Amazon Web Services and Nvidia have announced plans to deploy across AWS data centers in 2027 and 2028. That comes just five months after Amazon committed to deploying more than starting in 2026, effectively tripling its Nvidia GPU commitment in a matter of months.
The new chips will include Nvidia's Blackwell Ultra, Rubin, and Rubin Ultra GPUs, designed to power increasingly demanding AI workloads, from training frontier models to running AI agents and enterprise applications.
And Amazon says the reason is simple: demand has exceeded expectations.
According to Nvidia and AWS, startups, enterprises, AI labs, and governments are all scaling their AI workloads faster than expected. Customers are moving beyond AI experiments and into production, creating a huge need for more compute capacity.
This is bigger than Amazon simply buying more chips.
It is another signal that the AI infrastructure boom isn't slowing down.
Nvidia's latest numbers tell the same story. The chipmaker reported $96.2 billion in quarterly revenue, while its data-center business generated roughly $89 billion, up 117% year over year. Nvidia also expects revenue to reach $108 billion next quarter.
Amazon's decision suggests AWS expects its customers to keep consuming enormous amounts of AI compute well into the next several years.
And that matters because AWS is simultaneously developing its own AI chips.
Amazon has invested heavily in its custom Trainium chips as a way to reduce its reliance on Nvidia and offer customers cheaper alternatives for certain AI workloads.
Trainium is already being used by major AI labs, including Anthropic, while Amazon says its broader custom-chip business has passed a $20 billion annual revenue run rate.
Yet Amazon is still dramatically increasing its Nvidia footprint.
That tells you something about Nvidia's position in the AI hardware market.
Even the companies trying to build alternatives still need Nvidia's GPUs for some of their most demanding workloads.
The new partnership also goes beyond GPUs.
AWS and Nvidia plan to integrate Nvidia Vera CPUs, advanced networking, open models, data-processing software, and physical AI technologies into AWS infrastructure. Nvidia's technology will also be integrated with AWS's Nitro System and Elastic Fabric Adapter.
Amazon is essentially building AWS into an even larger AI infrastructure platform—one designed not just for today's generative AI models, but for the next wave of AI agents, robotics, scientific computing, and physical AI.
There is still a huge question hanging over all of this spending:
Will customers generate enough revenue from AI to justify the infrastructure being built today?
Amazon, Nvidia, Microsoft, Google, Meta, and other major players are collectively spending enormous amounts on data centers and AI chips.
If AI adoption continues accelerating, these investments could look incredibly smart.
But if AI spending eventually slows before customers generate sufficient returns, the industry could find itself with an enormous amount of expensive computing capacity.
For now, though, the evidence is pointing in the opposite direction.
Amazon didn't triple its Nvidia GPU commitment because it expects the AI boom to disappear.
It did it because AWS says demand is growing faster than expected.
And with 2 million more Nvidia GPUs scheduled to enter AWS data centers, Amazon is betting that the next phase of AI won't be about experimenting with the technology.
It will be about running AI at massive scale.
