Amazon triples GPU order from Nvidia as AI infrastructure race intensifies

Amazon and Nvidia are deepening their partnership with a major expansion announced during Nvidia's earnings call Wednesday. The deal commits AWS to deploy an additional 2 million Nvidia GPUs—including Blackwell Ultra, Rubin, and Rubin Ultra processors—across its data centers in 2027 and 2028.

The move comes just five months after Amazon agreed to roll out more than 1 million Nvidia GPUs starting this year. Nvidia's statement noted that customer demand has "exceeded expectations" since that earlier commitment, prompting the accelerated expansion.

A multibillion-dollar commitment#

While the companies did not disclose exact pricing, the deal is estimated to be worth tens of billions of dollars based on typical GPU costs. Neither organization revealed specific financial terms. Beyond chip purchases, the partnership extends to Nvidia's networking infrastructure that stitches together thousands of GPUs into unified systems, as well as software, CPUs, open models, and robotics platforms.

Nvidia says the decision to deepen collaboration stems from "surging demand" across startups, enterprises, research labs, and government organizations seeking AI computational power.

A paradox: partnership amid internal competition#

The timing raises an interesting dynamic: Amazon is simultaneously investing heavily in its own AI chips designed to reduce reliance on Nvidia and eventually compete directly. The company's custom Trainium processors provide an alternative to Nvidia's flagship offerings for deep learning workloads, while its Graviton CPUs—built on Arm architecture—challenge traditional server processors from Intel and AMD.

Amazon has positioned its custom chip business as a major growth driver, recently announcing it crossed a $25 billion annualized revenue run rate. This figure reflects over $225 billion in total commitments from AI labs including Anthropic and OpenAI. Yet despite these internal alternatives, Amazon continues to expand its Nvidia footprint significantly.

Vera CPUs and robotics integration#

The expanded deal also includes an unspecified volume of Nvidia's Vera CPUs, set to ship beginning in Q3. These chips represent part of Nvidia's strategy to capture what CEO Jensen Huang described as a "brand new $200 billion market opportunity" announced in May. Huang expects Vera deployment across every major cloud hyperscaler, with shipments already underway to partners including Oracle and SpaceXAI.