(San Jose) The American semiconductor specialist Nvidia on Monday unveiled its latest models of electronic chips intended to support the artificial intelligence (AI) revolution, of which it seeks to consolidate its position as a crucial supplier.
“We need bigger GPUs. So, ladies and gentlemen, I would like to present to you a very, very large GPU,” announced Jensen Huang, the group’s boss, during a conference for developers.
GPUs (graphics processing units) are chips with a computing capacity much greater than that of traditional microprocessors (CPU).
“I hope you realize this isn’t a concert, it’s a conference for developers,” Mr. Huang joked as he took the stage at the SAP Center in San Jose, Calif., a room of 18 000 seats usually hosting hockey games and concerts.
The event, dubbed the “Woodstock of AI” by Wedbush analyst Dan Ives, has become a must-see on the calendar of major new technology players due to Nvidia’s unique role in the AI revolution that has known the world since the appearance of ChatGPT at the end of 2022.
Nvidia’s powerful GPU chips, and its software, are one of the main ingredients of generative AI.
Its rivals, like Intel and AMD, are still struggling to match the power and efficiency of its flagship product, the H100, released in 2022.
Apple, Microsoft and Amazon have also developed chips with AI in mind, but they are, for now, forced to use Nvidia’s products to deliver on their AI promises.
Stock market highs
This leading position has propelled Nvidia’s stock to stock market heights, with approximately +250% over one year and even +80% since the start of 2024.
The Santa Clara group has overtaken Amazon in terms of market capitalization, arriving just behind Microsoft and Apple.
On February 23, it even crossed the symbolic threshold of 2,000 billion US dollars in valuation, an altitude that only Microsoft, Apple and the oil company Saudi Aramco have known.
But Nvidia does not intend to stop there and announced on Monday the release of an even more powerful processor, accompanied by ad hoc software, on a platform called “Blackwell”. In tribute to David Blackwell, the first black academic to join the National Academy of Sciences.
Blackwell GPUs must provide “superchips” for AI, four times faster than those of the previous generation used to train existing AI models, according to Nvidia.
“The rate of progress in computing is insane,” said Mr. Huang, whose group originally focused on video game graphics.
The group also claims that its “superchips” will be 25 times more efficient in terms of energy efficiency.
A preponderant element, because AI is strongly criticized due to its very heavy consumption of electricity compared to conventional systems.
Unlike its rivals Intel, Micron and Texas Instruments, Nvidia – like AMD – does not manufacture its own semiconductors, but uses subcontractors, such as Taiwan Semiconductor Manufacturing (TSMC).
Given geopolitical tensions between Taiwan and mainland China, this could be a weak point, and the United States has banned Nvidia from supplying its highest-performance chips to Chinese companies.
Gr00t
Nvidia announced other innovations in the field of artificial intelligence on Monday.
The Gr00t project — whose name is not inspired by a character from the Guardians of the Galaxyassures Nvidia — is described as “the world’s first model of human foundation”.
Robots equipped with Gr00t will be designed to understand what people say and imitate their movements, to learn from their own experience about how to interact with the world, Nvidia says.
The models “will allow the robot to learn from a few human demonstrations so that it can help with daily tasks and reproduce human gestures, just by observing us,” continues the group.
Nvidia is cooperating with Apple to equip the Apple firm’s virtual reality headset, Vision Pro, with AI, which has been on the market since the beginning of February. This collaboration comes at a time when Apple is under pressure to prove that it is not ahead of Amazon, Google, Meta and OpenAI in terms of AI.
Additionally, Nvidia unveiled Earth-2 Cloud, a platform for predicting climate change using simulations generated by AI supercomputers.