Microsoft Reportedly Tells Sales Team to Pitch Its Own AI Over OpenAI, Anthropic

Microsoft has told its sales force to position the company's own AI models against those of OpenAI, Anthropic and Google, according to a Bloomberg report summarized by TechCrunch on July 15. The message to sellers, delivered at an internal meeting on Tuesday, July 14, centered on the cost and efficiency of Microsoft's in-house models. For CIOs who bought Copilot partly because it gave them access to leading third-party models, the shift raises a practical question: which model is answering their employees' questions, and who decides when it changes?

What did Microsoft reportedly tell its salespeople?

The meeting was billed as a strategy session for Microsoft's new fiscal year, according to TechCrunch. Executives reportedly urged sellers to compare rival AI products unfavorably with Microsoft's, arguing that competitors sell components while Microsoft sells a complete system.

Jacob Andreou, the executive vice president for Copilot, reportedly presented a comparison between Copilot and Anthropic's Claude. Inside Microsoft's Office apps, he said, Anthropic's model was "slower and less accurate," and lacked the right security integrations, according to the report. TechCrunch said it had asked Microsoft and Anthropic for comment, and it noted that it had not independently confirmed Bloomberg's account. Microsoft had not publicly addressed the report in that story.

Why is Microsoft moving away from OpenAI and Anthropic models?

The sales message follows an earlier change in Microsoft's products. On July 7, Bloomberg reported that Microsoft had replaced OpenAI and Anthropic models with its own AI in some apps. TechCrunch described those swaps as affecting flagship apps such as Word and Excel and characterized them as a cost-cutting move.

The relationship with OpenAI has also loosened. According to TechCrunch, the two companies amended their partnership in April, dropping a clause that had given Microsoft exclusive access to OpenAI's API and models. That frees OpenAI to sell through Microsoft's rivals, and it gives Microsoft more reason to lean on models it controls. TechCrunch also pointed out that Microsoft shares have lagged over the past year as investors questioned its AI infrastructure spending, which makes margin on every Copilot query more important.

What does this mean for Copilot customers?

Nothing in the report says Copilot buyers will lose features or pay more. But it does suggest the model underneath a familiar interface can change for reasons that have more to do with the vendor's costs than with the customer's results. That has three consequences for enterprise teams.

First, output quality can shift without notice. A summarization or drafting workflow that was tuned against one model may behave differently on another. AI Connect speakers have warned about exactly this. In a AI Connect webinar on AI agent observability, Monte Carlo's Ethan Post said teams should test model changes against ground-truth evaluations before switching. When the vendor does the switching, the customer needs its own evaluations to notice.

Second, governance reviews may need updating. Risk and privacy teams often approve a tool based on which model provider processes the data. If a different model now handles some requests, those records should reflect it.

Third, the claims in a sales pitch are not benchmarks. Microsoft's comparison of Copilot and Claude was made by Microsoft, inside Microsoft's own apps. Buyers should ask for evidence on their own documents and tasks.

How should enterprises reduce dependence on a single AI vendor?

The episode strengthens the case for architecture that does not assume any one model or provider will stay fixed. "You don't want their SLAs to be your SLA," Joe McCarron, head of developer advocacy at Skyflow, said of model providers during a AI Connect webinar on scaling AI architecture. The panel recommended model-agnostic interfaces so teams can swap models without rebuilding applications.

That advice has been consistent at AI Connect events. In a 2024 panel on self-hosted LLMs and vendor lock-in, SandboxAQ's Stefan Leichenauer urged teams to test several vendors in proofs of concept and to ask how painful a later switch would be. And in a July 8 webinar, Mistral AI's Aaron Delp argued that enterprises should own their AI rather than rent it, a pitch that will sound more persuasive to some buyers after this week.

Practical steps include asking Microsoft account teams which models serve each Copilot feature and how customers are told about changes; keeping a small evaluation set of real business tasks and rerunning it monthly; and, for custom applications, routing model calls through a gateway that allows a change of provider without code rewrites.

The bigger picture is that the largest AI platform vendors are now competing with the model makers they once resold. Watch whether Microsoft offers enterprise customers clearer model choice inside Copilot, or whether in-house models become the default with third-party options at a premium. Either path will shape software budgets for fiscal 2027.

Frequently asked questions

Is Microsoft replacing OpenAI models in Copilot?

Bloomberg reported on July 7 that Microsoft replaced OpenAI and Anthropic models with its own AI in some apps; TechCrunch said those included Word and Excel.

What did Microsoft tell its salespeople?

At a July 14 meeting, executives reportedly urged sellers to stress the cost and efficiency of Microsoft's in-house models against rivals, according to a Bloomberg report summarized by TechCrunch.

How can enterprises avoid AI vendor lock-in?

Use model-agnostic interfaces, keep evaluation sets of real tasks, and ask vendors which models serve each feature and how changes are communicated.

Disclosure: AI Enterprise Magazine uses AI tools in its editorial process, including Anthropic's Claude models. Anthropic had no role in this article.

Filed under: AI Strategy & ROI, AI Governance & Trust