Amazon Deepens Nvidia Bet With Deal for 2 Million Next-Gen GPUs

Amazon is significantly expanding its reliance on Nvidia, agreeing to acquire an additional 2 million graphics processing units to power Amazon Web Services data centers through 2027 and 2028. The massive procurement deal, unveiled during Nvidia’s quarterly financial results, triples the cloud giant's previous commitment of 1 million chips made just five months prior.

The hardware package encompasses Nvidia’s forthcoming Blackwell Ultra, Rubin, and Rubin Ultra processors, as well as an unspecified deployment of Vera central processing units. While neither party disclosed the contract's explicit valuation, the scale of the order represents tens of billions of dollars in enterprise hardware spending.

Beyond Raw Silicon

The expanded pact broadens the technical integration between the two tech giants well beyond standard server racks. Amazon will incorporate Nvidia's end-to-end physical automation toolchain—including the Omniverse simulation engine, Cosmos world models, Isaac software suite, and Jetson edge computing boards—into its warehouse robotics infrastructure. Additionally, AWS will host Nvidia’s Nemotron family of foundation models on its Bedrock and SageMaker platforms.

The aggressive procurement highlights an ongoing paradox inside Amazon’s infrastructure strategy. AWS has invested heavily in designing proprietary silicon, marketing its Trainium accelerators as cheaper alternatives to Nvidia hardware and touting an in-house chip business running at a $25 billion annualized revenue pace. Yet when it comes to satisfying near-term appetite from enterprise clients, external AI labs, and sovereign entities, proprietary alternatives remain insufficient to meet overall demand.

Unrelenting Data Center Demand

Nvidia paired the announcement with another record quarter, posting $96.2 billion in second-quarter revenue, with data center sales surging 117% year-over-year to $89 billion. Looking ahead, the chipmaker projected third-quarter revenue of $108 billion as early Rubin production units begin shipping. To support this trajectory, Nvidia expanded its long-term manufacturing and supply commitments to $279 billion, up from $119 billion the prior quarter.

For enterprise cloud providers, the deal underscores how difficult it remains to displace Nvidia's dominance. Even with hyperscalers spending billions to develop in-house application-specific integrated circuits (ASICs) to curb margin pressure, customer preference for Nvidia's mature software ecosystem and the raw compute of its Rubin roadmap continues to dictate cloud capital expenditure priorities.