On Thursday local time, Nvidia CEO Jensen Huang acknowledged the long-term potential of orbital computing on the All-In podcast. "Of course we should start from the ground first, because we are already here," Huang said, but he also pointed out that it is still crucial to prepare for the construction of space infrastructure.
The concept is gaining traction as artificial intelligence workloads proliferate, prompting technology companies to explore alternatives to traditional energy-hungry data centers.
Huang Renxun pointed out that the cooling system is one of the biggest problems faced in building on-orbit data centers. On Earth, systems rely on conduction and convection to lose heat. But in space, these methods don't work.
"You can only use radiation," Huang said, adding that this method requires a "very large cooling area" to release the heat - which makes the system complex and expensive.
While not entirely impossible, he said the puzzle would take years to resolve.
While space has abundant solar resources and is vast, the cost of launching hardware and building infrastructure remains a significant obstacle.
Despite facing many challenges, Huang Renxun said that Nvidia has taken initial steps in the field of space computing.
“We’re already here,” he said, noting that the company has deployed CUDA-based systems on satellites for imaging and AI processing tasks.
The NVIDIA CUDA system allows developers to use NVIDIA graphics cards to do more than just graphics processing - it can also handle complex computing tasks faster than ordinary processors.
Processing the data directly in space (rather than sending it back to Earth) is the logical next step, Huang added. "This type of work should be done in space," he said.
“It’s going to take years,” Huang said. "It doesn't matter, I have plenty of time."
Earlier in Nvidia’s fourth quarter earnings call, Huang Jensen elaborated on his views on space data centers.
"The current economic benefits of space data centers are not ideal, but as time goes by, the situation will improve." He said.
Although cautious about the short-term prospects, Huang noted that space provides "abundant energy" and "ample space" for solar-powered AI satellites.
At the recent GPU Technology Conference, Huang highlighted the potential of orbital data centers, while Nvidia also released its Space-1 Vera Rubin module.
Huang added that Nvidia's THOR chip has passed "radiation certification," noting that Nvidia is already using satellites for image processing.