Mat Velloso spent the weekend on X telling anyone who would listen that Microsoft missed the AI wave. He has some standing here. Before becoming VP of product for the developer platform at Meta's Superintelligence Labs, he led AI developer products at Google DeepMind. And before that, twelve years at Microsoft, four of them as technical advisor to Satya Nadella himself.
That last detail is what makes the whole thread land harder than the usual ex-employee venting.
The 3.3% number
Microsoft has roughly 450 million paid Microsoft 365 commercial seats. On the FY2026 Q2 earnings call in late January, the company said it had converted 15 million of them to paid Copilot seats. That works out to a 3.3% paid adoption rate, and it is not the framing Microsoft would prefer. Against an estimated $37.5 billion in quarterly AI capex, it is a number that has analysts and shareholders asking what the conversion curve actually looks like from here.
The pattern shows up at the device layer too. Microsoft spent the last year leaning on OEMs to ship laptops with neural processing units inside, betting that on-device AI features would create a reason to buy a new PC. As Velloso put it, "nobody cares because not a single valuable usecase was built for those in Windows/Office." Harsh, but the consumer Copilot numbers don't push back hard. Sensor Tower data cited by CNBC earlier this year put Copilot's mobile daily active users at 6 million in February, against 440 million for ChatGPT and 82 million for Gemini.
And then there's Bing, which after years of being Microsoft's biggest AI bet still hasn't taken a percentage point of search share from Google. Velloso's point isn't that any single product is failing. It's that none of them are catching.
"Factory reset"
His framing for what's happening inside Redmond is factory reset, and he posted a list to back it up: Julia Liuson out at the developer division after 34 years, plus reassignments and departures across Xbox, GitHub, AI infrastructure, Teams, and OneNote. Liuson's resignation was reported in April, with Microsoft framing it as a planned retirement and Liuson staying on as an advisor through June. In March, CNBC reported that Microsoft consolidated its consumer and commercial Copilot engineering groups under new EVP Jacob Andreou, freeing Mustafa Suleyman to focus on building new models. Whether that counts as a strategic refocus or a quiet admission of structural problems is, I suppose, in the eye of whoever's holding the stock.
Frank X. Shaw, Microsoft's communications lead, pushed back publicly, accusing Velloso of stapling a "negative frame" onto what he says are normal corporate retirements. Velloso responded by pointing at the stock chart. Since he left Microsoft for Google in early 2024, he claims, Google's stock is up roughly 230% and Microsoft's is essentially flat. I haven't independently verified the exact percentages and the time window matters, but the kind of comparison that doesn't need to be precise to sting.
OpenAI is moving on Microsoft's turf
The most structurally interesting piece of Velloso's argument has nothing to do with leadership churn. Last week, OpenAI launched the OpenAI Deployment Company, a $4 billion enterprise consulting arm backed by TPG, Bain Capital, Brookfield, Goldman Sachs, SoftBank, McKinsey, and Capgemini. The unit absorbs roughly 150 forward-deployed engineers via OpenAI's pending acquisition of Tomoro, and it is explicitly aimed at moving Fortune 500 customers from pilot to production without going through Microsoft.
This is the part Microsoft should worry about. The company didn't win enterprise software through brilliant products. It won by having armies of people who would sit with CIOs and understand their stack. If OpenAI is standing up its own version of that motion, with the consulting firms as co-investors rather than partners, the Azure middleman role gets thinner. Anthropic is doing some version of the same thing, hiring its own deployment engineers under similar logic. Microsoft's response, so far, is to embed Claude inside Copilot via the Frontier program and hope the bundling holds.
The moat argument
Velloso, for what it is worth, doesn't think Microsoft is dying. He spent part of the same thread arguing the opposite. "The last mile is the hardest and Microsoft has the best distribution for that. Their moat is unbreakable." Replacing Active Directory, Exchange, Azure AD, and the rest of the enterprise plumbing at a Fortune 500 is a project measured in years, not quarters, and no AI lab is currently equipped to do it.
So both things can be true. Microsoft hasn't yet built a Copilot experience that 450 million seat-holders are reaching for. And Microsoft is still the company every CIO actually has to deal with. The factory-reset critique only matters if you assume the second sentence stops being true.
What to watch
Two things. First, Microsoft's June quarter, where commercial CEO Judson Althoff has told employees the company set ambitious targets for paid Copilot seats after UBS analysts publicly noted they expected more subscribers. Second, how many of DeployCo's engagements end up touching Microsoft infrastructure at all. If the answer is most of them, because Azure runs the underlying compute, the moat holds. If it's increasingly few, the picture changes fast.
