AWS Says Large Compute Capacity for 2026 Is Fully Booked as Power and Data-Center Constraints Persist
DataNews Editorial Team
AWS chief executive Andy Jassy said the company’s large compute capacity for 2026 is fully booked, with most 2027 capacity and a significant share of 2028 capacity already carrying customer commitments. Amazon is planning approximately $220 billion of 2026 capital expenditure across data centers, servers and networking equipment.
- AWS’s large compute capacity for 2026 is reported to be fully booked.
- Most 2027 large compute capacity is reported to be sold, while 2028 capacity already has significant customer commitments.
- Amazon plans about $220 billion in 2026 CAPEX spanning data centers, servers and network equipment.
- AWS identifies data-center resources, power and memory as continuing constraints.
- Reservations for future years should not be interpreted as commissioned capacity.
AWS is facing demand from AI laboratories and its conventional cloud business that is outpacing its infrastructure expansion, according to comments attributed to Amazon chief executive Andy Jassy. Jassy said all large AWS compute capacity for 2026 has been booked.
The source further reported that most large compute capacity for 2027 has already been sold, while a significant share of 2028 capacity has customer commitments. These are reservation and commitment indicators, rather than evidence of capacity already in service. AWS also intends to retain some capacity for on-demand computing and its Bedrock and EC2 services.
Jassy identified data-center resources, power and memory as constrained, and said he does not expect the shortages to end by 2027. The comments underscore that the supply constraint is not limited to data-center construction: power availability and memory supply are also relevant to the pace at which AI compute can be delivered.
Amazon plans approximately $220 billion of capital expenditure in 2026, including investment in data centers, servers and networking equipment. The figure is a planned company-wide CAPEX allocation, not confirmed spending and not a data-center-only investment figure. While the scale signals continued infrastructure investment, it does not by itself indicate when constrained compute capacity will become available.
For operators, investors and infrastructure suppliers, the development suggests that high-density AI capacity could remain supply-constrained despite major hyperscaler buildout programs. If these constraints persist, sites with secured power, grid interconnection and equipment supply could hold greater strategic value.
Jassy also said Trainium offers roughly 30–40% better price performance than comparable processors. Separately, he said 99% of AWS’s 1,000 largest EC2 customers use Graviton processors.
The statements indicate that demand for AI-oriented compute may continue to exceed available supply beyond a short-term cycle. Constraints in data-center availability, power and memory could sustain scarcity of high-density capacity even as hyperscalers expand CAPEX; this could increase the strategic value of pre-secured sites, grid connections and equipment supply chains.
- TechNews 科技新報 / ChineseSource →
This article was prepared by the DataNews editorial team based on the sources listed above.