Amazon triples Nvidia chip order amid 'surging demand'
Amazon and Nvidia are deepening their partnership with a deal to add 2 million more Nvidia GPUs to AWS data centers by 2027-2028, a deal worth tens of billions of dollars alongside broader tech integration.

Amazon and Nvidia announced an expanded partnership on Wednesday that will bring 2 million additional Nvidia GPUs — including Blackwell Ultra, Rubin, and Rubin Ultra chips — to Amazon Web Services data centers in 2027 and 2028. The announcement came just five months after Amazon committed to deploying more than 1 million Nvidia GPUs across AWS starting this year, with Nvidia noting that demand had exceeded its earlier expectations.
Neither company disclosed financial terms, but given typical GPU costs, the agreement is estimated to be worth tens of billions of dollars. The deal goes beyond hardware purchases: AWS will also integrate Nvidia's networking technology, open models, CPUs, data-processing software, and robotics platform.
Amazon still building its own chips
Notably, the expanded partnership arrives even as Amazon continues investing in its own chip lines, including Trainium processors for deep learning and Arm-based Graviton CPUs, aiming to reduce reliance on Nvidia. Amazon has said its custom chip business has surpassed a $25 billion annualized revenue run rate, fueled by $225 billion in commitments from AI labs such as Anthropic and OpenAI.
Still, Nvidia remains dominant in the space. Alongside the GPU shipments, Nvidia plans to supply Vera CPUs and support Amazon's warehouse robotics with its full physical AI stack, comprising Omniverse, Cosmos, Isaac, and Jetson.
Strong quarterly results
Nvidia reported second-quarter revenue of $96.2 billion, topping analyst expectations, with data center revenue reaching $89 billion — up 117% year-over-year. The company forecasts $108 billion in third-quarter revenue. Nvidia has also committed $279 billion toward securing supply and manufacturing capacity, a sharp increase from $119 billion the previous quarter.


