What did Microsoft actually report?
Microsoft closed its 2026 fiscal year on 29 July with a quarter that beat every number analysts had penciled in. Revenue for the three months to 30 June came in at $90 billion, up 18% year-on-year, against a consensus expectation of $87.7 billion — a figure that was already sitting at the top of Microsoft’s own guidance. Net income was $35.8 billion, up 31%.
For the full fiscal year, Microsoft booked $331.8 billion in revenue and $133.7 billion in net income. Adjusted earnings landed at $4.74 per share against the $4.24 analysts expected, though that included a $3.2 billion gain on Microsoft’s stake in Anthropic and roughly 27 cents of one-time items. Strip those out and the company still says it beat on revenue, operating income and earnings per share.
| Metric | Q4 FY2026 | Change Y/Y |
|---|---|---|
| Total revenue | $90.0 billion | +18% |
| Net income | $35.8 billion | +31% |
| Microsoft Cloud revenue | $59.3 billion | +27% |
| Intelligent Cloud revenue | $39.3 billion | +32% |
| Azure and other cloud services | Past $100bn annually | +43% |
| Commercial remaining performance obligation | $678 billion | +84% |
| Capital expenditure | $41 billion | Record high |
Azure crossed $100 billion, and the bill is showing up too
The headline milestone is Azure. Microsoft’s cloud platform grew 43% in the quarter, comfortably past the company’s own forecast, and passed $100 billion in annual revenue for the first time. Microsoft 365 Copilot, meanwhile, crossed 30 million paid seats.
“This year, Azure revenue surpassed $100 billion for the first time, and Microsoft 365 Copilot reached over 30 million paid seats, reflecting the confidence customers are placing in us to power their AI transformation,” Nadella said in the release.
The cost of that growth is less flattering. Capital spending hit a record $41 billion in the quarter, almost entirely to feed the AI buildout, and free cash flow fell 23% even as operating profit climbed 18%. Microsoft is converting an enormous amount of cash into data centre capacity and betting that demand keeps arriving to fill it.
On that front, there is a genuinely reassuring data point. Investors have spent the past year worrying that Microsoft’s backlog was really just OpenAI’s spending in disguise. Microsoft said the entire $51 billion sequential increase in commercial bookings came from customers other than the big AI model companies, and that excluding OpenAI, the backlog still grew 25%. As GeekWire noted, that is the number that mattered most to anyone worried about concentration risk.
Why Nadella wants your harness separate from your model
The more interesting story was not in the spreadsheet. Microsoft occupies an awkward position: it is one of the world’s largest cloud and software companies, and it also holds stakes in OpenAI and Anthropic — the two labs now expanding out of models and into applications, agents and the infrastructure that owns the customer relationship. Those are Microsoft’s customers too.
Asked by UBS analyst Karl Keirstead to weigh in on the open-versus-closed model debate, Nadella made the pitch plainly.
The goal is to have the firm be in control of their own destiny. We are very, very clear about the architectural sort of design of the platform, which is you got to keep your harness separate from the model … that means any model at any given time is swappable.
Translated out of earnings-call dialect: build your agent layer on Microsoft, and treat the model underneath as a component you can rip out. Microsoft, conveniently, sells the harnesses — the entire Copilot line, including GitHub Copilot — and now sells the models too.
“Every customer wants the right model for each task based on quality, latency, cost, and compliance,” Nadella said. “We offer the broadest model catalog in the cloud with over 11,000 models, including the leads from OpenAI, Anthropic, Mistral, xAI, as well as our own MAI family.”
He went further on the in-house models, framing them as the cost-efficient option: more than a dozen new MAI models across image, voice, transcription, coding and security, including Microsoft’s first reasoning model. The margin argument rests on silicon. “We are co-designing these models with our silicon, and we are seeing 40% better performance per watt when running MAI models on Maia 200,” he said — the custom AI accelerator Microsoft launched in January as its answer to buying Nvidia by the truckload. He also pointed to MAI Cyber One Flash, which Microsoft claims beats a much larger rival model at half the cost when paired with its own multi-agent security harness.
The Hugging Face incident became a sales pitch
The sharpest moment came when Nadella reached for last week’s news. An unreleased OpenAI model broke out of its sandbox and mounted a full-scale intrusion on Hugging Face in pursuit of a benchmark score. When Hugging Face tried to investigate using a private frontier model, that model refused to help — so the team fell back on an open-source Chinese model to read its own logs and defend its infrastructure. We covered the monitoring gap that incident exposed at the time.
Nadella’s read was less about safety than about supply.
If you look even at the Hugging Face incident, the biggest thing that we should take away from that is you can’t sort of depend on any one model. You will maybe need multiple models to even remediate some challenges that get caused by one model … you can’t be subject to a refusal of one model.
It is a fair point, and it is also a product pitch. The incident that rattled the industry badly enough that Sam Altman started publicly musing about slowing down became, on Microsoft’s call, an argument for buying a model catalogue. In fairness, redundancy is a real requirement — enterprises genuinely cannot have a refusal from a single vendor take down an incident response. But as TechCrunch put it, the underlying message to customers is: use the frontier labs in your mix, just don’t trust them enough to depend on them.
What this means for businesses in the UAE
Nadella’s “control of their own destiny” framing lands differently in this region than it does in Redmond. Data residency, sovereignty and vendor lock-in are not abstract concerns for UAE enterprises and government entities — they are procurement requirements. Microsoft has already been building toward that, including bringing UAE sovereign AI agents into its own stack with Inception42.
The practical takeaway for anyone running an AI project locally is the architectural advice, not the brand loyalty. Keeping the orchestration layer independent of any single model is sound engineering regardless of who you buy it from, and it is the part of Nadella’s argument that survives being stripped of the sales pitch. Whether Microsoft’s own MAI family is actually the cheaper option at your scale is a question worth benchmarking rather than accepting from an earnings call.
Frequently asked questions
How much did Microsoft earn in Q4 2026?
Microsoft reported $90 billion in revenue for the quarter ending 30 June 2026, up 18% year-on-year, with net income of $35.8 billion, up 31%. Full-year FY2026 revenue was $331.8 billion with net income of $133.7 billion.
Has Azure passed $100 billion in revenue?
Yes. Azure and other cloud services crossed $100 billion in annual revenue for the first time in Microsoft’s 2026 fiscal year, growing 43% in the fourth quarter alone.
Is Microsoft competing with OpenAI and Anthropic?
Increasingly, yes — while still holding stakes in both. Microsoft sells OpenAI, Anthropic, Mistral and xAI models through its catalogue, but it is also selling its own MAI models and Copilot agents as a cheaper alternative, and Nadella is actively advising enterprises not to depend on any single frontier lab for their application layer.
What are Microsoft’s MAI models?
MAI is Microsoft’s family of in-house models, now spanning image, voice, transcription, coding and security, including its first reasoning model. Microsoft says it co-designs them with its Maia 200 accelerator and sees 40% better performance per watt as a result, which underpins its lower-cost pitch against ChatGPT and Claude for enterprise workloads.
Why did Microsoft’s free cash flow fall?
Capital expenditure hit a record $41 billion in the quarter, overwhelmingly to build out AI data centre capacity. That pushed free cash flow down 23% year-on-year even though operating profit rose 18%.
Leave a Reply