Andy Jassy says AWS capacity for 2027 is "largely sold out" and a "good chunk" of 2028 capacity is also sold, though AWS reserves some capacity for on-demand (Catherine Perloff/The Information)

AWS capacity is the amount of computing power, storage, networking, and related infrastructure Amazon Web Services can make available to customers. It is the supply behind cloud services, including servers and the facilities that run them. Capacity matters because customers need it to operate applications and workloads.

When capacity is “largely sold out,” most planned capacity has already been promised to customers. The phrase does not mean every AWS service or every location is unavailable. It means AWS has little uncommitted capacity remaining in the affected future period. AWS also keeps some capacity for on-demand use.

Andy Jassy said AWS capacity for 2027 is “largely sold out.” He also said a “good chunk” of 2028 capacity is already sold. That points to unusually strong advance demand and suggests customers seeking large amounts of future computing power may face tighter availability or need to plan earlier.

The article gives a clear qualitative picture, not an exact percentage. Andy Jassy said AWS capacity for 2027 is “largely sold out.” That means most of the capacity AWS expects to have available in that year has already been committed. It does not provide a precise number of servers, data centers, or megawatts.

For 2028, Jassy said a “good chunk” of capacity is also sold. This indicates a substantial advance commitment, but the phrase does not define how large the share is. It could not be converted into a reliable percentage from the information provided. The article gives no exact figure for either year.

The key message is timing. Customers are reserving capacity years ahead, and AWS is already seeing strong commitments beyond 2027. AWS still reserves some capacity for on-demand use, so “sold out” is not an absolute claim about every available resource.

Cloud capacity is commonly bought in advance by organizations that expect sustained or rapidly growing computing needs. These can include large businesses, technology companies, and artificial-intelligence developers. The supplied article does not identify specific buyers, so those examples provide general context rather than article-specific facts.

What they reserve is future access to computing resources. That can include processing power, storage, networking, and the infrastructure needed to run workloads. In practice, a customer commits to using capacity during a future period, while AWS plans its facilities and equipment around that expected demand. This is different from purchasing and operating the hardware itself.

Jassy’s comments show that advance commitments now reach 2027 and 2028. The article does not say which customers made them or how much each reserved. It does show that AWS has sold most 2027 capacity and a substantial portion of 2028 capacity, while keeping some supply for on-demand use.

Reserved capacity is computing supply a customer commits to using in advance. This gives the customer planned access during a future period and gives AWS clearer demand for its infrastructure. On-demand capacity is capacity AWS keeps available for customers that need resources without making the same kind of long-term commitment.

For example, a customer expecting a major workload in 2027 could arrange access years ahead. Another customer might need servers suddenly and use whatever AWS has left for on-demand service. The central mechanism is allocation: advance commitments claim part of future supply, while AWS holds another portion back for more immediate or flexible requests.

The article says AWS capacity for 2027 is “largely sold out,” and a “good chunk” of 2028 capacity is sold. Yet AWS reserves some capacity for on-demand use. Therefore, advance sales do not mean every AWS resource is unavailable; they mean committed and uncommitted supply are being managed separately.

When most future capacity is committed, customers seeking additional computing power face a tighter supply of unreserved resources. They may have fewer choices about timing, location, or the amount of capacity they can obtain. The article does not describe a specific customer impact, but this is the practical meaning of capacity being largely sold out.

A customer that cannot secure a long-term commitment may try AWS’s remaining on-demand capacity. That supply is the portion AWS keeps aside for flexible demand. If the needed resources are unavailable, the customer may need to adjust timing, reduce the immediate requirement, or seek capacity elsewhere. These possible responses are general cloud-market implications, not details stated in the article.

Jassy’s comments show why planning matters. Most 2027 capacity is already sold, and a good chunk of 2028 capacity is committed. AWS retains some on-demand capacity, but the available buffer is separate from the capacity already promised to customers.

Artificial-intelligence workloads can create intense demand for cloud capacity because training and running advanced models require large amounts of computing power. They may also need substantial storage and high-speed connections between machines. This explanation uses established industry knowledge; the supplied article itself does not mention AI workloads or explain the cause of demand.

For example, an AI developer might need many specialized processors operating together to train a model. Later, serving that model to users can require more computing capacity. Cloud providers build and reserve infrastructure to support these workloads. The mechanism is simple: larger or more numerous workloads consume more capacity, leaving less uncommitted supply.

The article’s confirmed point is broader. Andy Jassy said AWS capacity for 2027 is “largely sold out,” while a “good chunk” of 2028 capacity is also sold. The source does not say how much of that demand comes from AI, so no AI-specific share can be stated.

Creating cloud-computing capacity requires physical infrastructure. Data centers provide secure places for equipment. Chips perform calculations. Electricity powers the machines, while cooling systems remove the heat they produce. Networks connect servers to one another and to customers. Storage systems and buildings are also part of the broader infrastructure.

The key mechanism is coordination. A data center cannot offer useful capacity if it lacks enough chips, power, cooling, or network connections. A shortage in any one area can limit the amount of computing AWS can provide. Building more capacity therefore requires planning and investment across several linked resources, not just adding servers.

The article does not list these physical components or explain AWS’s construction process. It does report that AWS capacity for 2027 is “largely sold out” and that a “good chunk” of 2028 capacity is sold. Those commitments concern future cloud supply, whose creation depends on this underlying infrastructure.