Amazon plans to purchase 2 million Nvidia chips to expand its data center capacity, Bloomberg reports. The order ranks among the largest single GPU commitments disclosed by a cloud provider and signals that hyperscaler spending on accelerated computing hardware continues to scale upward.

The reported figure — 2 million chips — sets the scale of the procurement. Nvidia's current data center portfolio, including its flagship AI accelerators, sits at the center of the order. Bloomberg did not disclose the specific chip models, the delivery schedule, or the total contract value in its initial report.

What does the purchase signal for the AI hardware market?

The size of the order matters for three reasons:

• It ties Amazon's infrastructure roadmap to Nvidia's supply chain for years, not quarters, given the manufacturing volumes required to fulfill a commitment of this magnitude.
• It confirms that demand for AI training and inference capacity at AWS-scale continues to outrun current data center footprints.
• It adds a data point to the broader hyperscaler capex trend, where the largest cloud operators have repeatedly raised hardware budgets to compete for AI workloads.

For Nvidia, a 2 million-chip customer commitment reinforces its position as the primary supplier of accelerators to the largest cloud platforms. For Amazon, the purchase deepens its inventory of compute capacity ahead of expected growth in AI-driven demand from AWS customers.

How does this fit Amazon's build-out strategy?

Amazon has framed its capital spending in recent quarters around data center expansion, power procurement, and networking upgrades needed to support generative AI workloads. A chip order of this scale implies corresponding investment in facilities, cooling, and power distribution — GPU-dense racks carry infrastructure costs well beyond the silicon itself.

The procurement also positions Amazon against its direct cloud rivals. Microsoft, Google, and Meta have each committed multi-billion-dollar sums to Nvidia hardware and to alternative silicon developed in-house. Amazon maintains its own custom accelerator line alongside Nvidia GPUs, and the reported order indicates it intends to scale both tracks simultaneously rather than substitute one for the other.

Bloomberg's report does not specify how the 2 million chips split across training and inference deployments, nor whether deliveries stretch across multiple product generations.

What remains unconfirmed?

Key commercial terms are not yet public. The report leaves open the per-unit pricing, the contract's timeline, and any allocation agreements tied to Nvidia's production capacity. Chip pricing for large buyers varies by generation and volume, so a total deal value cannot be derived from the unit count alone.

Nvidia has faced supply constraints during previous demand spikes. A commitment of 2 million units implies extended delivery windows and coordinated planning with Nvidia's manufacturing partners.

Amazon and Nvidia had not issued public statements elaborating on the order at the time of the report.

Why the number matters

Two million chips represents compute capacity measured in the hundreds of thousands of GPU-dense racks once deployed. Orders at that scale move the needle on:

• Global data center electricity demand
• Advanced packaging supply at Nvidia's contract manufacturers
• Competitive positioning among the top four cloud providers

The purchase confirms that the largest AI infrastructure race now runs through multi-year silicon procurement, with Amazon's 2 million-chip commitment standing among the most concrete markers of that shift to date.