Amazon: Cloud Capacity Sold Out Through 2027 as AI Demand Surges

Amazon CEO Andy Jassy said that cloud computing capacity at Amazon Web Services is fully sold out for 2026, nearly exhausted for 2027, and a large portion of 2028 is already reserved. Jassy explained on an Andreessen Horowitz podcast that the proliferation of artificial intelligence has generated widespread demand for cloud services.

Amazon Web Services allows companies to use computing and storage resources on a pay-as-you-go basis, and cloud infrastructure is also used to train AI models and run inference. As a result, the company is one of the largest customers for Nvidia chips and computer memory manufacturers in South Korea.

Jassy noted that Amazon's acquisition of semiconductor design firm Annapurna Labs about a decade ago was perhaps the best deal in the company's history, as it played a key role in developing the Trainium AI chip and Graviton server chip. He added that Amazon's investment in Anthropic also aided its in-house chip development efforts.

Amazon is building a vast global network of data centers, including a 100-megawatt AI center in Ulsan, South Korea, in partnership with SK Group, which is expected to begin phased operation in the second half of 2027. The company will announce its quarterly results on October 29.

What Do These Terms Mean?

Cloud Computing: Renting computing power and storage over the internet instead of buying and owning servers and data centers.

Inference: The phase where a trained model is used to answer a request or execute a task, which is the operational phase after training is complete.

Semiconductors: Electronic chips that power computers and data centers, including processors and AI chips.

Data Center: A facility housing thousands of servers, cooling systems, and power infrastructure, used to run cloud services and train models.