n Tuesday, Jensen Huang stated that Nvidia might just not be ready to expose a new chip technology that aims to cut energy usage, saying that it is
not yet reliable enough to be integrated into the flagship graphics processing units (GPUs).
Co-packaged optics use beams of laser light to send information on fiber optic cables between chips, creating connections faster and with superior energy efficiency compared to traditional copper cables.
Even more so, during one of the keynote addresses at Nvidia’s annual conference, which was held on Tuesday at a packed hockey stadium in San Jose, California, Nvidia’s CEO also added that his company would use the co-packaged optical technology in two new networking chips that sit in switches on top of its servers. He also added that the integration of new technology would also make the chips that are being used three and a half times more efficient than its previous models.
The switch chips will also come out later this year and extend into 2026 as theri step towards advancing the technology.
Yet, Huang told a group of journalists after his speech that even though Nvidia examined the possibility of using it more widely in its flagship GPU chips it had no current plans to do so, due to traditional copper connections that were “orders of magnitude” and even more reliable than today’s co-packaged optical connections.
Subscribe to our newsletter
While directly using optical connections between GPUs, Huang said “That's not worth it,” also adding “We keep playing with that equation. Copper is far better.". More so, Nvidia’s CEO reported that he has focused on providing a reliable product roadmap that Nvidia’s customers, such as OpenAI and Oracle could have the possibility to anticipate.
Huang said "In a couple of years, several hundred billion dollars of AI infrastructure is going to get laid down, and so you've got the budget approved. You got the power approved. You got the land built," also adding “What are you willing to scale up to several hundred billion dollars right now?".
Currently, Nvidia has a flagship of 72 products containing their chips in a single server, consuming 120 kilowatts of electricity. It is also worth mentioning that they are also generating heat that needs to be cooled with the help of a liquid cooling system that is similar to one that can be found in a car engine. It is also worth mentioning that the flagship server that will be ready for release in 2027 will also pack hundreds of its Vera Rubin Ultra Chips into a single one, also consuming 600 kilowatts of power.
