Superpower Daily
OpenAI’s Astra Is Claimed to Solve Non-Sofic Groups Problem, Ra...Model-Serving Platforms Reach 6.1% of AI-Spending Businesses, W...Perplexity Puts Computer Behind MCP, Opening 400+ Services to A...Databricks Says 300M Chart-JSON Pipeline Tops Four Multimodal B...Nvidia’s $63.1 Billion Receivables Add a Cash-Flow Caveat to It...Moonbug Starts AI Tests for Cocomelon Workflows With Legal and...OpenAI’s Astra Is Claimed to Solve Non-Sofic Groups Problem, Ra...Model-Serving Platforms Reach 6.1% of AI-Spending Businesses, W...Perplexity Puts Computer Behind MCP, Opening 400+ Services to A...Databricks Says 300M Chart-JSON Pipeline Tops Four Multimodal B...Nvidia’s $63.1 Billion Receivables Add a Cash-Flow Caveat to It...Moonbug Starts AI Tests for Cocomelon Workflows With Legal and...
Businesspublished

AWS Plans 2 Million More Nvidia GPUs for 2027–28 Despite Its Own Chip Effort

The planned purchase gives Nvidia a major multiyear customer commitment, but the economics behind sustained AI infrastructure spending remain under scrutiny.

By 2 min read
AWS Plans 2 Million More Nvidia GPUs for 2027–28 Despite Its Own Chip Effort
Sign in to save and react

Listen to this story

The audio brief

About 1:31
0:001:31
Read transcript
AWS plans to buy two million more Nvidia GPUs across 2027 and 2028, extending its partnership with the chipmaker even as Amazon develops competing AI hardware. The scale matters: this is a multiyear commitment to Nvidia capacity at a moment when the industry is testing whether massive AI data centers can earn back their cost quickly enough. And the package goes beyond graphics processors. AWS also plans to deploy Nvidia Vera central processing units, or CPUs, and use Nvidia technology for robotics. In other words, this is a broader platform relationship, not just an accelerator order. The unresolved question is Amazon’s silicon mix. Its in-house AI-chip effort continues, so the Nvidia purchases show ongoing demand, but they do not reveal how much AWS ultimately wants to build itself versus buy from Nvidia. The economics are still under scrutiny. CNBC’s Jim Cramer argues that Nvidia customers are seeing profits quickly enough to keep purchasing. CEO Jensen Huang has cited a claim that some customers recovered their AI-infrastructure investment in less than a year, even for data centers costing fifty billion dollars. But that was secondhand commentary, not a companywide disclosure. Nvidia also expects near-term growth from enterprises and neoclouds such as CoreWeave and Nebius. The key fact to watch is whether those customers—and AWS—can sustain spending as the new capacity comes online.

Story brief

3 key points

AWS is committing to a large Nvidia hardware footprint even as Amazon develops competing in-house AI chips. The purchase, spread across 2027 and 2028, covers 2 million additional GPUs plus Nvidia Vera CPUs and robotics technology, making it a broader platform partnership rather than a single-accelerator order. The deal is also a market test: continued hyperscaler spending depends on whether AI data centers generate...

  1. 01

    The planned hardware package extends beyond accelerators to Nvidia Vera CPUs and robotics technology.

  2. 02

    Jensen Huang’s sub-one-year AI-infrastructure return claim was secondhand, not a companywide customer disclosure.

  3. 03

    Amazon’s in-house AI-chip program continues alongside Nvidia procurement, leaving its long-term silicon mix unresolved.

AWS plans to buy 2 million additional Nvidia GPUs in 2027 and 2028 under an expanded partnership with Nvidia. The scale and timing make the agreement a significant multiyear commitment to Nvidia hardware as the industry tests how quickly AI infrastructure can pay for itself.

The order reaches beyond GPUs

The planned deployment includes more than accelerators. AWS also intends to install Nvidia Vera central processing units, or CPUs, and use Nvidia technology for robotics. That puts the partnership across several Nvidia product areas rather than framing it as a single component purchase.

The purchase tests the return-on-infrastructure case

CNBC’s Jim Cramer argues that Nvidia customers are seeing profits quickly enough to sustain their hardware purchases. His view speaks to a central debate around Nvidia: whether customers can generate sufficient returns from AI computing infrastructure to justify continued spending on chips and networking gear.

Nvidia CEO Jensen Huang offered a more specific, but secondhand, version of that argument. He said he had heard that some customers’ return on invested capital for AI infrastructure was less than a year, in the context of data centers costing $50 billion. The remark conveys management’s confidence, but it is not a companywide disclosure of customer returns.

Amazon’s silicon strategy remains dual-track

Amazon has invested heavily in developing its own AI chips. Its planned Nvidia purchases therefore show AWS intends to keep deploying Nvidia technology while it builds alternatives; the agreement does not resolve how those approaches will compete over time.

Nvidia expects growth beyond hyperscalers

The AWS commitment arrives alongside strong recent momentum. CNBC described Nvidia’s fiscal 2027 second-quarter results as better than expected, after a fourth consecutive quarter of accelerating year-over-year revenue growth. Finance chief Colette Kress said non-hyperscaler customers, including neoclouds such as CoreWeave and Nebius and enterprises, should drive growth in the current quarter.

Sources

  1. cnbc.comJim Cramer says this is why demand for Nvidia's chips keeps accelerating

Related coverage

Continue reading