Amazon triples Nvidia GPU order, adding 2 million chips for AWS through 2028

Amazon has committed to adding 2 million more Nvidia GPUs to its AWS data centers, roughly tripling a chip order signed just five months earlier, according to TechCrunch. The expanded partnership was announced Wednesday during Nvidia's quarterly earnings call and covers Blackwell Ultra, Rubin, and Rubin Ultra GPUs that will head to AWS facilities in 2027 and 2028.

Nvidia said demand since the original agreement — which covered more than 1 million GPUs — "has exceeded those expectations." Neither company disclosed financial terms, but TechCrunch estimates the deal is worth tens of billions of dollars given per-chip costs.

More than a chip purchase

The agreement goes well beyond Amazon buying hardware. According to TechCrunch, Nvidia said its networking gear — the equipment that links thousands of GPUs into a single system — along with its open models, CPUs, data processing software, and robotics platform will be integrated across AWS. The companies attributed the deeper tie-up to "surging demand" from startups, enterprises, AI labs, and governments.

Nvidia CFO Colette Kress said an unspecified number of Vera CPUs will ship alongside the new GPUs, some integrated into Rubin systems and others standalone. She added that Nvidia expects Vera to be deployed by "every major hyperscaler, neocloud, AI lab, and system OEM," with shipments already underway to lead partners that include Oracle and SpaceXAI. CEO Jensen Huang has separately described server CPUs as a "brand new $200 billion TAM" for the company.

The partnership also reaches into Amazon's warehouses and enterprise stack. Kress said Amazon plans to adopt Nvidia's full physical AI stack for its robot fleet: Omniverse for simulation and digital twins, the Cosmos world model platform, the Isaac robotics development platform, and Jetson edge hardware. On the enterprise side, AWS will host Nvidia's Nemotron family of open models on its Bedrock and SageMaker platforms.

Amazon's hedge keeps growing

The deal lands as Amazon builds silicon that could compete with Nvidia. AWS is in talks to sell its Trainium chips — positioned as an alternative to Nvidia's H100 and Blackwell for deep learning workloads — to other companies for their data centers, and its Arm-built Graviton CPU is seen as a challenger to server processors from Intel and AMD.

Amazon said on its last earnings call that its custom chip business has crossed a $25 billion annualized revenue run rate, supported by $225 billion in total commitments from AI labs including Anthropic and OpenAI, according to TechCrunch. Even so, the scale of the new Nvidia order suggests Amazon is not yet willing to lean entirely on its own accelerators to meet AI demand.

A record quarter behind it

The announcement coincided with Nvidia's second-quarter results: $96.2 billion in sales, beating analyst estimates, with data center revenue of $89 billion up 117 percent year over year. Nvidia guided to $108 billion in revenue for the third quarter, partly driven by its next-generation Rubin GPUs, which began production shipments this quarter.

To feed that pipeline, Nvidia has committed $279 billion to secure memory and manufacturing capacity for current and future data-center projects — up sharply from $119 billion last quarter — including $92 billion in projected spending for the rest of the fiscal year and another $87 billion in fiscal 2028.

Huang argued on the call that "AI is generating profitable tokens" and that more compute would yield more of them, framing the industry's core bet that infrastructure spending converts directly into returns.

Why it matters

Amazon tripling its GPU order within five months is one of the clearest signals yet that hyperscaler demand for AI compute is still accelerating rather than plateauing. It also shows the limits of Amazon's custom-silicon strategy: even a company investing heavily in Trainium and Graviton is locking in tens of billions more with Nvidia so that capacity never becomes the bottleneck. For Nvidia, the deal bolsters the case that Rubin demand will follow Blackwell's — but it raises the stakes on Huang's claim that compute reliably turns into profit, the question investors are watching as industry-wide commitments climb into the hundreds of billions.