Nvidia on Monday introduced a new software platform called Lepton aimed at creating a centralized marketplace for cloud-based access to its graphics processors, as demand for artificial intelligence computing power continues to surge.
The Lepton platform will allow cloud computing companies to sell GPU capacity in a single location, streamlining access for developers seeking Nvidia’s widely used chips for training AI models.
Nvidia GPUs dominate the AI training market, and a growing number of specialized cloud providers, known as “neoclouds”, such as CoreWeave and Nebius Group have emerged to rent out GPU capacity. Both CoreWeave and Nebius are among the initial partners on the Lepton platform, along with Crusoe, Firmus, Foxconn, GMI Cloud, Lambda, Nscale, SoftBank Corp, and Yotta Data Services.
Alexis Bjorlin, vice president of Nvidia’s cloud business, said the current method for securing GPU capacity is inefficient and fragmented. “It’s almost like everyone’s calling everyone for what compute capacity is available,” Bjorlin told.
“We’re just trying to make it seamless, because it enables the ecosystem to grow and develop, and it enables all of the clouds, the global clouds and the new cloud providers, access to Nvidia’s entire developer ecosystem.”
Major cloud providers such as Microsoft, Amazon Web Services, and Google Cloud have not yet joined the Lepton platform. Bjorlin noted, however, that the system is open to them should they choose to participate.
Over time, Lepton will offer developers the ability to search for Nvidia GPUs located in specific countries to comply with data residency rules. The platform will also support companies that already own Nvidia GPUs in locating additional capacity to rent.
“It’s a good move for them,” said Mario Morales, group vice president at research firm IDC. “Nvidia is close to about 5 million developers. So they want to figure out a way where they can give them (developers) more access to the technology.”
Nvidia has not disclosed the business model for Lepton, including whether it will collect fees or commissions from developers or cloud providers. Bjorlin said developers will continue to maintain their direct contractual relationships with the underlying compute providers.
