Five months ago, Amazon committed to deploying over 1 million Nvidia GPUs across AWS. That apparently wasn’t enough. According to TechCrunch, Amazon has now agreed to add another 2 million Nvidia GPU chips to its data centers, with the expanded order covering Blackwell Ultra, Rubin, and Rubin Ultra chips scheduled for delivery through 2027 and 2028. Nvidia made the announcement during its quarterly earnings call, noting simply that “demand has exceeded those expectations.” That’s an understatement worth tens of billions of dollars.
Neither company disclosed the financial terms, but GPU unit costs put this deal firmly in the tens-of-billions range. And this isn’t just Amazon buying more chips. The expanded partnership also brings Nvidia’s networking hardware, open models, CPUs, data processing software, and robotics platform into AWS infrastructure. That’s a much deeper integration than a standard chip procurement agreement, and it matters because it ties Amazon’s cloud business more tightly to Nvidia’s full technology stack.
What makes this particularly interesting is that Amazon has been actively trying to reduce its dependence on Nvidia. Its Trainium chips are a direct competitor to Nvidia’s H100 and Blackwell for deep learning workloads. Its Graviton CPU challenges Intel and AMD in general-purpose server compute. Amazon’s AI chief Peter DeSantis has even said AWS is in talks to sell Trainium to other companies for use in their own data centers. The custom chip business is not a side project. Amazon reported a $25 billion annualized revenue run rate from its chip operations, backed by $225 billion in total commitments from AI labs including Anthropic and OpenAI.
But Nvidia is still winning. Despite Amazon’s internal chip ambitions, the scale and speed of this expanded deal shows that Nvidia’s position in AI infrastructure is not under serious near-term threat. Nvidia also reported $96.2 billion in Q2 revenue, beating estimates, with data center revenue alone hitting $89 billion, up 117% year over year. It’s guiding for $108 billion in Q3, partly from early Rubin GPU shipments.
The partnership also extends into robotics and enterprise software. Amazon plans to adopt Nvidia’s full physical AI stack for its warehouse robots, including Omniverse, Cosmos, Isaac, and Jetson. On the enterprise side, AWS will offer Nvidia’s Nemotron open models through Amazon Bedrock and SageMaker. That’s a broader surface area than most AWS-vendor relationships.
Nvidia has now committed $279 billion to secure supply and manufacturing capacity, up from $119 billion last quarter. The company is clearly betting that AI compute demand keeps compounding. Jensen Huang put it plainly on the call: “If we had more compute, we could generate more profitable tokens.” The question investors are starting to ask is whether that logic holds as hundreds of billions pour into infrastructure with returns still being proved out. So far, the orders keep coming.
