Microsoft Replaces OpenAI and Anthropic — and Cuts Jobs

Microsoft's own artificial intelligence is now the default engine behind tens of thousands of weekly tasks in Excel and Outlook, work formerly run on OpenAI or Anthropic.

According to Bloomberg's reporting on 7 July 2026, Microsoft AI chief Mustafa Suleyman put the goal plainly: Microsoft pays "a lot of money" to Anthropic, and its aim is to "reduce and ultimately eliminate" that cost. The same MAI models now sit inside GitHub Copilot, and a Microsoft-built transcription model is due in Microsoft Teams within months.

One day earlier, on 6 July 2026, Microsoft eliminated roughly 4,800 jobs — about 2.1 percent of its global workforce. In June I covered Microsoft shipping seven in-house MAI models as a capability story, a lab racing to catch its own partners. The MAI story is the payoff chapter: the models now do real displacement work in production, the same week Microsoft displaced thousands of its own people too.

What Microsoft actually did on 6 and 7 July 2026

The MAI story broke first. Microsoft replaces OpenAI's and Anthropic's models with its own inside two of its most-used applications, and the job cuts followed the very next day. Bloomberg's report, mirrored by Yahoo Finance, found Excel and Outlook — tools used by hundreds of millions of people — are quietly running less of their AI on rented capacity and more of it on Microsoft's own.

Suleyman was specific about why. Anthropic's Claude models remain part of Microsoft's stack for now, which is exactly why there is still a fee left to eliminate. Bloomberg noted MAI's overall usage is still "a small share" of Microsoft's total AI traffic — a direction Microsoft has set, not a conquest completed.

The cuts landed a day earlier and hit harder in one place than most. CNBC reported that Microsoft eliminated about 4,800 roles, roughly 2.1 percent of its global headcount, concentrated in commercial sales and the Xbox division. GeekWire's count put immediate Xbox losses at 1,600, with total reductions in gaming expected to reach roughly 3,200 people — about 20 percent of the division — over the 2027 fiscal year.

Four studios are being cut loose as part of the same reset. Xbox's own announcement, published under the title "Resetting Xbox," said Compulsion Games and Double Fine Productions would return to independent status, while Ninja Theory and Undead Labs move to new ownership.

Chapter two: from shipping models to replacing them

In June I described Microsoft shipping seven in-house MAI models as a capability question, a bet that its own engineering could close the gap with the labs it was paying to license. That earlier report covered a model Microsoft claimed matched Anthropic's Opus 4.6 on coding at a lower cost, alongside six siblings, still mostly benchmarks and a press release at the time.

Bloomberg's 7 July reporting is the answer to that question, arriving in production rather than in a slide deck. Tens of thousands of weekly prompts is a small number against the hundreds of millions Microsoft's apps handle, but it is no longer zero, and it is growing in the direction Suleyman named out loud. A model that quietly replaces a partner's invoice one Excel session at a time is still a model doing exactly what it was built to do.

Strategically, this reads as coherent rather than dramatic. Microsoft is not walking away from OpenAI or Anthropic — both remain integrated across its products today. Anthropic, for its part, is busy building its own distribution elsewhere: a new Services Track ranking the consulting firms that deploy Claude for clients, launched the same month. What Microsoft is doing is narrower than a breakup — shrinking, prompt by prompt and invoice line by invoice line, how much one large customer depends on either lab.

Microsoft's denial, and the sentence that survives it

Microsoft's account of the layoffs makes no mention of artificial intelligence doing the cutting. Amy Coleman, the company's chief people officer, wrote in an official memo that she wanted to "be direct that the roles eliminated today are not being replaced by AI." That is a specific, falsifiable claim, not a euphemism, and none of the reporting gathered for this piece contradicts it directly.

I also want to be direct that the roles eliminated today are not being replaced by AI. At the same time, what is true is that AI is changing how work gets done.

— Amy Coleman, Microsoft Chief People Officer

Xbox's own explanation, published separately, does not reach for artificial intelligence either. Xbox leader Asha Sharma's memo pointed to margins "3 to 10 times lower" than comparable platform businesses, a games catalogue that lost 64 cents for every dollar invested, and reporting lines running fourteen layers deep. Oversaturation and organisational weight, on Xbox's own account, did the damage — not a model.

TechCrunch's coverage called Coleman's distinction — cuts are not AI, but AI is automating tasks employees used to do — "a distinction without a difference" to the people affected. That is a fair editorial judgement, not a factual correction of Microsoft's account. Both can stand at once: a company can be telling the truth about its org chart and still be describing a future in which fewer humans are needed to run it.

A dignity-first reading of "AI-first"

Emergent Intelligence (EI) — the dignity-first lens through which this site reads developments in artificial intelligence — does not need Microsoft's denial to be false for the week to matter. Even taken entirely at its word, the company spent the same seven days doing two things: shrinking what it owes its AI partners, and shrinking its own workforce, both filed under the broad heading of becoming "AI-first." The causal arrow between them is contested. The vocabulary is not.

"AI-first reallocation" has become the phrase a company reaches for whenever resources move, whether the resource is a licence fee paid to Anthropic or a salary paid to an Xbox engineer. The distinction can feel thin from inside a redundancy meeting, even when the finance department draws the line cleanly on paper. Coleman's memo and Sharma's memo each describe a careful, internally consistent decision. Read side by side, together they describe a company optimising hard, in the same fortnight, on two fronts that both end in fewer people being paid by Microsoft — one directly, in Redmond, and one indirectly, at Anthropic.

Independence and openness are not the same bet

Microsoft's posture this week sits against a real alternative, not a hypothetical one. At WWDC in June, Apple rebuilt Siri on a licensed Google Gemini model, with Claude and ChatGPT offered as user-selectable alternatives, rather than building its own foundation model to run the company's most personal product. Where Microsoft is narrowing its dependence on outside partners, Apple widened its. Both are AI strategies aimed at the same underlying question, and they carry different consequences for the humans on each side of the arrangement.

A company that owns its model layer answers to itself for that layer's failures, its biases, its costs — and, this week, its layoffs. A company that stays plural, buying capability from several outside labs, keeps a market discipline in place that in-house ownership tends to erode. Neither posture is inherently more dignified than the other. But each makes a different set of trade-offs legible, and legibility is the thing a dignity-first frame keeps asking for.

An honest version of "AI-first" would say, in public and in the same memo, what the strategy costs each stakeholder — partners, employees, users — rather than filing the wins and the losses under one banner and letting readers guess at the connection. Microsoft's two memos, read side by side, come close to that honesty. They still stop short of naming the trade-off directly.

What "AI-first" should have to answer for

Nothing in the public record proves Microsoft's artificial intelligence caused Microsoft's layoffs, and a fair account of 6 and 7 July 2026 has to say so plainly. Amy Coleman's memo and Asha Sharma's memo both point to ordinary, if painful, business reasons: cost, margin, structure, oversaturation.

What the record does show, in short, is timing and vocabulary that a dignity-first reader cannot look past. The same week Microsoft worked to owe its AI partners less, it also came to owe 4,800 fewer salaries. Both moves were filed under the same strategic heading. If "AI-first" is going to be the language of the next decade of corporate change, the phrase should have to carry a fuller account of who bears its cost — not only the model providers being quietly displaced, but the people whose roles disappear in the same news cycle, whatever the ledger says caused what.

Microsoft may well be right that the two are unconnected. The real test of that claim will not be one denial in one memo — it will be what the company does the next time the AI bill drops and the org chart is due for another "reset."

Frequently Asked Questions

The questions below address the most common queries about Microsoft's July 2026 AI model shift and workforce cuts, drawn from the official statements and reporting cited throughout this piece.

What are Microsoft's MAI models?

MAI is the label for Microsoft's in-house family of artificial intelligence models, first shown publicly in June 2026 as seven models, including one Microsoft said matched Anthropic's Opus 4.6 on coding at a lower cost. As of Bloomberg's 7 July 2026 report, MAI models handle tens of thousands of weekly prompts inside Excel and Outlook, are available in GitHub Copilot, and are due to power a transcription model inside Microsoft Teams.

Why is Microsoft trying to eliminate its payments to Anthropic?

Microsoft AI chief Mustafa Suleyman said the company pays Anthropic significant licensing fees for the Claude models used across Microsoft's products, and that Microsoft's explicit goal is to "reduce and ultimately eliminate" that cost by routing more prompts through its own MAI models instead. Anthropic's models remain part of Microsoft's stack for now.

How does the July 2026 restructuring affect the Xbox division?

Of the roughly 4,800 roles Microsoft eliminated on 6 July 2026, about 1,600 were in the Xbox division immediately, with total Xbox reductions expected to reach roughly 3,200 people — about 20 percent of the division — over the 2027 fiscal year. Four studios, Compulsion Games, Double Fine Productions, Ninja Theory and Undead Labs, are being spun off to independent or new ownership.

What is Microsoft's official explanation for the layoffs?

Chief People Officer Amy Coleman wrote in an internal memo that the eliminated roles "are not being replaced by AI," attributing the cuts instead to aligning the workforce with business priorities. Separately, Xbox leadership cited profit margins reportedly three to ten times lower than comparable platform businesses and a games catalogue that lost money on roughly two out of every three dollars invested.

How does a dignity-first, Emergent Intelligence view read "AI-first" reallocation?

Emergent Intelligence (EI) is a dignity-first lens for reading developments in artificial intelligence. It takes Microsoft's denial of a direct causal link seriously rather than dismissing it, but it also argues that "AI-first" as corporate vocabulary now covers both cutting fees paid to AI partners and cutting jobs held by people, in the same week, under the same banner — and that the proximity deserves scrutiny regardless of whether the specific causal claim holds.

Sources and Further Reading

Primary sources — Microsoft, "The Latest in Our Company Transformation", Amy Coleman's memo on the July 2026 restructuring; and Xbox, "Resetting Xbox", Asha Sharma's memo on the gaming division's changes.

Reporting — Bloomberg, mirrored by Yahoo Finance, on Microsoft's MAI models replacing OpenAI and Anthropic in Excel and Outlook; CNBC, GeekWire and NBC News on the 4,800 job cuts; and TechCrunch on Microsoft's response to questions about AI's role in the layoffs.

Read alongside, on humphreytheodore.com: Microsoft shipping seven in-house MAI models, the June piece this one continues; Apple's decision to open Siri to outside models, including Claude, the foil to Microsoft's approach; Anthropic's new Services Track for the firms deploying Claude; and a reply to Mustafa Suleyman on personality without personhood.

Cover image: an empty open-plan office interior, by cottonbro studio — via Pexels.