Amazon Nvidia chips order has tripled, adding 2 million GPUs to the cloud provider’s data centers, according to a joint statement released Wednesday.
Expanded deal doubles the AI hardware pipeline
The agreement calls for the delivery of Blackwell Ultra, Rubin and Rubin Ultra graphics processors to AWS facilities in 2027 and 2028. Those units are built to meet the heavy compute needs of training large language models and other AI workloads.
Five months earlier the retailer had pledged more than 1 million of the same class of chips, a commitment that the chipmaker said “demand has exceeded those expectations.” The new order effectively triples the original quantity.
Financial terms were not disclosed, but analysts estimate the purchase could be worth tens of billions of dollars given current unit pricing.
Related: Top Exec Leaves OpenAI Data Center Team
Beyond the raw hardware, the partnership will also integrate the firm’s networking fabric, open‑source AI models, CPUs, data‑processing software and a robotics platform across the cloud services.
Parallel push on in‑house silicon
While the retailer deepens its tie with the chipmaker, it continues to develop its own silicon. Its AI chief, Peter DeSantis, has indicated that the Trainium accelerator—intended as a direct alternative to the H100 and Blackwell GPUs—may be offered to external data‑center operators.
The Graviton line of CPUs, built on Arm architecture, is positioned as a challenger to traditional server processors from Intel and AMD. The retailer’s custom‑chip division recently reported a $25 billion annualized revenue run rate, driven by $225 billion in commitments from AI labs such as Anthropic and OpenAI.
For customers, the combined effect of the expanded GPU supply and the retailer’s own silicon roadmap could translate into faster model training times and lower cloud‑compute costs, especially for startups that lack large on‑premise clusters.
Related: How Industry Collaboration Is Promoted by FSI Networking Events ?
The chipmaker also plans to ship an unspecified quantity of its Vera CPUs, some paired with Rubin GPUs and others as standalone units, as part of the broader collaboration.
In the most recent quarter the firm recorded $96.2 billion in sales, with data‑center revenue hitting $89 billion—up 117 % from a year earlier.
Investors will be watching to see whether the added compute capacity translates into the expected boost in profitability as AI firms pour massive capital into infrastructure.
