Microsoft Trains Sales Team to Trash OpenAI, Anthropic

The Microsoft-OpenAI marriage is over in everything but name

According to Bloomberg, Microsoft held an internal FY27 strategy meeting this week. The message to its sales force was unmistakable: OpenAI and Anthropic are no longer partners — they're competitors to be beaten on every deal.

EVP Jay Parikh set the frame, per Bloomberg: "Everyone else is selling parts — 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."

Bloomberg reports EVP Jacob Andreou went further, delivering a slide deck that directly compared Microsoft 365 Copilot to Anthropic's Claude. His assessment, per Bloomberg: Anthropic's model was "slower and less accurate, and lacked the proper security integrations" when working inside Microsoft's Office apps.

Neither Microsoft nor Anthropic had responded to requests for comment at press time. OpenAI has not yet responded to this specific story — but its July 9 "preferred model" blog post now reads like a preemptive counter-punch.

What happened

Bloomberg's reporting marks the third major signal in two weeks that Microsoft is systematically decoupling from its AI partners:

1. July 8 — Bloomberg reported Microsoft is routing Office AI prompts from OpenAI and Anthropic models to its own MAI models, explicitly as a cost-cutting measure.
2. July 9 — OpenAI published a blog post asserting GPT-5.6 remains the "preferred model" for Microsoft 365 Copilot — an unusual public defense against the MAI-in-Office narrative.
3. July 15 — Microsoft is now training its salespeople to actively pitch against both OpenAI and Anthropic in enterprise deals, coaching them to position MAI models as faster, more accurate, and better integrated.

This follows last week's report that Microsoft is already routing Office AI prompts away from OpenAI and Anthropic to in-house MAI-Code-1-Flash models.

The backdrop: Microsoft and OpenAI revised their partnership in April 2026, giving OpenAI the right to sell its technology to Microsoft's competitors. Since then, Microsoft has accelerated its in-house MAI model family — MAI-Code-1-Flash is already generally available on GitHub Copilot, and seven new models were unveiled at Build 2026.

Why Microsoft's sales offensive matters

This is competitive warfare disguised as cost efficiency. Microsoft is simultaneously:

• The largest infrastructure backer of OpenAI
• A major Anthropic distribution partner via Azure Foundry
• Actively replacing both companies' models with in-house alternatives
• Training a global sales force to position those alternatives as superior

If Bloomberg's account is accurate, the contradiction is glaring: Microsoft runs OpenAI's models on its own infrastructure, profits from every API call, and is now telling enterprise buyers those same models are slower, less accurate, and less secure than the ones it built itself.

OpenAI had already tried to preempt this with its blog post declaring GPT-5.6 the "preferred model" for Microsoft 365 Copilot — but Microsoft's internal sales training tells a different story.

For the broader AI market, the signal is unmistakable. When the world's largest enterprise software company — which happens to be the biggest customer of both OpenAI and Anthropic — decides to vertically integrate, every enterprise AI deal becomes a three-sided fight. The partnership era of AI infrastructure is ending.

What changes for you

If you're an enterprise buyer evaluating Microsoft 365 Copilot, understand the lock-in you're signing up for. Microsoft's MAI models are purpose-built for the Microsoft stack and increasingly the only option surfaced inside Office apps. You're not choosing a model — you're choosing an ecosystem.

Before committing to Copilot at scale:

• Test whether MAI models meet your accuracy and latency requirements on your actual workloads — not vendor benchmarks.
• Evaluate whether losing model optionality inside Office matters for your compliance, security, or cost requirements.
• Watch whether Microsoft extends this routing strategy to Azure AI Foundry. Today it's Copilot. Tomorrow it could be the API gateway.

FAQ

Is Microsoft actually dropping OpenAI and Anthropic from its products?

Not formally. Both companies' models remain available through Azure AI Foundry. But inside Microsoft 365 Copilot — the product millions of enterprise users interact with daily — Microsoft is actively routing prompts away from partner models toward its own MAI family. The sales training confirms this is strategic, not circumstantial.

What are MAI models?

MAI (Microsoft AI) is Microsoft's in-house model family, built by the same team that develops Phi. MAI-Code-1-Flash shipped in June 2026 and is already GA on GitHub Copilot. Seven additional MAI models were announced at Build 2026, covering a range from small fast models to frontier-scale systems. The models are optimized for Microsoft's own infrastructure and products.

Should enterprises reconsider Microsoft 365 Copilot?

Not necessarily abandon it — but evaluate it with eyes open. Copilot is becoming a walled garden for AI models inside Office. If that's acceptable for your use case, the integration benefits may outweigh the lock-in. If model optionality matters, ask your Microsoft account team hard questions about routing transparency before you sign.