Amazon faces AI computing capacity shortages through 2027

Amazon Web Services (AWS) is facing significant computing capacity shortages driven by the surge in artificial intelligence demand. CEO Andy Jassy confirmed that much of the company’s cloud capacity is already sold through 2027, with substantial portions of 2028 also committed to large corporations and AI laboratories.

To meet this demand, Amazon plans to invest approximately $220 billion in capital expenditures for 2026, noting that further investment would not be a misuse of capital. However, infrastructure bottlenecks are emerging; some AWS data centers are currently unable to power newly acquired Nvidia GB300 server racks due to insufficient electrical capacity. This has led to the accumulation of “dark GPUs”—high-value hardware that is stored in warehouses or transit facilities because the necessary power, cooling, or data center space is not yet available.

Industry experts note that while the initial bottleneck for AI was chip supply, the focus is shifting toward power availability, transmission capacity, and data center construction cycles. This scarcity is expected to impact AI startups, potentially leading to higher and more rigid inference costs as they compete for limited access against large-scale enterprise clients.