Amazon and Nvidia announced the expansion of their partnership to add another two million GPU chips to Amazon Web Services data centers, with these units scheduled to reach the cloud infrastructure during 2027 and 2028. The plan includes Blackwell Ultra, Rubin, and Rubin Ultra chips designed to train and run artificial intelligence models.
The new agreement comes just five months after Amazon announced its intention to deploy more than one million Nvidia GPU units across AWS infrastructure, beginning this year. Nvidia said demand had exceeded previous expectations, while the two companies did not disclose the contract’s financial value. The article estimates that the deal’s value could reach tens of billions of dollars based on the cost of the GPU units, but this figure is not a value announced by either company.
A Partnership That Goes Beyond Buying Chips
The move is not limited to supplying graphics processing units. Nvidia said that networking technologies connecting thousands of GPU units within a single system, along with its open models, processors, data processing units, software, and robotics platform, would be integrated more broadly into AWS. The two companies attributed the expanded cooperation to what Nvidia described as “rising demand” from startups, enterprises, AI laboratories, and governments.
Nvidia will also supply an unspecified number of Vera processors, some integrated with Rubin and some standalone. Nvidia Chief Financial Officer Colette Kress said these processors are being adopted by major cloud service providers, AI laboratories, and system manufacturers, mentioning Oracle and SpaceXAI among the key partners.
What Does This Mean for Amazon?
At the same time, Amazon continues to develop its own chips to reduce its reliance on Nvidia and compete with it in some workloads. AWS is working on Trainium chips, which offer an alternative to H100 and Blackwell chips for deep-learning tasks, while also developing Arm-based Graviton processors for its servers. Amazon said its custom-chip business had exceeded an annualized revenue run rate of $25 billion, driven by total commitments of $225 billion from AI laboratories such as Anthropic and OpenAI.
In practice, the deal reveals a dual relationship: Amazon is building its own alternatives, but it is still expanding AWS’s reliance on the Nvidia ecosystem to meet rapidly growing demand. The open question is whether increased spending on computing will generate additional profits at the rate expected by AI companies and service providers.
Expansion into Robotics and Enterprise Services
Amazon will also adopt Nvidia’s physical AI suite to operate its warehouse-robot fleet. The suite includes Omniverse for simulation and digital twins, Cosmos for world models, Isaac for robot development, and Jetson for robotic processing and edge AI.
At the enterprise-services level, AWS will provide Nvidia’s open Nemotron models through Amazon Bedrock and SageMaker. Nvidia announced at the same event sales of $96.2 billion in the second quarter, including $89 billion from its data-center business, up 117% year over year, and forecast revenue of $108 billion for the third quarter. Further clarification from the two companies is needed regarding the timing of the new chips’ arrival and the agreement’s final value.