Amazon’s chief executive says AWS cloud capacity is largely sold out through 2027 as data centre space, power and memory run short

Tech and AI

Amazon's chief executive says AWS cloud capacity is largely sold out through 2027 as data centre space, power and memory run short

By Staff Writer  |  26 October 2026

Rows of cooling units and backup generators on the roof of a large data centre at dusk, with another hall under construction behind

Andy Jassy says demand from AI developers and businesses has filled 2026, most of 2027 and a good part of 2028, and that Amazon may spend more than the 220 billion dollars it planned for this year.

Amazon Web Services has sold almost all of the cloud computing capacity it expects to have through 2027, according to Amazon's chief executive, Andy Jassy. Speaking in a video podcast released on Friday 23 October, he said 2026 was sold out, 2027 was largely sold out apart from a share kept back for on-demand customers, and a good part of 2028 was already committed.

Jassy said Amazon was struggling to keep pace with demand from AI laboratories and from large businesses. Earlier this year the company said it would spend about 220 billion dollars on capital projects in 2026. He indicated that the figure could rise, saying Amazon could invest considerably more and that it would not be a poor use of capital.

Space, power and memory

Jassy said Amazon must balance the growth in AI computing against its conventional cloud business, which is also expanding, and that both depend on resources that are increasingly hard to secure: electricity, data centre space and memory chips.

He said there is only so much data centre space and power available today and only so much memory, and that he does not expect the shortage to end in 2027. AI accelerators need specialised high-bandwidth memory, while the conventional cloud services rely heavily on ordinary DRAM, so the company has to decide where the memory it can buy is put to work.

Jassy said 2026 is sold out, 2027 largely so, and a good part of 2028 is already committed.

The cost of running models

Capacity is one limit on how far AI spreads, and cost is another. Jassy said inference, the work a model does each time it answers a request, needs to become much cheaper before businesses use it everywhere they might.

It's still an order of magnitude too expensive in my opinion for it to be used as pervasively as possible.

Andy Jassy, Chief Executive Officer, Amazon

Amazon is developing its own Trainium AI chips partly to bring that cost down, and Jassy said they give 30 to 40 per cent better price-performance than comparable alternatives. The company reports its third-quarter results on 29 October.

For those who design and build data centres, the largest cloud provider is telling its customers that the limit on its growth for the next two years is the supply of finished halls, connected power and memory, and not demand.