding to token costs. Microsoft demoting the labs inside its own products is the same pressure, one layer up — the distributor engineering its way out of paying the labs at all.

The European Perspective