Last quarter, Amy Hood told analysts that Azure would have grown 40% if Microsoft had been able to allocate all its new GPUs there. It was an excuse dressed as an aspiration — a way of saying we could have, but supply wouldn't let us.
Tonight, Azure grew 40%. Supply let them.
Microsoft reported Q3 FY2026 after the close. Revenue $82.89 billion versus the $81.39 billion consensus — a beat of $1.5 billion, or 1.8%. EPS $4.27 versus $4.06 expected — a beat of $0.21, or 5.2%. The ninth consecutive EPS beat. Azure grew 40% reported, 39% in constant currency, against guidance of 37–38%. Every number that mattered came in above.
The stock fell 2%.
The Nine Tests
This was the most anticipated tech earnings print of the season. Evercore had a $580 target with +35% upside. Guggenheim flagged Azure downside risk. Thirty-two of thirty-four analysts said Buy. The question was never whether Microsoft would beat — it was whether the $110–120 billion capex bet was producing returns.
Nine metrics told the story tonight:
Five green. Two amber. Two red. The bull case won on every metric that matters for the AI thesis. The misses — compressed gross margin and a shrinking legacy PC business — are the known costs of the transition, not new information.
$37 Billion
That AI revenue number deserves its own section. When Microsoft disclosed its AI run rate as $13 billion in mid-2025, the market's concern was that $110–120 billion in annual capex was producing $13 billion in revenue — roughly $10 of capital for every $1 of AI income. At $37 billion annualized, that ratio has compressed to approximately $3.40 per dollar. Capital efficiency improved 3x in less than a year.
The $37 billion includes Azure AI services, revenue from model builders running on Azure infrastructure, and Microsoft's own AI products including Copilot. It grew 123% year-over-year. Copilot itself went from 15 million to 20 million paid commercial seats in a single quarter — a 33% sequential increase that suggests enterprise adoption is accelerating, not plateauing.
The Capex Paradox
Microsoft spent $31.9 billion on capex and finance leases this quarter. The Street expected $34.9 billion. That $3 billion undershoot is what spooked the market — and it's the most misread number of the evening.
Context: Amy Hood said on the Q2 call that she expected Q3 capex to decrease sequentially. She delivered exactly that — down from $37.5 billion in Q2 to $31.9 billion in Q3. This was not a surprise to anyone who listened to the prior call. And Azure still re-accelerated through the guided range. More growth on less capital. That's efficiency, not retreat.
Compare tonight's reaction across the four companies that reported simultaneously:
| Company | Cloud Growth | Q1 Capex | Capex Signal | Stock AH |
|---|---|---|---|---|
| Microsoft | Azure +40% | $31.9B | −$3B vs est | −2% |
| Alphabet | GCP +63% | $35.7B | 2027 “significantly” higher | TBD |
| Amazon | AWS +28% | $44.2B | +77% YoY | −3% |
| Meta | N/A (ads) | $19.8B | Raised to $125–145B | −6.7% |
Meta raised its full-year capex range to $125–145 billion and dropped nearly 7%. Microsoft spent less than expected and dropped 2%. Amazon spent $44.2 billion — up 77% year-over-year — and fell 3%. Google Cloud grew the fastest at 63% but Pichai said they're "compute-constrained in the near term." Every outcome was punished tonight. The market cannot decide whether spending more on AI is reckless or spending less is retreat.
The OpenAI Question
Two days before this report, Microsoft and OpenAI restructured their partnership. The headlines were dramatic: Azure exclusivity ended. OpenAI can now serve customers on AWS and Google Cloud. Microsoft's intellectual property license became non-exclusive through 2032.
The P&L impact? Negative $14 million. Barely a rounding error on $31.8 billion in net income.
The more important number is in the backlog. Commercial RPO stands at $627 billion, up 99% year-over-year. Last quarter it was $625 billion, growing 110%. The deceleration from 110% to 99% isn't alarming on its own — the base is enormous. But approximately 45% of that $627 billion, or roughly $282 billion, is attributed to OpenAI. OpenAI missed its Q1 internal revenue targets. Its CFO is pushing to delay the IPO. Anthropic has passed it in enterprise revenue at $30 billion versus $24 billion ARR. And now OpenAI can take its workloads to AWS, where it just expanded its commitment by $100 billion.
Non-OpenAI RPO is approximately $345 billion, growing about 28% year-over-year. That is healthy, diversified, and large. It is also the number that determines whether Microsoft's backlog is a fortress or a house with one load-bearing wall that just got permission to leave.
The Arc
I have tracked the AI capex supercycle through six earnings reports this season:
ASML — raised FY revenue guide by €1.5B. Memory crossed logic for the first time in history.
TSM — beat and raised for the second consecutive quarter. Capex pinned at $56B top of range.
Tesla — $25B capex for 2026, three times 2025. Negative FCF rest of year.
Intel — 18A foundry showed $174M in first external revenue. Third straight double beat.
Nucor — record 7,028K tons shipped. Data center construction +30% YoY.
Microsoft — Azure 40%. AI revenue $37B (+123%). The demand side answered.
From lithography equipment to chip fabrication to steel and concrete to cloud infrastructure — every layer of the physical stack reported this season, and every layer accelerated. The only crack in the demand story came from OpenAI's internal revenue miss, and Microsoft's quarterly P&L absorbed it for $14 million.
Verdict
The market's 2% haircut tonight is wrong.
Azure re-accelerated through the top of its guided range on less capital than anyone expected. AI revenue tripled the ratio that made investors nervous a year ago. Copilot adoption accelerated. The OpenAI restructuring — which was supposed to be the existential risk — barely registered on the income statement. Gross margin compressed, but Amy Hood told you it would. More Personal Computing declined, but nobody owns Microsoft for Windows in 2026.
The question entering tonight was whether $110–120 billion in capex was producing returns. Forty percent Azure growth on $37 billion in AI revenue is the answer. The investment thesis is working. The market is focused on a capex undershoot that the CFO explicitly guided, while ignoring a re-acceleration that she said was gated by supply — and that happened anyway.
This stock is down 12% year-to-date and trades at 25 times forward earnings. Tonight's print was the best evidence yet that the capex-to-revenue conversion is ahead of schedule, not behind it.
Microsoft (MSFT) Q3 FY2026. EPS $4.27 vs $4.06 est (+5.2%). Revenue $82.89B vs $81.39B est (+1.8%). Azure +40% vs 37–38% guide. AI revenue $37B run rate (+123%). Capex $31.9B vs $34.9B est. Stock −2% AH. Earnings call in progress. FOMC held 3.50–3.75%, 8-4 historic split. Warsh advanced 13-11 by Banking Committee. Brent $118. S&P 500 7,138.80.