Amazon AWS cloud capacity sold out through 2027, Andy Jassy says amid AI boom

Amazon CEO Andy Jassy said on Friday, October 23, that Amazon Web Services (AWS) has largely sold out its cloud computing capacity through 2027, with a significant portion of its 2028 capacity already committed. Artificial intelligence companies and enterprises are increasingly bidding for computing resources, leading to an increase in pressure on cloud infrastructure providers, as mentioned in the comments, according to Reuters.

On a video podcast hosted by Andreessen Horowitz co-founder Ben Horowitz and general partner Erik Torenberg, Jassy said that Amazon is considering spending more than its $220 billion in capital investment in 2026 on infrastructure projects. The business is increasingly being demanded to perform AI computing, but a shortage of memory, electricity and data centre capacity is posing challenges to its growth.

Amazon AWS capacity booked as AI demand grows

The company has booked the majority of its capacity for 2027, and a significant amount for 2028, Jassy said. This includes Amazon Bedrock and Amazon Elastic Compute Cloud (EC2), which the company still plans to allocate some resources for on-demand consumption.

Infrastructure is not just about data centres, Jassy said. Power supplies and memory parts are also in tight supply, and he does not foresee the memory crunch coming to an end in 2027.

AWS supplies cloud computing infrastructure to businesses of all sizes and around the world. It's also using its EC2 service to launch virtual servers for running apps, and Bedrock is a service that enables customers to create generative AI apps with varying AI models.

Amazon may increase AI infrastructure spending

For 2026, Amazon was expected to spend approximately $220 billion on capital expenditures, driven by the growth of its infrastructure as demand for AI services continues to increase. Jassy indicated that further investment could be justified, given that the company's cloud capacity is already largely committed.

The investment follows a flurry of announcements from tech firms to establish data centres and acquire hardware needed for training and running AI models. However, there are ongoing problems with limited power supplies, data centre space, and supplies of specialised memory to expand capacity.

AI accelerators utilise high-bandwidth memory (HBM), which is especially crucial for handling complex AI tasks. Dynamic random-access memory (DRAM) is the other type of memory used by traditional cloud services, which exacerbates the shortage of memory.

Amazon targets lower AI computing costs with trainium

Jassy also emphasised the importance of making AI inference more cost-effective, the mechanism by which AI models respond or execute tasks. While the inference cost is still an order of magnitude too high to make AI as ubiquitous as it can be, he said.

Amazon is building its own Trainium AI chips to be more efficient and cost-effective. Jassy said the chips offer a 30% to 40% improvement in price-performance compared with comparable alternatives.

Amazon is expected to release financial results on October 29, 2026, in its third quarter. Investors will be paying close attention to its plans for cloud growth and meeting surging demand, as well as its plans for its AI infrastructure spending.