Nvidia (NASDAQ: NVDA) supplies the world's best graphics processing units (GPUs) for data centers, which are the primary chips used in artificial intelligence (AI) training and inference workloads. Demand for this hardware is outstripping supply as technology giants spend record amounts of money to build infrastructure in the race for AI supremacy.
Nvidia was a $360 billion company when the AI boom started gathering momentum in early 2023, but it's now worth almost 15 times more, with a market capitalization of $5.3 trillion at the close of trading last Friday. According to a series of comments by CEO Jensen Huang last week, there might be plenty of growth left in the tank. Read on.
Missed AI's "Act 1"? Act 2 Could Be 15x Bigger. Most investors think they missed the AI boat because they didn't buy Nvidia in 2005. But according to our analysts, we're only at the end of "Act 1"—the R&D phase. "Act 2" is the global rollout. Continue »
Nvidia's chip sales could double next year
Nvidia's Blackwell Ultra architecture is at the foundation of its flagship GB300 GPU, which is currently the most popular data center chip in the AI industry. It offers up to 50 times more performance in certain configurations compared to the company's original Hopper-based H100 GPU from 2022, highlighting the sheer pace of innovation.
But Nvidia just started shipping its next-generation data center system called Vera Rubin. It includes the Rubin GPU, the Vera central processing unit (CPU), and a series of upgraded networking components, and it can provide up to 30 times more performance than the equivalent Blackwell Ultra system. Moreover, it can reduce inference token costs by a staggering 97%.
Inference tokens are the text, images, and computer code an AI model generates in response to a prompt, so Vera Rubin will significantly reduce the cost of deploying AI software. This will likely encourage more AI adoption while also making data center operators more profitable, which will spur further demand for Nvidia's chips.
A few months ago, Jensen Huang said every frontier AI company planned to adopt Vera Rubin at launch, which wasn't the case for Blackwell Ultra, so this new architecture was already expected to be a bigger commercial success than its predecessor. But Huang offered even more color during a speech in Scotland last Thursday, saying Nvidia could sell twice as many chips next year compared to this year.
Nvidia doesn't disclose how many chips it sells annually. However, it's on track to deliver around $411 billion in total revenue during its current 2027 fiscal year (which ends on Jan. 31, 2027), based on Wall Street's average estimate. Moreover, management recently told shareholders to expect 70% revenue growth in fiscal 2028, implying the company could bring in almost $700 billion next year.