Microsoft's fiscal 2026 results showed Azure revenue growth of 39%, a figure CFO Amy Hood attributed in part to GPU capacity constraints — telling investors growth would have exceeded 40% had the newest generation of GPUs been available in sufficient quantity. The disclosure came even as Microsoft is spending at a record pace on AI infrastructure, with capital expenditure reported around $190 billion for the year.
The capacity crunch isn't unique to Microsoft. Reporting from April 2026 described AI startups facing price increases as high as 32% and wait times stretching to a year for cloud GPU access, with providers including Microsoft and Amazon accused of prioritizing internal compute needs over third-party customer demand.
Why it matters
Azure customers building AI products are directly exposed to Microsoft's internal capacity allocation decisions in a way that customers running ordinary compute or storage workloads are not: if Microsoft's own Copilot and OpenAI-partnership workloads are competing for the same scarce GPU inventory as paying Azure customers, that allocation choice shapes who gets fast access to the newest hardware and who waits.
The stock market reaction underscores how closely investors are watching this trade-off: Microsoft's stock was reported down more than 24% over the trailing year as of mid-2026, a steeper decline than peers in the so-called Magnificent Seven group of large tech companies, despite record AI spending.
How it works
Pricing data compiled in September 2026 illustrates just how much cloud GPU costs vary by provider even for identical hardware: the same H100 GPU node was priced at $55.04 per hour on AWS, $88.49 on Google Cloud, and $98.32 per hour on Azure, according to a FinOps cost-governance report — a gap of roughly 1.79x between the cheapest and most expensive provider for comparable hardware. A separate market overview from Cast.ai similarly tracked 2026 GPU cloud pricing trends across major providers, noting prices moving with AI demand across regions and instance types.
Those price gaps reflect not just raw hardware costs but each provider's own internal demand pressure — a cloud provider simultaneously running large internal AI workloads on the same GPU fleet it sells to customers has less spare capacity to offer externally, which shows up as both higher prices and longer provisioning queues for customers.
The competing read
Microsoft's public position is that AI infrastructure investment, including the roughly $190 billion 2026 capex figure, is building the capacity that will eventually resolve the shortage, and that near-term Azure growth being capped by supply rather than demand is actually evidence of how strong the underlying business is. Skeptics point to the stock's decline and to third-party accounts of startups facing higher prices and long waits as signs that Microsoft's balancing act between fueling its own AI products and serving external cloud customers is currently tilted toward the former, at a real cost to the latter.
What happens next
How quickly the capacity picture eases depends largely on the pace of new GPU generations reaching data centers and how much of that new supply Microsoft continues to route internally versus offering to Azure customers. Until that balance shifts, cloud customers building AI products should expect continued volatility in provisioning timelines and per-hour GPU pricing, and comparison shopping across providers is likely to remain worthwhile given the near-doubling in per-hour costs already observed between providers for equivalent hardware.