Nvidia CEO makes trillion-dollar forecast for AI chip sales

SAN JOSE, California Nvidia chief executive officer Jensen Huang, addressing crowds at the company’s biggest annual event on March 16, unveiled a variety of new products while predicting that its flagship AI processors would help generate US$1 trillion (S$1.3 trillion) in sales through 2027.

During a 2½-hour keynote address, Mr Huang announced plans to push deeper into central processing units (CPUs) – Intel’s home turf – and introduced semiconductors made with technology acquired from start-up Groq. The company even said it was developing chips for data centres in outer space.

At the heart of Mr Huang’s message: Demand for computing power continues to soar, and Nvidia is uniquely equipped to meet the challenge. “I believe that computing demand has increased by one million times in the last two years,” he said. “It is the feeling that we all have. It is the feeling every start-up has.”

Mr Huang is contending with increasingly sceptical investors, who want more evidence that Nvidia’s booming sales growth will last. The trillion-dollar sales forecast, fuelled by orders for the company’s Blackwell and Rubin chips, offered evidence that demand remains solid. Still, the outlook does not represent a major acceleration in sales growth. The company had previously forecast that data centre gear would bring US$500 billion in sales through the end of 2026. The latest forecast extends the outlook by another year, doubling the cumulative amount.

After initially rising as much as 4.8 per cent, the company’s shares soon pared their gains on March 16. The stock was up 1.6 per cent to US$183.19 at the close in New York.

“The update should ease fears of a pullback in 2027 as Rubin enters the cycle, though it may also reset market expectations higher and raise the bar again,” Bloomberg Intelligence analyst Kunjan Sobhani said in a note.

New products and partnerships were a highlight of GTC, an annual gathering that draws fervent crowds to San Jose, California. The new Groq chip is designed to boost the responsiveness of artificial intelligence systems. The company also showed off a computer made up of general-purpose CPUs. That opportunity is “for sure” a multibillion-dollar business, Mr Huang said. 

The GTC roll-out is Nvidia’s latest bid to promote AI computing and keep customers loyal to its technology. The company uses the event to announce partnerships with companies in a range of industries, aiming to show the increasing benefits of AI. A flood of spending on AI chips has turned Nvidia into the world’s most valuable company. But it is facing mounting competition from rivals like Advanced Micro Devices, as well as its own customers attempting to produce in-house chips to handle AI. 

Nvidia has accelerated its technology development in recent years, seeking to replace its entire product line-up on an annual basis while adding new components. The next design of its flagship AI processors – appearing in the second half of 2026 – is called Vera Rubin, named for the pioneering astronomer whose observations provided evidence supporting the existence of dark matter. 

Rubin will be followed by a generation named after the American physicist Richard Feynman, who died in 1988. It will have customised high-bandwidth memory, the company said, without giving further details.

While Nvidia continues to post sales growth that is the envy of the chip industry, its stock rally has stalled in recent months. The shares were down 3.4 per cent in 2026 heading into the GTC presentation, leaving the company’s market value at a still-unrivalled US$4.4 trillion.

Mr Huang announced that the Groq 3 LPU will now be part of Nvidia’s product catalogue. An LPU, or language processing unit, is a specialised chip that is good at accelerating large language model inference – the process of generating responses to AI prompts. Such semiconductors have fast memory built onto the chip that helps them generate text almost instantly. Nvidia will offer it as a coprocessor, complementing work done by its accelerators. The latter components are better at dealing with more complex, multistage tasks.

In December, Nvidia announced what it called a licensing deal with Groq that gave it the rights to use certain technologies and designs. While the start-up still exists, its founders and a big chunk of its engineers joined Nvidia in what was practically an acquisition.

Nvidia has accelerated the engineering work done by Groq to bring its product to market more quickly. The chip will be manufactured by Samsung Electronics, with Groq-based systems coming in the second half, Mr Huang said. 

The chipmaker said that its forthcoming CPU, branded Vera, is more capable than previous versions of that chip. As AI data centres become increasingly complex, the orchestration of work divided between various types of computers and software – a job performed by general-purpose CPUs – is becoming more important, Nvidia said.

Vera will combine the attributes of CPUs used in data centres, gaming PCs and laptops, the company said. It will be able to deal with many inputs simultaneously while quickly handling single complex ones. It also will require less electricity, according to Nvidia.

The company also plans to start selling computers made up entirely of the CPUs, a new approach for the chip giant. Such machines can be used in combination with other Nvidia-based computers or work independently.

Nvidia has pushed beyond its hallmark of graphics processing units, or GPUs, which are used to train and run artificial intelligence software. It now offers complete computer systems featuring processors, networking and software.

The chipmaker also provides AI models and other software on an open-source basis, meaning that customers can tinker with the technology as they see fit. The company even offers versions tailored for specific uses, aiming to help industries that it sees as ripe for disruption by AI. BLOOMBERG