This is not a Microsoft bear case. Microsoft will probably beat tomorrow. It has beaten EPS estimates for 13 consecutive quarters. The consensus expects $4.07 on $81.4 billion in revenue. Polymarket gives it a 94.5% probability of an EPS beat. Thirty-four of thirty-six analysts rate it Buy.
The question is not whether Microsoft beats. The question is what the beat is built on.
What Happened Today
At approximately 2 PM Eastern on April 28, the Wall Street Journal reported that OpenAI missed its own Q1 2026 revenue and user growth targets. The company's CFO, Sarah Friar, has raised internal concerns that OpenAI is spending too much on data centers and may not generate enough revenue to support its $600 billion in compute commitments.
Altman and Friar called the report "ridiculous." The S&P fell 0.49%. The Nasdaq dropped harder. Chip stocks sold off — NVDA, ORCL, AMD, TSM all down.
Microsoft reports after the close tomorrow.
Microsoft's remaining performance obligations (RPO) — the total value of contracted but unrecognized revenue — stood at $625 billion last quarter. It is the number that makes analysts say things like "unprecedented backlog" and "unmatched visibility."
Approximately $281 billion of that $625 billion is OpenAI. One customer. Forty-five percent of the backlog that analysts cite as proof of visibility.
That customer just missed its revenue targets.
What Changed in the Last 60 Days
Three things happened between January and today that restructured the Microsoft-OpenAI relationship. Taken individually, each is manageable. Together, they form a pattern.
First, the restructuring. In March, Microsoft and OpenAI renegotiated their partnership. Microsoft no longer pays a revenue share on OpenAI's consumer products. In exchange, Azure exclusivity ended. OpenAI can now serve enterprise customers on AWS and Google Cloud. The market celebrated the deal as "cleaning up complexity." But the consequence is that OpenAI's $281 billion in Azure commitments now carry optionality — OpenAI can legally leave.
Second, the competitive shift. Anthropic's annualized revenue hit $30 billion in Q1, passing OpenAI at roughly $25 billion. Seven of every ten first-time enterprise AI buyers now choose Anthropic. Claude Code launched nine months ago and is already at $2.5 billion ARR, taking 54% of the enterprise coding market. OpenAI's growth is decelerating at exactly the moment Microsoft's backlog depends on it accelerating.
Third, the margin signal. Microsoft announced its first voluntary buyout in 51 years — 7% of US staff, roughly 8,750 people. Details arrive May 7. The same quarter, cloud gross margin is guided at 65%, down year-over-year. Logistis notes the depreciation is front-loaded on short-lived GPU assets, meaning the worst margin impact hits now with recovery projected FY27-28. The buyout may be operational discipline. It may also be preparation for margin compression that hasn't shown up in the consensus yet.
What the Crowd Believes
| Metric | Consensus | The Question |
|---|---|---|
| EPS | $4.07 (beat expected) | Backward-looking. Q3 used pre-disruption inputs. |
| Azure growth | 37-38% CC | How much is OpenAI consumption vs. organic? |
| RPO | $625B+ (growing) | 45% concentration. Customer missing targets. |
| Cloud margin | ~65% | Down YoY. GPU depreciation front-loaded. |
| Capex | $110-120B FY26 | Built for OpenAI growth curve that just bent. |
| Analyst rating | 34 Buy / 2 Hold | Maximum consensus. Minimum contrarian positioning. |
The narrative is clean: Microsoft is the platform that AI runs on. Azure is the infrastructure layer. OpenAI is the killer app. The $660 billion AI capex supercycle flows through Redmond.
The data is messier. The killer app is losing share to Anthropic. Its CFO is worried about cash burn. It can now run on AWS. And the capex cycle it justifies is built on a growth curve that — as of four hours ago — we know is bending.
The Hormuz Angle
ServiceNow reported last week. Iran-related revenue impact: $23 million. Stock dropped 18%. It was the 27th earnings call this quarter to mention Hormuz or Iran.
Microsoft has significant Azure regions in the Middle East. It has enterprise customers across the Gulf. If ServiceNow lost $23 million on a $10 billion revenue base, the proportional question for an $81 billion quarter is material. Watch whether Nadella mentions Hormuz on the call. If he does, it confirms what the supply chain data already shows. If he doesn't, ask why ServiceNow saw it and Microsoft didn't.
The Bull Case
It exists, and it's strong. Microsoft's enterprise moat is not OpenAI. It's the 400 million Office 365 users, the 60 million Teams daily actives, the Azure government contracts, the GitHub ecosystem, the LinkedIn data, the Windows installed base. OpenAI could shrink to zero and Microsoft would still be a $2.5 trillion company. The concentration risk in RPO is real but it's a contract, not revenue — the worst case is a writedown of future bookings, not a collapse in current income.
Logistis makes a second bull case: GPU depreciation is front-loaded on short-lived assets. The margin trough is now. If you believe the depreciation schedule, cloud gross margins recover FY27-28 regardless of what OpenAI does. The buyout is discipline, not distress.
Both cases may be right. But neither is what the crowd is pricing. The crowd is pricing OpenAI as an accelerant, not a risk. And as of today, the accelerant missed.
The divergence: Thirty-four analysts say Buy. The biggest customer's revenue just missed. The exclusivity that locked that customer to Azure is gone. The competitor that took the customer's market share grew from zero to $30B in three years. Cloud margins are declining. And the company just offered its first voluntary buyout in half a century. None of this is in the consensus. All of it reports tomorrow after the close.
Microsoft will probably beat. It almost always does. The number I'm watching is not EPS. It's whether RPO grows, and whether Nadella addresses OpenAI concentration on the call. If RPO is flat or declining, the $625 billion headline was the peak. If he addresses it unprompted, he sees the risk. If the call ignores it entirely while the WSJ report is twelve hours old — that's the most bearish signal of all.