Hem
AI-kapplöpningenFördjupning

Nvidias vd: ”AI kräver hundra gånger mer datorkraft”

CEO Jensen Huang talks during the keynote address of Nvidia GTC Tuesday, March 18, 2025, in San Jose, Calif. (Nic Coury / AP)

Världen har gravt underskattat AI:s behov av datorkraft. Det sa Nvidias toppchef Jensen Huang under en fullsatt konferens i veckan – ett event som han själv kallade ”Super Bowl för AI”.

– Mängden datorkraft som behövs är utan tvekan hundra gånger större än vi trodde vid den här tiden förra året.

För att möta efterfrågan lanseras Rubin, en ny chipserie som ska vara upp till fjorton gånger kraftfullare än nuvarande Blackwell. Nvidia meddelade också samarbeten med GM och Disney, skriver Wall Street Journal.

The Wall Street Journal

Nvidia CEO Says AI Computing Needs to Surge 100-Fold

Jensen Huang announces new Rubin AI chips, corporate partnerships in bid to shore up investor confidence in AI boom.

By Steven Rosenbush and Isabelle Bousquette

The Wall Street Journal, 18 March 2025

Nvidia Chief Executive Jensen Huang tried to quell investor concerns about the artificial intelligence boom Tuesday at an event he dubbed “the Super Bowl of AI.”

The world will need 100 times more computing power for advanced AI than it considered necessary a year ago, he said at a conference that has grown from a sleepy chip developer gathering in Silicon Valley into a rock-concert-like event that filled a hockey arena.

AI expansions into models that can “reason” and act as “agents” to carry out tasks for humans will require far greater computational firepower, which comes from the kinds of chips that Nvidia manufactures.

“This last year, this is where almost the entire world got it wrong,” he said. Nvidia shares have been volatile since the release of DeepSeek in January, a model released by a Chinese startup that said it had built sophisticated AI models that required fewer of Nvidia’s chips. That led some investors to question whether future AI systems will need fewer Nvidia chips for training and day-to-day operations.

CEO Jensen Huang talks during the keynote address of Nvidia GTC, March 18, 2025, in San Jose, Calif. (Nic Coury / AP)

So-called reasoning models spend more time “thinking” about a problem before delivering an answer. They break each prompt down into steps, a process that is best used for complex problems, according to companies building those models. They might generate so much data that computing speeds will need to increase to process the data quickly for users. Users won’t want to wait 10 times longer to get an answer that relies on 10 times more data, he said.

Huang also spent time hyping up AI agents, which are designed to complete tasks—such as booking a flight or making a dinner reservation—for humans. Agents, which often rely on reasoning models, require a lot more computational power, which means sustained or even higher demand for Nvidia’s chips, Huang said.

“The amount of computation needed is easily 100 times more than we thought we needed at this time last year,” Huang said, reiterating a point he has made in recent months.

“Everyone, pay attention. This could very well be the largest industry of all”

Jensen Huang, CEO of Nvidia

He unveiled a new line of even more powerful chips called Rubin that are expected to be available in the second half of 2026. The AI superchip, named after the astronomer who discovered dark matter, is set to have more than three times the computing capabilities of the Ultra chips. An Ultra version of the Rubin will offer 14 times the performance of the Blackwell, he said.

The Rubin will need to be a monster hit to meet sky-high investor expectations for the chip maker. Analysts expect this series of Nvidia’s chips to generate nearly $40 billion in revenue in their first year of sales and more than $95 billion in their second year, according to consensus estimates from Visible Alpha. That is more than the annual revenue of nearly three-quarters of companies on the S&P 500.

Nvidia shares were down by more than 3% after the presentation concluded, slightly more than the tech-heavy Nasdaq Composite, where there was a broad tech selloff along with declines in the S&P 500 after turmoil returned in the Middle East.

(Nic Coury / AP)

The company also announced a partnership with General Motors to help provide underlying technology for its autonomous driving fleet, as well as a partnership with Disney and Google’s DeepMind called “Newton” that can simulate physical systems.

“The time has come for robots,” Huang said, noting there is a growing shortage of human labor. This is an area where Nvidia can build on its legacy as a gaming company, which demanded it create virtual worlds in which objects and people behaved as they would in the real world.

With a “slow-thinking” capability, Nvidia’s physical AI can help robots perceive and reason about their environment, and fast-thinking capability allows it to take actions.

“Everyone, pay attention. This could very well be the largest industry of all,” Huang said.

Before the keynote began, Huang took the stage in his trademark leather jacket and spent time ramping up the crowd, which had helped push up San Jose hotel prices up to $2,500 a night as the city was festooned in Nvidia colors.

Restaurants and coffee shops will earn their entire month’s rent over the course of a few days, San Jose Mayor Matt Mahan said in an interview. Hosting the event will be a test for how the area handles the Super Bowl and World Cup next year.

Omni är politiskt obundna och oberoende. Vi strävar efter att ge fler perspektiv på nyheterna. Har du frågor eller synpunkter kring vår rapportering? Kontakta redaktionen