The hockey stadium in San Jose, California was packed yesterday as if it were the Hockey World Cup. In front of an audience of software developers from all over the world, a chip game was played instead, with the US champion, the CEO of Nvidia Jensen Huang, which has tried to respond, from a distance, to the attacks of its Chinese adversary DeepSeek, followed by his compatriot AlibabaAt stake are billions of investment in the US technology sector, represented by Magnificent Seven.
THEthe anticipation was high after Chinese competitors announced in January that they were ready with a technology that requires much less investment. But Juang has not entirely succeeded, with its new products, in convincing the public, both developers and Wall Street, not even investors: Nvidia shares lost 3,50% yesterday.
During his keynote speech at the annual conference Huang said yesterday that his company will use the optical technology packaged in two new network chips installed in switches above its servers, explaining that this technology will make chips three and a half times more efficient from an energy standpoint compared to their predecessors. But what Hang sees as a revolution needs longer times in the rush the markets are in: switch chips will be available between the end of this year and 2026, he reports Reuters. Huang told a group of reporters after his talk that while Nvidia has explored using it more widely in its flagship GPU chips, it has no plans to do so at this time because traditional copper connections are “orders of magnitude” more reliable than today’s packaged optical connections. “It’s not worth it,” Huang said of using optical connections directly between GPUs. “We keep playing with that equation. Copper is much better".
Silicon Valley entrepreneurs and investors have placed their hopes in optical technology
Huang then said that he was focused on providing a reliable product roadmap that Nvidia customers, such as OpenAI and Oracle, they might prepare. “In a couple of years, several hundred billion dollars of AI infrastructure will be allocated, and so you have the budget approved. You have the energy approved. You have the land built,” Huang said. “What are you willing to do right now?”
Silicon Valley entrepreneurs and investors have placed their hopes of optical technology, which they believe will be fundamental for the construction of ever bigger computers for the systems of artificial intelligence, which, Huang said yesterday, will still be necessary even after the progress of companies like DeepSeek, because artificial intelligence systems will need more computing power for process the answers.
Startup including Ayar Labs, Lightmatter and Celestial AI have raised hundreds of millions of dollars in venture capital, some of it from Nvidia itself, to try to put packaged optical connections directly onto AI chips. Lightmatter and Celestial AI are both aiming for public offerings. Copper connections are cheap and fast, but they can only carry data a few meters at most. While this may seem trivial, it has had a huge impact on Nvidia’s product lineup over the past five years.
The new Rubin family introduced
L'current flagship product from Nvidia contains 72 of its chips in a single server, consuming 120 kilowatts of electricity and generating a there was heat to request a system of cooling liquid-like that of a car engine. The flagship server unveiled yesterday for a 2027 launch will contain hundreds of its own chips True Rubin Ultra in a single rack and will consume 600 kilowatts of power.
Insert more than double the number of chips in the same space in two years will require enormous engineering feats by Nvidia and its partners. Such efforts are driven by the fact that AI computing work requires moving a lot of data back and forth between chips, and Nvidia is trying to keep as many chips as possible within the relatively short range of copper connections.
Mark Wade, CEO of Ayar Labs, which has received venture support from Nvidia, said the chip industry is still trying to figure out how to produce co-packaged optics at lower costs and with mechanical greater reliability. While the transition could Don't arrive before 2028 or beyond, Wade said, the chip industry will have no choice but to abandon copper if it wants to continue building ever-larger servers. “You just have to look at the power consumption going up and up on racks with electrical connections,” Wade told Reuters in an interview on the sidelines of Nvidia's conference. "Optics is the only technology that gets you off that train."
