Amazon Boosts Nvidia Chip Orders Threefold as AI Demand Rises

The relationship between Amazon and Nvidia has recently grown substantially tighter. On August 26th, the two businesses announced a further collaboration, which includes an agreement to supply Amazon's data centres with an additional 2 million GPU chips from Nvidia. Nvidia Blackwell Ultra, Rubin, and Rubin Ultra GPUs are examples of GPUs that can manage the intensive computing requirements of training and operating AI models.

The chips are scheduled to be sent to the data centres of Amazon Web Services in 2027 and 2028. This statement was made during Nvidia's quarterly earnings call and follows Amazon's agreement to deploy over 1 million Nvidia GPUs across AWS infrastructure starting this year, which was made just five months ago. Since then, demand has surpassed those predictions, according to a statement from Nvidia.

Amazon Investing Heavily on AI

The declaration is noteworthy for more than only its size and growth rate; it also goes beyond Amazon's purchase of additional Nvidia chips. Additionally, this is transpiring concurrently with Amazon's investment in its own AI processors, which may serve as competitors. According to Nvidia, their technology will be integrated across AWS.

This includes their networking hardware, data processing software, open models, CPUs, robotics platform, and data processing software, which can connect thousands of GPUs into one system. Businesses cited "surging demand" from government agencies, AI research facilities, startups, and enterprises as a driving factor in their determination to collaborate more closely.

This new alliance comes as Amazon is stepping up its own development of artificial intelligence chips, namely central processing units (CPUs), the brains of server computers. In an effort to compete with Nvidia and reduce its reliance on the chip titan, Amazon has been manufacturing its own processors. AWS is reportedly in discussions to offer its Trainium chips to other organisations for usage in data centres. These chips are a direct competitor to Nvidia's H100 or Blackwell chips for deep learning applications. This was said by Amazon's AI chief, Peter DeSantis. Intel's and AMD's long-standing dominance in the server processor market is being threatened by Amazon's Graviton CPU, which is manufactured by ARM.

Future Business Strategies of Amazon and Nvidia

Additionally, Amazon's enterprise products and warehouse robots are part of the alliance. The entire physical AI stack from Nvidia will be used by Amazon to power their robot fleet, according to Nvidia CFO Colette Kress. This stack consists of the following platforms: Omniverse, which is used for simulations and digital twins; Cosmos, for world models; Isaac, for robotics development; and Jetson, for computing hardware for robots and edge AI.

Also this week, Nvidia debuted Jetson 2.0, an updated robotics computer aimed at "entry-level edge AI". AWS will provide support for Nvidia's Nemotron open model family on its managed foundation model platform, Bedrock, and its managed cloud service, SageMaker, for enterprise customers. Additionally, Nvidia surpassed analyst expectations for the second quarter, reporting sales of $96.2 billion on 26 August. With $89 billion in sales, increasing 117% year over year, data centre revenue accounted for the bulk of Nvidia's quarterly revenue.