(Bloomberg) — Nvidia Corp., whose products have fueled a flood of artificial intelligence spending, said new types of AI models that produce more complex answers will only increase the need for computing infrastructure.
Most Read from Bloomberg
Chief Executive Officer Jensen Huang said that concerns caused by Chinese startup DeekSeek’s R1 AI model — that fewer chips and powerful servers would be needed for such software in the future — were misplaced.
“The understanding of R1 was completely wrong,” Huang said at a meeting with analysts and investors at his company’s GTC conference Wednesday in San Jose, California. “Computation demand is much higher.”
Shares of Nvidia rose as much as 2.4% in New York trading. They had been down 14% this year heading into the session.
The company is trying to persuade a wider group of industries to invest in AI equipment, promising that the economic benefits of the technology are at hand. Nvidia, which became the world’s most valuable chipmaker after a dizzying growth run over the last two years, is facing more investor scrutiny in 2025. The key concern is whether customers will maintain their spending on AI infrastructure.
DeepSeek stoked those fears earlier this year when it released its AI model, saying it produced the powerful technology for less money. But Nvidia’s largest customers have reaffirmed their spending plans since then. And analysis by Bloomberg Intelligence this week found that expenditures by the biggest data center operators is actually climbing faster than anticipated.
At the analyst meeting, Nvidia’s Huang was asked about the efforts of customers to develop their own components — chips that might replace his AI accelerators in data centers. Companies like Alphabet Inc.’s Google have been working with Broadcom Inc. to develop their own application specific integrated circuits, or ASICs, for this area. But Huang countered that many ASICs are designed but not always actually put into data centers.
Those big customers need better chips to generate more revenue from their infrastructure, not cheaper ones to save costs, he argued.
“All of those companies are run by great CEOs who are really good at math,” he said. “The effects are not just cost. It’s a different calculus.”
Competitors’ chips can’t match Nvidia’s Hopper design, its previous generation, he said. And the current Blackwell platform is 40 times more powerful.
2025-03-19 17:09:00
Source link
