Amazon cloud capacity sold out through 2027, CEO Jassy says

“Sold out” means Amazon has promised most of its planned AWS computing capacity to customers through 2027. It does not mean every server is permanently unavailable. Jassy said Amazon keeps some capacity for on-demand use and for services including Bedrock and EC2.

For example, an AI lab or large business may reserve computing capacity years ahead for model development or applications. That commitment gives the customer access to expected capacity, while reducing the amount Amazon can offer to new buyers. The key issue is supply against demand, not a retail-style empty shelf.

Jassy said 2026 capacity is already sold out, 2027 is largely sold out, and a good chunk of 2028 is sold out. The situation shows how strongly AI infrastructure demand is pressing on AWS. It also explains why Jassy said Amazon could invest much more in capacity without it being a bad use of capital.

Amazon said earlier this year that it expects to spend about $220 billion on capital expenditure in 2026. That figure represents a huge planned investment, but Jassy indicated demand could justify spending even more on cloud capacity.

The timing of commitments is equally striking. Jassy said Amazon’s cloud-computing capacity for 2026 is already sold out. He described 2027 as largely sold out, apart from some capacity retained for on-demand use and services such as Bedrock and EC2. He also said a good chunk of 2028 is sold out.

These figures show that Amazon is planning infrastructure years ahead of customer use. They also indicate that investment does not immediately remove shortages. New capacity must be built and matched with demand, while major customers are already making long-term commitments. Amazon is scheduled to report third-quarter results on October 29.

The main users are AI labs and enterprises, according to Andy Jassy. They need large amounts of cloud infrastructure as demand for AI systems grows. Their commitments help explain why AWS capacity is being reserved years in advance.

EC2 is an AWS service that provides cloud servers for running applications. Bedrock gives customers access to AI models so they can build generative AI applications. A business might use EC2 to run its software and Bedrock to add AI capabilities, although the article does not describe any specific customer.

The key mechanism is reservation of shared cloud capacity. When large customers commit to substantial amounts, Amazon has less uncommitted capacity available for new or short-term buyers. Jassy said Amazon is struggling to keep pace with “insatiable demand” from AI labs and enterprises. The capacity pressure therefore affects both specialist AI companies and broader business users.

The article gives a direct explanation: AI labs and enterprises have created insatiable demand for AI infrastructure. These organizations are seeking enough computing capacity to develop and operate AI systems, and their needs are growing faster than Amazon can expand supply.

The mechanism is a race between demand and construction. Customers make large, advance commitments, while Amazon must add cloud capacity through capital investment. Jassy said even a much larger investment could be a good use of capital because 2026 capacity is already sold out. That statement shows how quickly demand is absorbing planned supply.

The article does not identify particular AI projects, workloads, or technical bottlenecks behind the growth. More broadly, AI development commonly requires substantial computing resources, but that additional explanation is established background rather than a detail provided here. The immediate result described is AWS struggling to keep pace with customers’ requirements.

When computing capacity is committed years ahead, Amazon gains clearer visibility into future demand and revenue. However, less capacity remains available for customers that arrive later or need sudden increases. This creates a tighter market for cloud resources.

For example, an AI company without an advance reservation might have to delay a project, seek capacity elsewhere, or use a smaller deployment. A general business could face similar constraints if it needs more servers quickly. These are possible consequences of limited supply, not outcomes specifically reported in the article.

Amazon may respond by investing more, which Jassy said could be a sensible use of capital. AI companies may secure capacity earlier or diversify their infrastructure. Other cloud customers may need more planning. The article establishes the underlying pressure: 2026 capacity is sold out, 2027 is largely sold out, and a good chunk of 2028 is committed.

If AWS capacity is unavailable, customers can generally compare other public-cloud providers, lease equipment from specialized hosting companies, or build capacity in private data centers. They can also combine several providers instead of relying on one platform. The article itself does not name any alternative.

A customer could also reduce its computing requirement. It might select a smaller AI model, use a model designed for a narrower task, or run some workloads on existing equipment. These approaches may lower the amount of scarce capacity needed, though they can involve trade-offs in performance, cost, or flexibility.

The immediate AWS situation makes such planning more relevant. Amazon says 2026 capacity is sold out, 2027 is largely sold out, and a good chunk of 2028 is committed. Still, the article does not compare providers or say that customers are moving elsewhere. These alternatives are established industry options, not reported developments in this article.

In general, cloud capacity is built from data centers containing servers, processors, storage, networking equipment, cooling systems, and electricity infrastructure. The article specifically discusses cloud-computing capacity but does not list its physical components. Amazon’s AWS services turn that infrastructure into resources customers can use remotely.

AI applications commonly require many powerful chips and servers for tasks such as training models, serving responses, and handling large user demand. Those machines consume electricity and need space, cooling, and fast connections. This explains why adding capacity is a major infrastructure project rather than a simple software change. These details are established background, not specifics supplied by the article.

The article’s reported reality is that demand from AI labs and enterprises is outpacing Amazon’s ability to keep up. Jassy said 2026 capacity is sold out and 2027 is largely sold out. Amazon’s planned capital spending and possible additional investment are responses to that capacity pressure.