Amazon and Nvidia have expanded their partnership, with Amazon ordering an additional two million Nvidia GPUs for AWS data centres in 2027 and 2028 after demand for AI computing capacity exceeded earlier expectations. The move strengthens Nvidia’s position in cloud AI infrastructure despite Amazon’s efforts to develop competing in-house chips, and signals continuing large-scale demand from startups, companies, AI labs and governments.

The latest order is triple the more than one million Nvidia GPUs Amazon agreed to deploy only five months earlier, and is likely worth tens of billions of dollars, although financial terms were not disclosed. AWS will use Blackwell Ultra, Rubin and Rubin Ultra GPUs, alongside Nvidia networking, software and other technologies; Amazon also plans to adopt Nvidia’s robotics stack for warehouse automation. Amazon’s own custom-chip business has reached a $25 billion annualised revenue run rate, supported by $225 billion in commitments from AI customers including Anthropic and OpenAI.