The artificial intelligence (AI) infrastructure boom is running into a physical limit that money alone can’t solve: electricity. Hyperscalers can spend hundreds of billions of dollars on servers, but those servers still need power, and new data-center capacity can take years to connect to the grid. That is changing the economics of AI computing. The scarce resource is increasingly not just chips or capital, but energized megawatts.
That creates an opening for Nvidia (NVDA), whose next-generation Vera Rubin platform is designed to squeeze more computing — and more revenue — from every watt. Let’s take a closer look.
More News from Barchart
The $40 Billion Opportunity
During Nvidia’s second-quarter fiscal 2027 earnings call, CEO Jensen Huang explained that the company’s revenue opportunity has climbed from roughly $18 billion per gigawatt with Hopper to $25 billion with Grace Blackwell and now $40 billion with Vera Rubin. CFO Colette Kress confirmed the same progression.
That is a remarkable change in only a few generations.
|
Platform |
Revenue Opportunity Per GW |
|
General-purpose computing |
$3 billion to $5 billion |
|
Hopper |
~$18 billion |
|
Grace Blackwell |
~$25 billion |
|
Vera Rubin |
~$40 billion |
The important point for investors is that Nvidia isn’t simply selling a faster GPU. It is selling an entire AI factory that includes the Vera CPU, Rubin GPU, NVLink, networking, and other infrastructure. That allows the company to capture more of the spending attached to each gigawatt of deployed capacity.
More Computing From Scarce Electricity
Power efficiency is what makes the model work. Nvidia says Vera Rubin can deliver 30 times higher throughput per megawatt and 35% lower token costs than Grace Blackwell Ultra. Separately, Nvidia reported that early Vera Rubin benchmarks delivered 10 times the throughput per megawatt of the Grace Blackwell NVL72 system.
In plain English, the same electrical connection can support dramatically more AI work. That is key because the customer isn’t buying a chip for the sake of owning a chip. They want to turn computing capacity into revenue. Huang said hyperscalers now have backlogs totaling about $2 trillion, while compute is already translating directly into higher revenue and earnings for those companies. That creates a powerful incentive to deploy the most productive hardware available.