Nvidia delivers 1 million AI chips to Amazon in multi‑year cloud computing deal

Nvidia’s Vera Rubin POD, the architecture that underpins the new shipment, combines seven Hopper‑based GPUs into five rack‑scale systems that together form a single AI supercomputer, according to Nvidia’s technical blog. The design delivers up to 2 exaflops of AI performance per pod, a scale that Amazon can replicate across its global cloud infrastructure. By distributing a million of these chips over several years, Amazon will be able to spin up thousands of such pods, dramatically expanding its capacity for generative‑AI workloads, large‑model training, and inference services for enterprise customers.

The deal, first reported by Meyka and later echoed by CXO Digitalpulse, is structured as a multi‑year supply agreement that will see the chips delivered by the end of 2027. Nvidia will ship the units in batches that align with Amazon’s data‑center rollout schedule, allowing the cloud provider to integrate the hardware into its existing hyperscale fleet without major redesign. The agreement also includes a service component: Nvidia will provide software stacks, including the Nvidia AI Enterprise suite and the latest CUDA and TensorRT versions, to ensure that Amazon’s customers can leverage the full performance envelope of the Vera Rubin pods from day one.

Analysts see the partnership as a bellwether for the broader AI‑infrastructure market. Bloomberg’s recent coverage of Nvidia’s trillion‑dollar revenue forecast through 2027 highlights that the company expects AI chip sales to dominate its growth trajectory. Supplying a million chips to a single cloud operator represents a sizable slice of that projected market, reinforcing Nvidia’s position as the de‑facto supplier for high‑performance AI compute. The scale of the contract also underscores Amazon’s commitment to keeping its AWS platform competitive against rivals such as Microsoft Azure and Google Cloud, which have secured their own large‑scale GPU deals in recent months.

Regulatory context may shape the timing and geography of deliveries. Reuters reported that the U.S. government is considering new export controls on advanced AI chips, which could affect the flow of Nvidia’s products to overseas data centers. While the current agreement is slated for completion by 2027, any tightening of export rules could compel Nvidia and Amazon to adjust supply chains, potentially shifting more production to domestic fabs or to partners like Foxconn, which Reuters noted is already slated to use Nvidia chips in its own manufacturing lines.

The partnership also has implications for cost structures in the cloud. TechCrunch’s recent analysis of billion‑dollar AI‑infrastructure deals points out that the amortization of such high‑performance hardware can drive down per‑inference costs for customers, making AI services more accessible to a broader range of enterprises. By locking in a million‑chip supply, Amazon can spread capital expenditures over a longer horizon, potentially offering more competitive pricing for its AI‑as‑a‑service offerings. This could accelerate adoption of large‑model APIs and custom training pipelines across sectors that have previously been constrained by compute budgets.

Overall, the Nvidia‑Amazon agreement illustrates how the AI boom is translating into concrete, hardware‑driven growth. The Vera Rubin POD’s exascale capabilities, combined with a multi‑year supply chain and integrated software support, give Amazon a decisive edge in delivering next‑generation AI services. At the same time, the deal reinforces Nvidia’s trajectory toward a trillion‑dollar AI‑chip revenue target, while exposing both firms to evolving regulatory landscapes that could reshape the global AI supply chain in the years ahead.