Amazon and Nvidia have significantly deepened their ties. On Wednesday, the two tech giants unveiled an expanded collaboration that includes an agreement to bring an additional 2 million Nvidia GPU chips into Amazon’s data centers. These GPUs—engineered to handle the intense computational workloads required for training and operating AI models—include Nvidia’s Blackwell Ultra, Rubin, and Rubin Ultra lines. The chips are slated to arrive at Amazon Web Services (AWS) facilities in 2027 and 2028.

The announcement, delivered during Nvidia’s quarterly earnings call, comes just five months after Amazon committed to deploying more than 1 million Nvidia GPUs across its AWS infrastructure starting this year. According to Nvidia, demand has already “exceeded those expectations” since that initial agreement.

Neither company disclosed the financial details of the deal, leaving the exact returns for Nvidia unclear. However, given the unit costs of these GPUs, industry observers estimate the partnership could be worth tens of billions of dollars. What makes this announcement particularly striking is not just its sheer scale or the speed at which it materialized, but also that it goes beyond Amazon simply purchasing more Nvidia chips—even as Amazon continues to invest in its own potentially competing AI hardware.

Nvidia also confirmed that its broader technology portfolio will be integrated across AWS. This includes networking hardware that links thousands of GPUs into a unified system, as well as its open models, CPUs, data processing software, and robotics platform. The two companies cited “surging demand” from startups, enterprises, AI labs, and even governments as a key factor driving their closer collaboration.

This expanded partnership unfolds as Amazon ramps up its own AI chip initiatives, particularly around CPUs—the general-purpose processors that serve as the backbone of servers. Amazon has been developing its own chips to reduce reliance on Nvidia and potentially compete with the chipmaking titan. Amazon’s AI chief, Peter DeSantis, has revealed that AWS is in discussions to sell its Trainium chips—a direct alternative to Nvidia’s H100 or Blackwell processors for deep learning tasks—to other companies for data center use. Additionally, Amazon’s Arm-based Graviton CPU is viewed as a contender against traditional server chips from Intel and AMD.

Amazon has reported steady growth in its custom chip business, noting on its latest earnings call that it surpassed a $25 billion annualized revenue run rate, fueled by $225 billion in total commitments from AI labs such as Anthropic and OpenAI. Yet, despite these efforts, Nvidia remains the dominant force in the AI chip arena.

With the 2 million GPU chips heading to AWS starting in the third quarter, Nvidia also plans to deliver an unspecified number of Vera CPUs—some integrated with Rubin, others sold standalone—according to Nvidia CFO Colette Kress. Nvidia CEO Jensen Huang has ambitious plans for the Vera CPU line, boasting in May that he had uncovered a “brand new $200 billion TAM” for the company. Beyond AWS, Kress noted that Nvidia expects Vera to be adopted by “every major hyperscaler, neocloud, AI lab, and system OEM,” with shipments already underway to leading partners, including Oracle and SpaceX AI.

The partnership also extends into Amazon’s warehouse robotics and enterprise offerings. Kress said Amazon plans to deploy Nvidia’s full physical AI stack to power its fleet of robots. This stack includes Omniverse (a simulation and digital twin platform), Cosmos (a world model platform), Isaac (a robotics development platform), and Jetson (computing hardware for robots and edge AI). This week, Nvidia introduced an updated version of Jetson designed as a more accessible robotics computer for “entry-level edge AI.”

On the enterprise front, AWS will host Nvidia’s Nemotron family of open models on Amazon Bedrock, its managed foundation model platform, as well as SageMaker, its managed cloud service.

Financially, Nvidia reported second-quarter sales of $96.2 billion, surpassing analyst expectations. Data center revenue accounted for the bulk of that total at $89 billion, a 117% increase year-over-year. Nvidia projects third-quarter revenue to reach $108 billion, with some of that coming from its next-generation Rubin GPUs, which began production shipments this quarter. Investors have been closely watching Rubin’s initial Q3 sales as a signal of sustained demand for Nvidia’s upcoming hardware.

To secure supply and manufacturing capacity for current and future data-center projects, Nvidia has committed $279 billion—up sharply from $119 billion last quarter—as the chipmaker looks to lock in memory and production capabilities to meet AI demand over the coming years. That commitment includes $92 billion in projected spending for the remainder of the fiscal year and an additional $87 billion in fiscal year 2028.

“The thing that matters for the industry is that AI is now doing productive and useful work,” Huang said during Wednesday’s call. “AI is generating profitable tokens…If we had more compute, we could generate more profitable tokens, which results in more profit for all of the services. This is the exact phase where we’re at, which is the reason why everybody’s leaning in.”

Investors will be monitoring whether additional compute capacity translates as neatly into profits as AI companies pour hundreds of billions of dollars into infrastructure.