Amazon Triples Nvidia GPU Order to 2 Million Chips for AI Expansion
The cloud giant is integrating Nvidia's full AI stack into its data centers and warehouse robotics.
Amazon is dramatically scaling its AI infrastructure by tripling its order of Nvidia GPUs, adding 2 million chips to its data centers over the next two years. The move comes as demand for generative AI capabilities from enterprises, governments, and startups continues to outpace previous projections.
According to TechCrunch, the expanded order targets deployment in 2027 and 2028 and includes high-end hardware such as the Blackwell Ultra, Rubin, and Rubin Ultra models. The partnership extends beyond raw compute; Amazon will also integrate Nvidia's physical AI platform—including Jetson, Omniverse, and Isaac—into its warehouse robotics operations. Additionally, AWS will now serve Nvidia's Nemotron family of open models via Amazon Bedrock and SageMaker.
The Compute Arms Race
This expansion follows a previous agreement to deploy over 1 million GPUs starting in 2026. The surge in demand coincides with a period of explosive growth for Nvidia, which reported Q2 revenue of $96.2 billion. Data center revenue alone accounted for $89 billion of that total, representing a 117% increase year-over-year. Nvidia CEO Jensen Huang emphasized the economic driver behind this growth, stating that more compute allows for the generation of "more profitable tokens," which in turn increases profit for AI services.
Strategic Hedging
What makes the deal notable is Amazon's simultaneous pursuit of silicon independence. The company continues to scale its own custom chip business, including Trainium and Graviton, which has reached an annualized revenue run rate of $25 billion. Amazon is even exploring the possibility of selling its Trainium chips to external companies. By doubling down on Nvidia while developing its own hardware, Amazon is hedging its bets to ensure it can meet immediate capacity needs without sacrificing long-term autonomy.
Industry Implications
This partnership signals that Nvidia's dominance remains intact despite the efforts of "hyperscalers" to build competing in-house silicon. The integration of Nvidia's software and robotics ecosystem into Amazon's physical logistics suggests a strategic shift where AI is moving from the cloud into the physical world of warehouse automation.
What to Watch
Industry observers will be monitoring whether Amazon's internal chip efforts can eventually reduce its reliance on Nvidia's high-cost hardware. While the 2 million GPU order secures Amazon's capacity through 2028, the long-term trajectory of the partnership will depend on the performance of Amazon's own AI silicon and the continued pace of enterprise AI adoption.