Amazon CEO Jassy: AWS capacity sold out through 2027, with much of 2028 also locked up
Amazon CEO Andy Jassy revealed in an interview on the Andreessen Horowitz podcast that AWS's large-scale compute capacity for 2026 has been fully booked, 2027 is essentially sold out, and a considerable portion of 2028 has also been locked in early by customers. Amazon disclosed earlier this year that its 2026 capital expenditure will reach about $220 billion, mainly directed at data centers, servers and networking equipment, but Jassy said data center, power and memory resources are tight, and the shortage may not end in 2027. He also said that 99% of AWS's top 1,000 EC2 customers already use the in-house Graviton processor, that Trainium chips offer about 30% to 40% better cost-performance than comparable processors, and that AI inference costs still need to fall by roughly ten times to drive enterprise-scale adoption.
Why read this
AI score 70
Jassy laid out the timeline for AWS capacity sellouts and cost-performance data for in-house chips, which can be used to gauge the compute supply-demand gap and the direction of inference cost reduction.
Topics
Body · AI translation
According to Reuters, Amazon CEO Andy Jassy recently revealed during an interview on the Andreessen Horowitz podcast that, driven by continued explosive demand for artificial intelligence (AI) infrastructure, most of Amazon Web Services' (AWS) cloud computing capacity has been sold out through 2027, and a considerable portion of its 2028 capacity has already been locked in by customers in advance.
In the interview, Jassy explicitly stated that all of AWS's large-scale compute capacity for 2026 has been booked. "2027 is essentially sold out. We always keep some capacity for on-demand business and inference services like Bedrock or EC2, but the bulk capacity is basically all sold. A considerable chunk of 2028 has also been sold." He added that demand for compute from AI labs has reached an "unprecedented" scale, and combined with the continued growth of enterprise cloud business, AWS is finding it difficult to keep pace with customer demand even with full-scale expansion.
The scale of this supply-demand gap is evident from Amazon's capital expenditure plans. Amazon disclosed earlier this year that its 2026 capital expenditure will reach approximately $220 billion, mainly directed toward data centers, servers, and networking equipment.
However, Jassy also acknowledged that even such massive investment is still insufficient to meet demand. "The company could absolutely spend more money, and it would not be an unwise use of capital."
Currently, what constrains the expansion of cloud compute is not only funding. Jassy pointed out that data centers, power, and memory resources are all in tight supply, and he does not believe the related shortages will end in 2027; supply constraints for some components may even persist for several years. With limited memory supply, AWS needs to make trade-offs in allocating high-bandwidth memory (HBM) used for AI accelerators versus dynamic random-access memory (DRAM) that supports traditional cloud business.
Facing cost pressure, Amazon has continued its in-house development path in general-purpose CPUs and AI chips. Jassy revealed that 99% of AWS's top 1,000 EC2 customers already use its self-developed Graviton processor, which offers about 40% better price-performance than other x86 processors. Now, this strategy is being replicated in the AI accelerator space. He said the Trainium chip offers about 30% to 40% better price-performance than processors of the same tier.
But Jassy also acknowledged that AI inference costs still have significant room to fall. "In my view, for AI to be used as widely as possible, the current cost is still an order of magnitude too high." He believes inference costs need to drop by about ten times to truly drive large-scale adoption of enterprise AI applications. This judgment forms the core logic behind Amazon's continued bet on self-developed chips: by lowering the cost per unit of compute, it can open up broader enterprise market demand.