Amazon Triples Its Nvidia Chip Order to Keep Pace with Surging AI Demand

In a move that reflects the intensity of the fierce race for AI infrastructure, Amazon and Nvidia have announced a remarkable expansion of their strategic partnership. The new agreement includes adding two million graphics processing units (GPUs) from Nvidia to Amazon's data centers, in a deal estimated to be worth tens of billions of dollars, coming just a few months after a previous agreement that was considered massive at the time.

Details of the Expanded Deal

The two companies revealed the agreement during Nvidia's quarterly earnings call. It includes chips from advanced models such as Blackwell Ultra, Rubin, and Rubin Ultra—processors specifically designed to handle the heavy computational workloads associated with training and running AI models. These chips are scheduled to arrive at Amazon Web Services (AWS) data centers during 2027 and 2028.

What stands out is the rapid growth of this collaboration; just five months ago, Amazon had agreed to deploy more than one million Nvidia GPUs across its infrastructure starting this year. However, Nvidia noted in its statement that «demand exceeded those expectations,» prompting both parties to double the scale of their commitment.

Beyond Buying Chips

The partnership is not limited to selling more chips; it extends to integrating Nvidia's technology ecosystem more broadly within AWS, including:

• Networking equipment that connects thousands of GPUs into a single integrated system.
• Open-source models and CPUs.
• Data processing software and a robotics platform.

The two companies indicated that the «escalating demand» from startups, major enterprises, AI labs, and even governments was a key driver for deepening this collaboration.

Amazon Also Bets on Its Own Chips

Interestingly, this expansion is happening while Amazon intensifies its efforts to develop its own chips, seeking to reduce its dependence on Nvidia and even compete with it. The company's AI executive noted that talks are underway to sell Trainium chips—the direct alternative to Nvidia chips for deep learning workloads—to other companies. Graviton processors, built on Arm architecture, are also a competitor to traditional server processors from Intel and AMD.

Amazon announced that its custom chip business is growing at a rapid pace, surpassing an annual revenue rate of $25 billion, driven by total commitments worth $225 billion from labs such as «Anthropic» and «OpenAI.» Nevertheless, Nvidia remains the undisputed dominant player in the world of AI chips.

Vera Processors and Robotics

Alongside the GPUs, Nvidia plans to ship an unspecified number of Vera processors, some integrated with Rubin chips and others standalone. CEO Jensen Huang holds great ambitions for these processors, having previously spoken of a potential market worth $200 billion.

The partnership also extends to Amazon's warehouse robots, where the company intends to adopt Nvidia's full physical AI ecosystem, including the Omniverse, Cosmos, Isaac, and Jetson platforms. On the enterprise side, AWS will offer open-source Nemotron models via the Amazon Bedrock and SageMaker platforms.

Record Numbers for Nvidia

Nvidia recorded sales of $96.2 billion in the second quarter, exceeding analyst estimates, with data center revenue accounting for the largest share at $89 billion—a 117% growth compared to the previous year. The company expects its revenue to reach $108 billion in the third quarter.

Huang said that what matters to the industry today is that AI «has begun to do useful and productive work,» adding that greater computing power means higher productivity and larger profits for all services. The most important question that investors are waiting to see answered remains: will these massive investments in computing power actually translate into tangible profits?

✦ بقلم فريق دروب أيديا