Amazon just massively increased its Nvidia chip orders

Amazon and Nvidia just deepened their relationship significantly. On Wednesday, the two companies announced an expanded partnership. The deal includes a major addition. Amazon will add another 2 million Nvidia GPU chips to its data centers.

These GPUs handle heavy compute demands. They’re designed specifically for training and running AI models. The lineup includes Nvidia Blackwell Ultra, Rubin, and Rubin Ultra GPUs. These chips will head to Amazon Web Services data centers. That rollout is planned for 2027 and 2028.

Demand Has Exceeded Expectations

This announcement came during Nvidia’s quarterly earnings call. It arrives just five months after an earlier deal. Amazon had previously agreed to deploy more than 1 million Nvidia GPUs across AWS infrastructure. That earlier deployment started this year. Nvidia addressed this rapid escalation directly. The company said in a statement that “demand has exceeded those expectations.”

Neither company disclosed specific financial terms for this deal. It’s unclear exactly what return Nvidia will see. Still, based on typical GPU unit costs, this deal is worth tens of billions of dollars.

More Than Just Chips

This announcement stands out for a few reasons. It’s notable because of its sheer size. It’s also notable because of how quickly it grew. But there’s another important detail here. The partnership extends well beyond Amazon simply buying more Nvidia chips. This is happening even as Amazon invests in its own competing AI chips.

Nvidia shared more details on Wednesday about what’s included. The company said its broader technology stack will now be integrated across AWS. That includes networking hardware. Specifically, this refers to hardware that connects thousands of GPUs into a single system. It also includes Nvidia’s open models. CPUs are part of the deal too. So is Nvidia’s data processing software. Its robotics platform will be integrated as well.

Both companies pointed to a common driver behind this decision. They cited “surging demand” from multiple sources. That includes startups, enterprises, AI labs, and even government entities.

Amazon’s Own Chip Ambitions

This expanded partnership comes at an interesting moment for Amazon. The company has been ramping up its own AI chip efforts. That’s particularly true for CPUs. These are the general-purpose processors found at the heart of most servers.

Amazon has been developing its own chips for a specific reason. The goal is to reduce dependence on Nvidia. In some ways, Amazon is even positioning itself to compete directly with the chip giant. Peter DeSantis, Amazon’s AI chief, has spoken about this strategy. He’s said AWS is in talks to sell its own Trainium chips to other companies.

Those chips serve as a direct alternative to Nvidia’s H100 or Blackwell chips for deep learning workloads. Amazon’s Arm-built Graviton CPU is viewed similarly. Many see it as a challenger to traditional server chips from Intel and AMD.

Amazon has shared positive updates about this custom chip business. On its last earnings call, the company reported crossing a significant milestone. It reached a $25 billion annualized revenue run rate for its custom chips. That growth was driven largely by $225 billion in total commitments. Those commitments came from AI labs like Anthropic and OpenAI.

Nvidia Remains the Dominant Force

Still, it seems clear that Nvidia remains the dominant force in AI chips.

Alongside the 2 million GPU chips heading to AWS starting in the third quarter, Nvidia has more planned. The company will also send an unspecified number of Vera CPUs. According to Nvidia CFO Colette Kress, some of these will be integrated directly with Rubin. Others will function as standalone units.

Nvidia CEO Jensen Huang has ambitious plans for these Vera CPUs. Back in May, he made a bold claim. He said he had identified a “brand-new $200 billion TAM” for the company.

Beyond AWS, Kress shared more context on Wednesday. She said Nvidia expects Vera to be adopted broadly. That includes every major hyperscaler, neocloud, AI lab, and system OEM. According to Kress, shipments are already underway to lead partners. Those partners include Oracle and SpaceXAI.

Expanding Into Robotics and Enterprise Tools

This partnership extends into other areas too. That includes Amazon’s warehouse robots. It also includes various enterprise offerings.

Kress explained that Amazon plans to adopt Nvidia’s full physical AI stack. This technology will power Amazon’s fleet of robots. The stack includes several key components. That includes Omniverse, Nvidia’s simulation and digital twin platform. It includes Cosmos, the company’s world model platform. Isaac, Nvidia’s robotics development platform, is part of the stack too.

So is Jetson, computing hardware built for robots and edge AI. This week, Nvidia introduced a new version of Jetson. It’s designed to be a more accessible robotics computer. The goal is to support “entry-level edge AI” applications.

On the enterprise side, another integration is planned. AWS will serve Nvidia’s Nemotron family of open models. This will happen through Amazon Bedrock, Amazon’s managed foundation model platform. It will also extend to SageMaker, Amazon’s managed cloud service.

Nvidia’s Quarterly Earnings

Nvidia also shared its broader financial results on Wednesday. The company reported sales of $96.2 billion for the second quarter. That figure beat analyst estimates. Data center revenue made up the bulk of Nvidia’s sales for the quarter. That segment alone brought in $89 billion. That’s up 117% compared to a year earlier.

Nvidia shared expectations for the upcoming quarter too. The company expects revenue to reach $108 billion in the third quarter. Some of that growth will come from Nvidia’s next-generation Rubin GPUs. According to the company, production shipments began this quarter. Investors have been closely watching for early Rubin sales data. Strong Q3 numbers would signal continued demand into Nvidia’s next hardware generation.

A Massive Investment in Future Capacity

Nvidia also revealed a massive financial commitment. The company has committed $279 billion to secure supply and manufacturing capacity. This covers current and future data-center projects. That figure is up substantially from $119 billion just last quarter. Nvidia is working to secure memory and manufacturing capacity. The goal is meeting AI demand over the next several years. This commitment breaks down into two parts. That includes $92 billion in projected spending for the rest of this fiscal year. It also includes another $87 billion planned for fiscal year 2028.

Betting on Profitable Tokens

“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 watching closely going forward. The key question is whether additional compute actually translates into additional profits. This matters especially as AI companies continue pouring hundreds of billions of dollars into infrastructure.

Amazon just tripled its Nvidia chip orders over rising AI demand