Microsoft reported a $37B annual AI revenue run rate and a $627B commercial backlog. Azure grew 40% year over year, but the numbers reveal a widening gap between AI demand and physical data center capacity.
Key facts
- Microsoft AI run rate: $37B, up 123% YoY
- Azure revenue growth: 40% YoY
- Commercial RPO backlog: $627B, up 99%
- Delivery window stretched from 6 to 18 months
- Demand exceeds capacity by 3-to-1 in key regions
Microsoft's Intelligent Cloud segment generated $34.7 billion in quarterly revenue, up 30%, with Azure driving most of that gain [According to the source]. The 40% Azure growth reflects demand for GPU-backed workloads, including model training and inference. The AI business alone hit a $37 billion annual revenue run rate, up 123% year over year.
The $627B Backlog: Demand Already Sold, Not Yet Delivered
Commercial remaining performance obligations (RPO) — revenue under contract but not yet delivered — surged 99% to $627 billion. This backlog is a direct proxy for infrastructure that Microsoft has sold but cannot yet deliver because power, cooling, and data center capacity are still under construction.
“What used to be a six-month delivery window has stretched to 18 months or more,” said Steven Dickens, president and analyst at HyperFrame Research [According to the source]. He noted that demand outstrips available capacity by nearly three to one in key Tier-1 regions.
The Bottleneck Shifted From Chips to Power and Cooling
The constraint is no longer solely semiconductor supply. “It’s across the entire stack — power, memory, skills, and data center capacity — not just one vector,” Dickens said. AI clusters push rack density higher, raising power draw per facility and increasing reliance on liquid cooling, each factor extending build cycles compared with traditional cloud deployments [According to the source].
Microsoft did not disclose capital expenditure or expansion timelines in the release. The company's Fairwater AI data center launched ahead of schedule in late April, but the backlog suggests that incremental capacity gains are still dwarfed by demand.
Unique take: Microsoft is now effectively capacity-constrained in AI, not demand-constrained. The $627B backlog means the company has already sold more AI compute than it can physically deliver. This shifts the competitive dynamics: neoclouds like Nebius, which claimed first NVIDIA GB300 cloud access, can fill the gap by offering faster deployment timelines for GPU workloads.
What to watch
Watch Microsoft's next quarterly capex disclosure and any announced data center expansions beyond Fairwater. The key metric is whether the RPO-to-delivery ratio improves or widens further — a proxy for whether capacity investment is keeping pace with AI demand.
