The relationship between Amazon and Nvidia has taken a significant step forward. The two companies announced on Wednesday a major expansion of their existing partnership, centered on a deal for Amazon to add roughly 2 million Nvidia graphics processing units to its data center infrastructure - effectively tripling a commitment the companies had made just five months earlier.
The chips at the heart of the agreement - Nvidia's Blackwell Ultra, Rubin, and Rubin Ultra GPUs - are built specifically to handle the intensive computational workloads involved in training and operating artificial intelligence models. According to the announcement, these processors are scheduled to be delivered to Amazon Web Services data centers throughout 2027 and 2028.
The news broke during Nvidia's quarterly earnings call. It follows an agreement reached just five months ago in which Amazon committed to deploying more than 1 million Nvidia GPUs across AWS infrastructure beginning this year. Since that earlier deal was struck, Nvidia stated that actual demand has outpaced what either company had projected at the time.
Neither Amazon nor Nvidia disclosed the financial terms of the expanded arrangement. However, given the known market pricing for high-end GPU units, industry observers widely estimate the total value of the deal runs into the tens of billions of dollars.
More Than a Chip Purchase
What makes this agreement particularly notable is that it goes well beyond a straightforward hardware procurement deal. The partnership now encompasses a much broader integration of Nvidia technology across Amazon's cloud platform. This includes Nvidia's high-speed networking hardware - the interconnect systems that link thousands of GPUs into unified computing clusters - as well as its open AI models, central processing units, data processing software, and robotics platform, all of which will be woven into the AWS ecosystem.
Both companies attributed the decision to deepen their collaboration to what they described as "surging demand" from a wide range of customers, including technology startups, large enterprises, dedicated AI research labs, and government agencies.
Amazon's Own Chip Ambitions Run in Parallel
The expansion of the Nvidia partnership is happening simultaneously with Amazon's continued push to develop its own silicon. Amazon has been investing heavily in proprietary chip development with the explicit goal of reducing its reliance on Nvidia and, in some cases, competing directly with the company in the data center market.
Amazon's AI chief, Peter DeSantis, has indicated that AWS is in active discussions to sell its Trainium chips - designed as a direct alternative to Nvidia's H100 and Blackwell chips for deep learning applications - to external companies for use in their own data centers. The company's Arm-based Graviton CPU is similarly positioned as a challenger to conventional server processors from Intel and AMD.
Amazon has reported that its custom chip business is gaining meaningful traction. During its most recent earnings call, the company noted that its chip-related operations had crossed a $25 billion annualized revenue run rate, supported by $225 billion in total commitments from prominent AI organizations including Anthropic and OpenAI.
Despite this parallel development effort, the scale of the new Nvidia deal makes clear that Nvidia remains the dominant force in AI chip supply, and that Amazon sees no conflict in pursuing both strategies at once.
Nvidia's Vera CPUs Enter the Picture
In addition to the 2 million GPU units set to arrive at AWS beginning in the third quarter, Nvidia also plans to supply an unspecified quantity of its Vera CPUs. According to Nvidia CFO Colette Kress, some of these processors will be integrated directly with Rubin GPU systems, while others will be deployed as standalone units.
Nvidia CEO Jensen Huang has spoken ambitiously about the Vera CPU product line. Earlier this year, he described it as representing a brand new $200 billion total addressable market for the company - a significant expansion of Nvidia's ambitions beyond the GPU segment it has long dominated.
Kress confirmed on Wednesday that Nvidia anticipates Vera CPUs will be adopted by virtually every major cloud provider, AI-focused infrastructure company, artificial intelligence research lab, and original equipment manufacturer in the server market. She noted that shipments to lead partners - including Oracle and SpaceX's AI division - are already underway.
Robotics and Enterprise Platforms Also Included
The expanded agreement extends into two additional areas: warehouse robotics and enterprise AI services.
On the robotics front, Kress said Amazon intends to adopt Nvidia's full physical AI software and hardware stack to power its fleet of warehouse robots. That stack includes several distinct platforms:
- Omniverse - Nvidia's simulation and digital twin environment
- Cosmos - its world model platform for physical AI
- Isaac - its dedicated robotics development platform
- Jetson - computing hardware built for edge AI and robotic applications
This week, Nvidia also introduced a new iteration of the Jetson platform, specifically designed to serve as a more affordable and accessible robotics computer for what the company described as "entry-level edge AI" use cases.
On the enterprise side, AWS will host Nvidia's Nemotron family of open AI models through two of its managed services: Amazon Bedrock, the company's managed foundation model platform, and SageMaker, its cloud-based machine learning service.
Nvidia Posts Record Quarterly Results
Alongside the partnership announcement, Nvidia reported its financial results for the second quarter, which came in ahead of analyst expectations. The company recorded total sales of $96.2 billion for the quarter. Data center revenue accounted for the bulk of that figure, reaching $89 billion - representing a 117 percent increase compared to the same period one year earlier.
Looking ahead, Nvidia projected that third-quarter revenue would reach approximately $108 billion. A portion of that anticipated revenue is expected to come from initial sales of its next-generation Rubin GPUs, with the company confirming that production shipments of Rubin hardware have already begun this quarter. Investors have been closely monitoring early Rubin sales figures as an indicator of whether strong demand will carry through into Nvidia's next hardware generation.
To secure the manufacturing capacity and component supply needed to meet projected AI demand over the coming years, Nvidia said it has committed $279 billion in total procurement and manufacturing agreements - a substantial increase from the $119 billion figure reported the previous quarter. That total includes $92 billion earmarked for spending during the remainder of the current fiscal year, along with an additional $87 billion planned for fiscal year 2028.
Huang on the State of AI Infrastructure Investment
During Wednesday's earnings call, Nvidia CEO Jensen Huang offered his assessment of where the AI industry currently stands and why infrastructure investment continues to accelerate.
"The thing that matters for the industry is that AI is now doing productive and useful work. 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."
Whether that logic holds up in practice will be a key question for investors and analysts in the months ahead. As AI companies collectively commit hundreds of billions of dollars to infrastructure buildout, the market will be watching closely to determine whether additional computing capacity translates as directly into additional revenue and profit as Huang and others have suggested.
