Amazon and NVIDIA have announced an expansion of their partnership. Under the latest agreement, AWS will add 2 million GPUs to its data centers in 2027 and 2028, covering product lines such as Blackwell Ultra, Rubin, and Rubin Ultra. This decision comes just five months after Amazon’s previous large-scale procurement commitment.
Place another order within 5 months
The information was disclosed during NVIDIA's earnings call. NVIDIA stated that Amazon's previously planned deployment scale was no longer sufficient to meet demand, so both parties decided to further increase procurement volumes. Customers driving this demand growth include startups, large enterprises, AI labs, and government agencies.
This partnership is about more than just increasing the number of GPUs. NVIDIA stated that its network interconnect hardware, open models, CPUs, data processing software, and robotics platforms will also be more widely integrated into the AWS ecosystem.
- The number of newly added GPUs is 2 million.
- Delivery is concentrated in 2027 and 2028.
- Amazon has previously committed to deploying over one million GPUs.
AWS's in-house chips continue to advance
In recent years, Amazon has been accelerating its efforts to develop proprietary chips, aiming to reduce its reliance on NVIDIA and establish competition in certain areas. Peter DeSantis, head of AWS AI, previously stated that AWS is in discussions with other companies to sell Trainium chips for deep learning tasks in data centers.
In addition to AI accelerators, Amazon’s Arm-based Graviton CPUs are also continuing to expand, targeting the traditional server chip market. Amazon previously disclosed that its custom chip business has reached an annualized revenue scale of $25 billion, backed by a total of $225 billion in commitments from AI labs such as Anthropic and OpenAI.
However, based on the scale of the latest collaboration, NVIDIA remains AWS’s core supplier for expanding AI computing power. NVIDIA’s Chief Financial Officer, Colette Kress, stated that in addition to GPUs, the company will also supply AWS with a certain number of Vera CPUs, some of which will be integrated and deployed with Rubin.
Partnership expanded to include models and bots
The scope of the partnership is also expanding into enterprise services and warehouse automation. AWS will offer NVIDIA’s Nemotron open models on Amazon Bedrock and SageMaker to provide managed AI services for enterprise customers.
Regarding robotics, Amazon plans to adopt NVIDIA’s full stack of physical AI technologies for its warehouse robotics systems. The related products include Omniverse, Cosmos, Isaac, and Jetson, covering simulation, world models, robotics development, and edge computing.
NVIDIA is simultaneously increasing its production capacity.
NVIDIA also reported on the same day that its second-quarter revenue reached $96.2 billion, exceeding market expectations, with data center revenue at $89 billion, up 117% year-over-year. The company expects third-quarter revenue to reach $108 billion, with part of the growth driven by the new Rubin GPU.
To meet AI infrastructure demands over the coming years, NVIDIA has committed $279 billion to secure supply chains and manufacturing capacity, significantly higher than the $119 billion in the previous quarter. The market’s next focus will be on whether this massive investment in computing power can continue to translate into actual profits from cloud services and AI products.