Microsoft is coaching its salespeople to talk down the models it still runs on

Microsoft executives spent an internal meeting on Tuesday teaching the sales force how to run down OpenAI, Google, and Anthropic, which would be unremarkable except that two of those companies supply the models inside Microsoft’s own products.

The session was billed as a strategy kickoff for the fiscal year that began this month, and it leaned on cost and completeness. The company’s in-house models are cheaper and more efficient, the argument goes, and they arrive attached to everything else Microsoft sells.

It is the sales-floor expression of a shift that has been running all year, as Microsoft works its own MAI models into products that used to run on other people’s.

“Everyone else is selling parts,” executive vice president Jay Parikh told the room. “We’re selling the full end-to-end system. That’s the story that we all need to get out there and tell in FY27.”

Jacob Andreou, another executive vice president, went further with a presentation setting Copilot directly against Anthropic’s Claude. Inside Microsoft’s office apps, Andreou said, Anthropic’s model was slower and less accurate, and lacked the proper security integrations.

Claude is currently embedded in Copilot. So the pitch, stripped down, is that Microsoft’s salespeople should tell customers that one of the engines under the bonnet is the weaker option, while continuing to ship it.

Sales teams are coached to knock competitors all the time, and there is nothing improper here. What makes it notable is the target. Microsoft spent years and a great deal of money building dependence on exactly these firms, and it is now selling against them.

The direction has been visible for months. Microsoft began swapping OpenAI and Anthropic models out of Word and Excel earlier this month, a move reported as cost-cutting.

Its AI chief has said outright that the company wants to eliminate what it pays Anthropic.

None of which requires a conspiracy. Anthropic charges for its models, Microsoft would rather not pay, and the cheapest way to stop paying is to convince the buyer that the thing you were reselling was never that good.

Anthropic is the expensive dependency, and Microsoft knows the bill better than almost anyone, because its own engineers were among the heaviest Claude users outside Anthropic’s customer base. It has since quietly pulled back from Claude Code internally.

The structural change underneath all of this happened in April, when Microsoft and OpenAI amended their partnership and dropped the exclusivity clause, freeing OpenAI to sell to Microsoft’s competitors.

Once your partner can sell to your rivals, it is no longer obvious why you should not sell against your partner. Amazon began offering OpenAI models through AWS shortly after.

There is a market read here too. Microsoft has spent the past year defending its AI capital spending to investors who are visibly unconvinced, and a $357bn drawdown concentrates the mind.

Telling the sales force to argue that the in-house models are not merely cheaper but better is a way of arguing that the spending bought something.

Whether customers accept the argument is a separate matter. Claude has become the default in enterprise AI coding tools, and buyers who chose it did so with the benchmarks in front of them.

A Microsoft slide asserting the opposite, delivered by a Microsoft salesperson with a Microsoft quota, is not the same evidence.

TechCrunch approached Microsoft and Anthropic. Neither had commented at the time of publication.

The three-way hedge is still intact for now. Microsoft holds a large OpenAI stake, ships Anthropic’s models in Copilot, and sells its own MAI line against both.

That is a defensible position while it lasts, and Tuesday’s meeting is a reasonable guide to which of the three the company would prefer to be selling in a year.