Microsoft has spent years reselling other companies’ artificial intelligence. At an internal strategy meeting for its 2027 financial year, first reported by Bloomberg on 15 July, its executives told salespeople to start selling against it.

The reported shift matters because it comes from the vendor with the most to lose from model-first buying. Microsoft resells OpenAI’s and Anthropic’s models inside its Copilot assistant and its Azure cloud, yet it is now coaching sellers to argue that the underlying model is beside the point, and that cost, integration and security decide the purchase instead. For marketing teams weighing AI tools, that is the buying criterion the best-resourced platform in the market now wants them to adopt.

What is Microsoft telling its sales teams?

Microsoft is instructing its sales force to position its own AI as a complete system rather than a single clever model, according to Bloomberg’s account of the meeting. The session, billed as a strategy briefing for the financial year that began on 1 July, leaned on the cost and efficiency of Microsoft’s in-house models against those of its rivals. “Everyone else is selling parts — we’re selling the full end-to-end system,” executive vice-president Jay Parikh reportedly told the room, framing that as the line sellers should carry through the year.

One executive went further. Jacob Andreou, an executive vice-president working on Copilot, reportedly presented a direct comparison with Anthropic’s Claude, telling colleagues that inside Microsoft’s Office apps Claude was “slower and less accurate, and lacked the proper security integrations.” That characterisation comes from an internal sales presentation, not an independent benchmark, and both Microsoft and Anthropic declined to comment. Anthropic’s models are themselves available to enterprise customers inside Microsoft 365 Copilot, which makes the criticism a pitch against a product Microsoft also resells.

Why is Microsoft turning on its own AI suppliers?

Microsoft is trying to shift enterprise customers onto its own models and cut its dependence on the suppliers whose technology it has leaned on. A report earlier in July said the company had begun swapping OpenAI’s and Anthropic’s models out of flagship apps such as Word and Excel in favour of its in-house MAI models, described as a cost-cutting move. The sales messaging follows the same logic: sell what Microsoft owns, and keep more of the margin.

The commercial backdrop is a partnership that has loosened. Microsoft holds a reported 49% profit-sharing stake in OpenAI and once had exclusive access to its models; in April the two companies amended their agreement, dropping the exclusivity clause and freeing OpenAI to sell through Microsoft’s competitors. Competing against a company it part-owns is awkward, but the sum in play is large. Microsoft’s AI business passed a $37bn annual revenue run rate in its most recent quarter, up 123% year on year, and steering customers to its own models protects more of that revenue while trimming the fees it pays outside suppliers.

Cost is the argument Microsoft expects to land. Chief executive Satya Nadella reportedly told the meeting that helping customers manage AI spending would become a priority over the coming year, and pointed to Unilever, which he said had saved about $300m after moving an automated claims-processing system from an unnamed frontier model to a cheaper Microsoft one. The figure is Microsoft’s own and the model it replaced was not named, but it illustrates the pitch: when two systems do a job well enough, the cheaper one wins.

Does model quality still decide the AI purchase?

Model quality is becoming one factor among several rather than the whole decision, and that reframing is the part marketing teams should register. For the past two years, AI vendor pitches have turned on benchmark scores and model capability. Microsoft is betting that as models converge, buyers will care more about what a tool costs to run, how cleanly it fits the software they already use, and whether it meets their security and data-residency rules — the ground on which an incumbent platform, not a specialist model, tends to win.

For marketers choosing or recommending AI tools, that points to shortlists built around integration depth, total cost of ownership and compliance, not model leaderboards alone. It also raises the bar on the claims themselves. Microsoft’s line that Claude is “slower and less accurate” is a sales argument, not a tested result; the security point carries more weight, since Claude runs on Amazon and Google infrastructure outside Azure, which can matter for teams bound by strict data-residency requirements. Marketers are better placed than either vendor to test such assertions in their own stack — a company talking down a rival it also resells is making a case, not reporting a finding.

The same shift reaches marketers’ own messaging. Teams that have built product positioning on a named model — “powered by GPT”, “built on Claude” — now face a better-funded narrative that treats the model as interchangeable and sells the platform wrapped around it. The counter is not a louder capability claim but evidence: what the tool does in a real workflow, at what cost, under whose security terms.

Microsoft’s 2027 financial year runs to next June, and its coming earnings calls will show whether the cost message moves enterprise spending or merely reassures investors funding its AI build-out. Whether buyers accept that the model no longer decides the purchase is the question the pitch leaves open.