A consortium led by Google, Intel, and Qualcomm aims to break Nvidia's stranglehold on software for AI chips with a new open-source platform. The goal is to move AI development away from Nvidia's proprietary CUDA platform.
Nvidia is best known for its AI hardware. But a big reason for its dominance in chips is its CUDA API. Nvidia CEO Jensen Huang introduced it to the market in 2007, and it has been the standard for more than a decade.
More than four million developers worldwide use CUDA to create AI and other applications. Even if a company could build comparable or better AI chips, it would be impossible to catch up to this software infrastructure without massive investment.
For good reason, Nvidia manager Manuvir Das recently described Nvidia as "80 percent a software company" that makes the most of its hardware for AI training and inference.
Chip giants want to establish an alternative to Nvidia Cuda
The UXL coalition of technology companies, including Arm, Qualcomm, Google and Intel, wants to break this dominance. They want to focus their resources first on the most pressing computing problems, such as the latest applications in artificial intelligence and high-performance computing.
"We're actually showing developers how you migrate out from an Nvidia platform," said Vinesh Sukumar, Qualcomm's head of AI and machine learning, in an interview with Reuters.
The starting point is Intel's OneAPI technology. Building on this, the companies plan to develop a set of open source software tools through the UXL Foundation that can be used to address AI accelerators from different manufacturers. The goal is to create an open ecosystem that promotes productivity and hardware choice.
The UXL Foundation's Technical Steering Group plans to define the technical specifications in the first half of the year and have them ready by the end of the year. The goal is to create a stable foundation that can handle code from different companies and run on any hardware.
In addition to the founding members, cloud providers such as Amazon and Microsoft, as well as other chipmakers, will be involved. UXL was launched in September 2023 and will have long-term support for Nvidia hardware and software.