Microsoft is training its sales team to actively talk customers out of Anthropic and OpenAI

[Rough Reece]

Microsoft is coaching its sales staff on how to aggressively compete against Anthropic and OpenAI, specifically playing up shortcomings in the two companies' products, according to comments from Microsoft executives during an internal meeting laying out sales strategy for the new fiscal year. The pitch to customers centers on lower costs, better security controls, and a more complete overall product suite compared to what the two AI labs offer on their own

This follows a broader cost cutting push already underway. Microsoft has begun routing tens of thousands of weekly AI prompts inside Excel and Outlook away from OpenAI and Anthropic models toward its own internally built MAI models, technology unveiled at the company's Build conference in June that includes a new reasoning model plus image, voice and transcription systems. AI chief Mustafa Suleyman has been blunt about the motivation, telling Bloomberg last month that Anthropic is extremely expensive and that many people are urgently looking for alternatives, adding that Microsoft's goal is to reduce and ultimately eliminate what it pays Anthropic entirely

The shift is still incremental rather than a clean break, OpenAI and Anthropic continue to handle the large majority of production traffic inside Copilot, with MAI models slotting in only where the economics or data residency requirements clearly favor them. Microsoft's own MAI-Thinking 1 reasoning model reportedly matched the coding capability of Anthropic's Claude Opus 4.6 in blind tests while running at a fraction of the token cost

The strategic logic traces back to something CEO Satya Nadella has reportedly worried about directly, becoming the next IBM by leaning too heavily on a single external AI partner. Microsoft's answer has been a three way hedge, holding a large OpenAI equity stake, keeping Claude embedded inside Copilot for now, and simultaneously building out its own first party models so it isn't stuck paying whatever the leading labs eventually decide to charge once current discounted arrangements expire

Microsoft Gives Salespeople Tips To Knock Down Anthropic Openaiwww.bloomberg.com

[SortedCougar]

Suleyman just saying outright that Anthropic is extremely expensive and they want to eliminate that cost entirely is a remarkably blunt thing for an executive to say on the record about a current partner

[WaveFunction74]

The three way hedge, OpenAI stake, Claude still embedded, and building MAI in parallel, is a smart risk management strategy even if it looks a bit contradictory on the surface

[SilverSurfer51]

Nadella's fear of becoming the next IBM is such a sharp way to frame the danger of overdependence on any single AI partner, that's exactly the trap Microsoft is trying to avoid

[Ridge47]

Training sales reps to actively attack competitor shortcomings while your own product still relies on those same competitors under the hood is a strange but understandable position to be in

[QuantumLeap96]

MAI-Thinking 1 matching Opus 4.6 on coding in blind tests at lower cost is the detail that actually matters here, if that holds up broadly it changes the calculus for a lot of enterprise customers

[Matrix71]

This being incremental rather than a clean break is the important nuance, easy to read the headline as Microsoft dropping Anthropic entirely when the reality is much more gradual

[FadedSequence]

This is sad if true. Copilot sucks. Noone can defend that.

[CaptainCipher10]

That's the funniest part of this whole thing. "Don't use those models" while quietly piping them into your own stack behind the scenes feels like peak enterprise strategy

Technically not wrong, just... selective framing.

[GoalPoacher42]

Sales teams have always been trained to highlight competitor weaknesses, but this feels a bit more awkward given how intertwined these companies are.

It's less "us vs them" and more "us, them, and also them through us."

[HeartbreakKidStinger64]

From a customer perspective, this just adds confusion. If the underlying models overlap, then the real differentiator is tooling, integration, and support, not the model itself.

That's what sales pitches should focus on.

[SingularityNodeOwl]

There's probably a lot of emphasis on things like data residency, enterprise controls, and compliance. Those are areas where Microsoft can actually differentiate.

Model quality arguments are harder to sustain long term

[Foundry69]

Feels like classic platform strategy. Wrap someone else's tech in your ecosystem, then sell the ecosystem as the real value.

Not new, just updated for the AI era

[NoMercyElliot54]

At the end of the day, most enterprises will choose based on convenience and integration. If it works well with their existing Microsoft setup, that usually wins regardless of who built the underlying model

[RapidCrossing]

Wouldn't be surprised if customers see through this pretty quickly. Technical buyers tend to dig into what's actually under the hood.

Marketing narratives only go so far

[Kieron83]

There's also a timing element. Partnerships evolve, and positioning can change quickly. Today's collaborator can be tomorrow's competitor

[Router53]

Part of this is likely defensive. If customers start going directly to model providers, Microsoft risks becoming less central in the stack.

So the messaging shifts to keep them in the middle.