The number, buried in Microsoft’s latest quarterly filing, is now the most cited figure in the debate over whether the AI boom is real: $678 billion in contracted but unrecognized revenue.
Microsoft’s remaining performance obligations, the formal name for its backlog of signed contracts, reached $678 billion in the quarter, up sharply from a year earlier, according to the company’s financial report. The figure represents cloud and AI services that customers have committed to buy but that Microsoft has not yet delivered or billed. Yahoo Finance reported that the stock firmed after the data was published, a sign that investors took the number as evidence that demand for AI services is not a passing fad.
The backlog has become the core evidence in the argument over AI capital spending. For two years, the market has debated whether the hundreds of billions of dollars flowing into data centers, accelerators and energy infrastructure will ever produce returns. Skeptics pointed to the gap between the hype and the revenue on companies’ income statements. Microsoft’s backlog answers that objection directly: enterprise customers have signed contracts, committed dollars and locked in capacity, and that money is already on the books.
The growth of the backlog tells the story. Microsoft’s remaining performance obligations have expanded steadily as the company has converted OpenAI’s capacity needs, Azure commitments and enterprise AI deals into long-term contracts. The current total, $678 billion, is more than double the level of two years ago, and the pace of growth has accelerated as businesses moved from experimenting with AI to signing multi-year agreements for cloud capacity and model access.
Wall Street’s interpretation has shifted accordingly. The narrative around AI spending has moved from anxiety about whether the money would come back to confidence that demand is contracted and durable. Analysts who follow Microsoft said the backlog provides unusual visibility into the next several years of revenue, cushioning the company against a slowdown in new business even if the economy weakens. The stock’s muted but positive reaction to the filing reflected that comfort.
The composition of the backlog matters as much as its size. A large share is tied to the OpenAI relationship, under which Microsoft provides cloud capacity for the lab’s models and, in turn, resells access to those models to its own customers. Those contracts are long and expensive, and they guarantee Microsoft a stream of capacity payments. The rest comes from enterprises that have adopted Azure for general cloud migration, AI workloads and Microsoft 365 bundles, a base that has proven sticky through economic cycles.
The risks are visible to anyone who reads the footnotes. Backlogs are contracts, not revenue, and they can be renegotiated, delayed or canceled. If the economics of AI shift, if model demand disappoints or if customers decide they have overcommitted to cloud capacity, the backlog would shrink before the cash ever arrives. Microsoft’s history is full of contracted commitments that never converted at full value, and the AI versions of those contracts are the largest ever recorded.
There is also the question of what the backlog says about the industry, not just Microsoft. If one company holds $678 billion in committed AI revenue, the market reasons, the total pipeline across the cloud providers must be enormous. That reasoning has supported the valuations of chip makers, data center operators and power companies that feed the AI build-out. A slowdown in the conversion of Microsoft’s backlog would ripple through that entire ecosystem.
The comparison to the previous technology cycle is instructive. In the late 1990s and again in the 2010s, cloud providers booked large backlogs that grew for years before the industry matured. Each cycle produced a correction when the growth assumptions were tested. AI’s backlogs are growing faster than those earlier cycles, which is precisely why the skeptics remain vocal.
The comparisons sharpen the picture. Amazon and Google, Microsoft’s main rivals in cloud, report their own backlogs, and both have posted strong growth, but Microsoft’s total is the largest in the industry. The company’s position as the capacity provider for OpenAI, the lab behind the most widely used AI products, gives it a contracted revenue base that peers cannot match. Analysts also point to the mix: a larger share of the backlog is now tied to AI-specific services, such as model access and inferencing, rather than to traditional infrastructure, which means the figure tracks the AI cycle directly rather than lagging it.
For now, the numbers are on the bulls’ side. Microsoft’s backlog is not a forecast or a hope; it is a stack of signed contracts with named customers, and it keeps growing. The company’s finance chiefs describe the figure as the strongest indicator of demand they track. The market, having spent two years debating whether AI demand is real, now has a single number it can point to, and that number went up again.
