1 hour ago
22
Microsoft has a GPU problem, and it’s not the kind you’d expect. The company has the chips. It has the customers. What it doesn’t have is enough electricity to plug everything in.
An estimated $80 billion in unfulfilled Azure orders are currently bottlenecked by power grid limitations, according to CEO Satya Nadella. That’s not a software bug or a supply chain hiccup. It’s a physics problem, and it’s forcing one of the world’s most valuable companies to watch demand pile up while processors collect dust in warehouses.
The numbers behind the backlog
Microsoft’s commercial remaining performance obligations, essentially the total value of contracted but not yet delivered services, hit $678 billion by the end of its fiscal year 2026. That figure represents an 84% year-over-year increase.
The bottleneck is straightforward: data centers need enormous amounts of power, and the electrical grid can’t deliver it fast enough. Microsoft added 1 gigawatt of new data center capacity in its most recent quarter alone and opened 31 new data centers in Q4 of its fiscal year. To put 1 GW in perspective, that’s roughly enough electricity to power a mid-sized American city.
The company plans to double its data center capacity within two years. Goldman Sachs projects that global data center power demand will increase by more than 160% by 2030, with some estimates pushing that figure into the 165-175% range. The culprit is AI workloads, which consume vastly more electricity per computation than traditional cloud tasks.
Capital expenditures tell the same story from the spending side. Microsoft invested approximately $116 billion across FY2026, with quarterly outlays generally exceeding $30-40 billion.
OpenAI’s outsized role
A significant chunk of Microsoft’s commercial backlog traces back to its partnership with OpenAI. Industry estimates suggest OpenAI-related commitments comprise somewhere between 30-45% of the $678 billion in remaining performance obligations. But Nadella has been quick to note that sequential growth is increasingly driven by non-OpenAI customers, a signal that enterprise AI adoption is broadening well beyond the ChatGPT ecosystem.
The grid gap becomes an industry problem
Microsoft’s power constraints aren’t unique. They’re just the most visible example of a challenge facing every major cloud provider and AI infrastructure company. The gap between what tech companies want to build and what the electrical grid can support is widening.
Goldman Sachs’ 160%-plus demand growth projection implies that the problem will get worse before it gets better. AI model training runs are growing exponentially in their power requirements, and inference workloads, the electricity consumed every time someone asks an AI a question, scale directly with adoption.
For investors watching the AI infrastructure buildout, the $80 billion backlog number confirms that demand for AI compute is real and growing. But it also signals that revenue recognition will be constrained by physical infrastructure for the foreseeable future, capping how quickly companies like Microsoft can convert demand into earnings.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
English (US) ·