With new software

Tech alliance aims to break Nvidia’s AI dominance

According to experts, Nvidia controls around 80 percent of the global market for high-performance AI processors.

(Bild: AFP)

The triumph of so-called Generative Artificial Intelligence is closely linked to Nvidia's high-performance processors. According to experts, the US company dominates around 80 percent of the global market. However, resistance is now forming: an alliance of top-class technology companies wants to break Nvidia's dominance.

However, it is not focusing on AI chips themselves, but on the software used to program them. "We are showing developers how they can break away from Nvidia's platform," says Vinesh Sukumar, Head of AI at semiconductor provider Qualcomm, in an interview with the Reuters news agency.

Nvidia can draw on almost 20 years of experience with its CUDA developer software. The program package is regarded as the company's secret weapon for customer loyalty, as it facilitates the programming of AI and other applications, provided that Nvidia chips are used. This business segment is an important source of revenue for the company. CUDA costs 4500 dollars per year and per Nvidia chip when used in a private data center or one dollar per hour for use by a cloud provider.

Competition from Qualcomm, Intel and Google
The UXL Foundation, which was founded last September and includes Qualcomm, the chip manufacturer Intel and the internet company Google, wants to change this with a competing product. The OpenAPI technology developed by Intel, which is to be expanded into a program package, serves as the basis. This open-source software, whose code is accessible to everyone, will work with AI chips from different manufacturers. "It's about creating an open ecosystem to promote productivity and hardware choice," explains Bill Hugo, Head of Technology for High Performance Computing at Google.

According to further information, the UXL experts want to finalize the technical specifications of the software in the first half of 2024. The details should be "fully developed" by the end of the year. UXL is dependent on information from numerous companies to ensure that the developer software supports as many available chips as possible. The alliance is therefore also courting cloud providers such as Amazon and Microsoft as well as other processor manufacturers. In the medium to long term, it should also be possible to use the UXL software for Nvidia chips.

Big competition
The UXL Foundation's project is just one in the battle for the biggest possible slice of the AI pie. In recent months, venture capitalists have invested a total of more than four billion dollars in almost 100 start-ups that also want to compete with Nvidia with AI developer software. This is the result of an analysis carried out exclusively for Reuters by the industry service Pitchbook.

However, experts predict that many of these start-ups will lose the battle against Nvidia. The success of the UXL Foundation is also questionable. This is because CUDA's already wide range of functions is constantly growing, partly thanks to suggestions from users. "But that's not the only thing that matters," emphasizes Jay Goldberg, head of the consulting firm D2D. "The key is the fact that people have been using CUDA for 15 years and have developed their own programs around it."