Amazon just tripled its order of Nvidia chips over ‘surging demand’

The strategic alliance between Amazon and Nvidia has reached a new level of intensity. In a major announcement delivered during Nvidia’s latest quarterly earnings call, the two tech giants revealed a significantly expanded partnership that will see Amazon integrate an additional 2 million Nvidia GPU chips into its AWS (Amazon Web Services) data centers. This massive procurement, slated for deployment throughout 2027 and 2028, underscores the relentless hunger for high-performance compute power required to train and deploy the next generation of artificial intelligence models.

A Massive Expansion of AI Infrastructure

This latest commitment arrives just five months after Amazon initially agreed to deploy over 1 million Nvidia GPUs across its infrastructure. According to Nvidia, the rapid escalation in the order size is a direct response to market realities.

"Demand has exceeded those expectations," Nvidia stated, noting that the surge is being driven by a diverse ecosystem of startups, enterprise corporations, AI research labs, and government entities.

While the companies have opted to keep the specific financial terms of the deal under wraps, the scale of the procurement—involving cutting-edge hardware like the Nvidia Blackwell Ultra, Rubin, and Rubin Ultra GPUs—suggests a transaction valued in the tens of billions of dollars.

Beyond the Chips: A Holistic Integration

The significance of this announcement extends far beyond simple hardware acquisition. The partnership is evolving into a comprehensive integration of Nvidia’s entire technology stack into the AWS ecosystem. This includes:

• Networking Hardware: Advanced systems designed to interconnect thousands of GPUs into a single, cohesive processing unit.
• Software and CPUs: Integration of Nvidia’s data processing software, open-source models, and the new Vera CPUs.
• Robotics and Physical AI: Adoption of Nvidia’s full robotics suite, including Omniverse (digital twins), Cosmos (world models), Isaac (robotics development), and Jetson (edge AI hardware).

By adopting this full-stack approach, Amazon is positioning AWS as the premier destination for companies looking to leverage Nvidia’s most advanced AI tools, from enterprise-grade cloud services to physical warehouse automation.

The Balancing Act: Custom Silicon vs. Nvidia Dominance

Interestingly, this deepening reliance on Nvidia occurs even as Amazon aggressively pursues its own custom silicon strategy. Amazon’s AI chief, Peter DeSantis, has been vocal about the company’s efforts to reduce its dependence on external chipmakers.

Amazon’s internal chip business, which recently crossed a $25 billion annualized revenue run rate, is anchored by two primary pillars: 1. Trainium: A custom AI chip designed to compete directly with Nvidia’s H100 and Blackwell series for deep learning workloads. 2. Graviton: An Arm-based CPU that challenges traditional server processors from Intel and AMD.

Despite these internal successes, Nvidia remains the undisputed leader in the AI hardware space. The inclusion of Nvidia’s Vera CPUs in the new AWS deal highlights that even as Amazon builds its own alternatives, it recognizes the necessity of maintaining access to Nvidia’s specialized, high-performance architecture to satisfy its massive client base.

Financial Momentum and Future Outlook

Nvidia’s financial performance continues to mirror the explosive growth of the AI sector. The company reported $96.2 billion in sales for the second quarter, comfortably beating analyst expectations. The data center segment was the primary engine of this growth, contributing $89 billion—a staggering 117% increase year-over-year.

Looking ahead, Nvidia anticipates revenue to climb to $108 billion in the third quarter, bolstered by the initial rollout of its next-generation Rubin GPUs. To ensure it can meet this insatiable demand, Nvidia has committed a massive $279 billion to secure manufacturing capacity and memory supply for future projects, a significant jump from the $119 billion earmarked just one quarter prior.

The "Profitable Token" Era

During the earnings call, Nvidia CEO Jensen Huang offered a clear perspective on why the industry is pouring hundreds of billions into infrastructure.

"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."

As the industry transitions from experimental AI to "profitable tokens," the race for compute power has become the defining characteristic of the modern tech landscape. Whether this massive capital expenditure translates into sustained, long-term profitability remains the primary question for investors, but for now, Amazon and Nvidia are betting heavily that the demand for AI compute is only just beginning to peak.