Amazon Deepens Nvidia Ties With Massive 2 Million-GPU Expansion

Amazon is dramatically expanding its Nvidia hardware footprint, agreeing to add another 2 million GPUs to Amazon Web Services data centers in a move that underscores how quickly AI compute demand is still growing. The new order, announced Wednesday, comes only months after Amazon said it would deploy more than 1 million Nvidia GPUs across AWS infrastructure.

The size and speed of the deal matter for two reasons: they show that cloud providers are still racing to secure scarce AI chips, and they highlight how even Amazon, which is building its own processors, continues to rely heavily on Nvidia for the most demanding AI workloads.

The expanded arrangement was disclosed during Nvidia’s quarterly earnings call and includes chips from the company’s current and next-generation road map, including Blackwell Ultra, Rubin, and Rubin Ultra. The GPUs are slated for AWS data centers in 2027 and 2028.

Neither company revealed the financial terms, but the scale of the order suggests a contract worth tens of billions of dollars once chip prices and associated infrastructure are taken into account.

What makes this latest announcement especially notable is that it goes far beyond a simple chip purchase. Nvidia said its networking gear, CPUs, data processing software, open models, and robotics tools will also be folded more deeply into AWS offerings, turning the relationship into a broader platform partnership.

That shift arrives at a moment when Amazon is still developing its own AI silicon to reduce dependence on Nvidia. Rather than signaling a retreat from in-house chip ambitions, the new deal shows that Amazon is pursuing a two-track strategy: build custom hardware where it can, while buying leading-edge Nvidia systems where performance and customer demand justify the expense.

What Amazon and Nvidia agreed to

Amazon’s latest commitment adds 2 million more Nvidia GPUs to AWS, bringing the cloud giant’s purchase plans much closer to the center of the AI infrastructure boom. The chips will be deployed across Amazon data centers beginning in the latter half of the decade, supporting training and inference for large AI models.

Nvidia said the order was driven by what both companies described as “surging demand” from a wide range of customers, including startups, large enterprises, AI labs, and government buyers. That phrasing is important because it suggests the demand is not isolated to one sector or one class of customer; it is broad-based and still accelerating.

The new deal follows an earlier agreement from roughly five months ago in which Amazon committed to deploying more than 1 million Nvidia GPUs across AWS infrastructure starting this year. Since then, Nvidia said demand has outpaced even those elevated expectations.

The partnership also includes Nvidia’s Vera CPUs, with some units integrated alongside Rubin chips and others shipped as standalone processors. Nvidia chief financial officer Colette Kress said those processors are already being sent to selected partners and will be deployed widely across major cloud and AI customers.

Why this deal matters for the AI chip race

The Amazon-Nvidia expansion is more than a supply agreement; it is a snapshot of the state of the AI infrastructure market. The leading cloud companies are under pressure to keep adding compute as model builders demand more capacity, while chip makers are fighting to stay indispensable in a market where customers increasingly want optionality.

For Amazon, the deal confirms that its own chip development has not yet displaced Nvidia as the default choice for the most compute-intensive AI tasks. For Nvidia, it reinforces the company’s dominant position in data-center AI even as rivals, including customers, invest in alternatives.

Amazon has spent years developing custom silicon to reduce its dependence on external suppliers and improve economics inside AWS. Its Trainium chips are marketed as an alternative to Nvidia accelerators for deep learning workloads, while Graviton CPUs offer a homegrown option for general-purpose server computing. Together, those products are part of Amazon’s push to control more of the stack and potentially sell those capabilities more broadly.

Yet the scale of Wednesday’s announcement suggests the market for premium AI compute is still so strong that Amazon cannot rely on its own chips alone. Customers want access to the best available hardware, and Nvidia continues to set the performance benchmark for many AI training and deployment jobs.

How Amazon’s chip strategy fits around Nvidia

Amazon is not abandoning its custom-chip plan. Instead, it is layering Nvidia purchases on top of an internal roadmap designed to diversify its supply, improve margins, and compete more effectively with other cloud players.

The company has said its custom-chip business is growing quickly, and on its most recent earnings call it pointed to a $25 billion annualized revenue run rate driven by commitments tied to AI customers such as Anthropic and OpenAI. Those numbers illustrate how large Amazon’s in-house opportunity could be, even if Nvidia remains the preferred partner for cutting-edge workloads.

The strategic question is whether Amazon can use its own silicon to capture cost-sensitive workloads while leaving high-performance, high-margin AI jobs to Nvidia. If that division holds, Amazon could protect AWS economics without sacrificing customer demand for premium accelerators.

In practice, the new agreement implies that Amazon sees enough near-term demand to justify buying aggressively from both camps.

What Nvidia is adding beyond GPUs

Nvidia’s role in the expanded AWS partnership extends well past graphics processors. The company said Amazon will also integrate its networking systems, CPUs, software tools, open models, and robotics technologies into AWS environments.

That matters because modern AI infrastructure is no longer just about the chip itself. To build and run large-scale systems, cloud providers also need the fast interconnects, orchestration software, simulation tools, and AI frameworks that help thousands of processors function as a unified system.

Nvidia’s stack includes hardware and software layers that make it easier for customers to deploy everything from model training clusters to robot-control systems. In other words, the company is selling not only acceleration, but an ecosystem.

For Amazon, that broader integration could make AWS more attractive to customers that want a relatively seamless path from experimentation to production. It also gives Nvidia more influence over the architecture of the AI cloud market, even when the chips are hosted inside a rival cloud platform.

How big is the financial commitment?

The companies did not disclose price tags, but the deal is enormous by any reasonable estimate. AI GPUs remain among the most expensive chips in the world, and the associated systems include networking and support infrastructure that multiply the total cost.

Based on market pricing and the sheer volume of units involved, analysts would likely view the order as worth tens of billions of dollars. That makes it one of the largest cloud infrastructure commitments publicly tied to a single chip supplier in the AI era.

For Nvidia, the new agreement adds more visibility into future revenue at a time when investors are scrutinizing how long the AI buildout can continue. For Amazon, it is another large but strategically necessary bet that the demand curve for AI services will continue to rise.

The deal also reflects a practical reality: cloud providers cannot build every layer of their AI stack fast enough to keep up with customer demand if they rely only on internal development. Buying from Nvidia buys time, capacity, and credibility.

Why Nvidia says demand is still accelerating

Nvidia said the deeper partnership with Amazon came after demand exceeded the earlier expectations set just months ago. That is a meaningful signal because it suggests AI buildouts are still moving faster than even the industry’s most bullish forecasts.

Nvidia said the increase follows a period in which customer demand outstripped what the companies had initially expected.

The company pointed to demand from startups, enterprises, AI labs, and governments. Together, those groups indicate a broad market for AI compute that spans experimentation, production, public-sector adoption, and frontier model development.

Jensen Huang used Wednesday’s call to argue that the industry has moved into a phase where AI systems are already doing productive work and generating returns. In his view, more compute translates into more output, more profitable tokens, and ultimately more revenue for AI service providers.

Investors have heard similar arguments before, but the question now is whether the economics will keep working as spending rises into the hundreds of billions of dollars.

What does Nvidia expect next?

Nvidia expects its third-quarter revenue to reach $108 billion, with some of that growth tied to Rubin, its next-generation GPU platform. The company said it has already begun production shipments of Rubin this quarter.

That timeline is important because it shows Nvidia is not simply riding demand for its current chips; it is using the AI buildout to lock in sales for future product generations as well. If customers commit now to Rubin and Rubin Ultra, Nvidia gains a longer runway of expected revenue.

For AWS, that means future capacity planning will need to account not only for current-generation clusters but also for upcoming waves of hardware. Large cloud providers increasingly have to think several years ahead when negotiating chip supply.

How Amazon’s robotics plans fit into the deal

Amazon’s robotics business is also becoming part of the Nvidia partnership. Nvidia said Amazon plans to adopt its full physical AI stack to support warehouse robots, including Omniverse, Cosmos, Isaac, and Jetson.

Those tools serve different roles across simulation, world modeling, robotics development, and edge computing. Together, they can help companies test robot behavior in digital environments before deploying systems in real facilities.

That is especially relevant for Amazon, which operates one of the world’s largest and most automated logistics networks. Warehouse robotics can improve throughput, reduce manual handling, and help the company manage the complexity of fulfillment at scale.

Nvidia also used the week to introduce a new version of Jetson aimed at making robotics computing more accessible for entry-level edge AI applications. That product positioning suggests the company wants its stack to reach beyond elite model developers and into practical industrial deployments.

Why physical AI matters to Amazon

Physical AI refers to systems that can perceive, reason, and act in real-world environments. For Amazon, that means robotics inside warehouses, automation in fulfillment centers, and potentially more advanced edge devices that can operate closer to the action.

If Nvidia’s stack helps Amazon simulate and train robots more efficiently, the benefits could extend beyond internal operations. A more robust robotics platform could also become part of AWS offerings for enterprise customers interested in automation.

This is one of the clearest signs that the Amazon-Nvidia relationship is expanding beyond pure cloud compute and into industrial AI.

What the numbers say about Nvidia’s momentum

Nvidia’s quarterly results added another layer to the story. The company said it recorded $96.2 billion in sales in the second quarter, beating analyst estimates. Data-center revenue accounted for $89 billion of that total, up 117% from a year earlier.

Those figures confirm that Nvidia’s business is still dominated by AI infrastructure, with cloud and enterprise demand continuing to drive the overwhelming share of sales.

The company also disclosed that it has committed $279 billion to secure supply and manufacturing capacity for current and future data-center projects. That figure is substantially higher than the $119 billion it cited in the prior quarter, reflecting how aggressively it is lining up memory, manufacturing, and logistics support for the next few years.

Of that total, Nvidia says $92 billion is projected to be spent during the rest of the current fiscal year, with another $87 billion expected in fiscal year 2028.

Those commitments show how capital-intensive the AI infrastructure race has become. It is no longer just a chip design contest; it is a battle for fabs, packaging, memory, networking, and power supply.

How does this affect the broader AI market?

The Amazon-Nvidia expansion is likely to reverberate across the rest of the cloud and AI ecosystem. Competitors will watch closely to see whether AWS can monetize the extra capacity fast enough to justify the investment.

The deal also adds pressure on rivals such as Microsoft, Google, Oracle, and smaller neocloud providers to secure their own supply of next-generation AI hardware. If the largest cloud providers are still scrambling for more chips, the rest of the market will likely feel the squeeze.

For AI startups and model developers, the announcement is encouraging in one sense: more capacity should eventually mean more access to powerful infrastructure. But it also highlights the likely cost of entry, since premium AI training remains concentrated among the few companies with the capital and supply relationships to buy chips at this scale.

The concentration of hardware supply also raises competitive questions. If one supplier controls the majority of the fastest chips and multiple cloud platforms depend on it, the AI market can become more vulnerable to pricing power, delays, and bottlenecks.

Timeline of the Amazon-Nvidia buildup

The pace of the partnership’s growth helps explain why the latest announcement is drawing so much attention. In just a few months, Amazon’s Nvidia commitment has gone from large to enormous.

What investors will be watching next

For shareholders, the central question is not whether AI demand exists, but whether the economics justify the pace of spending. Nvidia is clearly benefiting from the buildout, but cloud customers must still turn that hardware into profitable services.

Amazon’s challenge is to prove that buying more Nvidia capacity will translate into stronger AWS growth, higher utilization, and better customer retention. At the same time, it must continue developing its own chips so it is not locked into a single supplier for the most valuable workloads.

Investors will also watch whether future quarters confirm Huang’s claim that more compute produces more profitable AI activity. If usage scales as promised, the industry’s spending spree may still look rational. If not, the infrastructure rush could eventually face margin pressure.

For now, however, the message from Amazon and Nvidia is clear: the AI hardware race is far from cooling, and the biggest buyers are still preparing for much more demand ahead.

The latest partnership shows a market in which scale, supply, and long-term chip access can be as important as model quality. Amazon wants the flexibility to buy, build, and sell AI infrastructure at once. Nvidia wants to remain the foundation beneath that entire stack. Wednesday’s announcement suggests both companies believe the next phase of AI growth will reward that arrangement.

Key facts at a glance

• Amazon will add 2 million Nvidia GPUs to AWS data centers.
• The chips include Blackwell Ultra, Rubin, and Rubin Ultra.
• Delivery is scheduled for 2027 and 2028.
• The deal expands beyond GPUs to CPUs, networking, software, and robotics tools.
• Amazon is still developing its own chips, including Trainium and Graviton.
• Nvidia reported $96.2 billion in quarterly sales, with $89 billion from data centers.

As cloud providers continue to chase scarce AI infrastructure, the Amazon-Nvidia partnership is becoming one of the clearest examples of how much compute the industry thinks it will need — and how far ahead it is willing to pay for it.