Nvidia leads the AI hardware market, but as the generative AI market grows and matures, there is a push for more competition in AI hardware and software.
Nvidia has become a key figure in the GenAI hardware market thanks to its chips, which power many of the GenAI tools and applications produced today.
The company partners with the biggest cloud providers -- Google, Microsoft and AWS. Recently, Facebook parent company Meta revealed plans to purchase 350,000 Nvidia H100 GPUs by the end of the year.
However, a foundation consisting of Nvidia's competitors says the company's AI dominance is a problem for developers and aims to challenge its leadership, Reuters reported this week.
UXL Foundation
The Unified Acceleration (UXL) Foundation, part of the Linux Foundation, emerged in September 2023. Its goal is to prevent AI developers from building apps specifically for one vendor's GPUs, and to instead foster the use of open standards and open source software to build applications for multivendor, multiarchitecture ecosystems. Its members include Arm, Intel, Qualcomm, Samsung, VMware and others.
UXL members say the increasing dependence on GPUs creates a challenge for software developers in that they have to build apps for specific architectures. Using open standards would free developers from having to "think about what type of CPU they are compiling for when writing software," according to a February blog post by Rod Burns, vice president of ecosystem at Codeplay Software and a member of the UXL steering committee.
Developing software for heterogeneous AI architectures, the group says, starts with Intel's OneAPI, an open standard technology that supports building apps across different accelerators, including GPUs, CPUs and FPGAs.
More options in the market
While it seems UXL's strategy targets Nvidia's leadership, it's about providing more options to the market, according to industry experts.
"The pie is getting bigger," independent AI analyst Mark Beccue said. "There's only so much that Nvidia can do."
Nvidia has gained momentum as one of the first to build GPUs for different purposes. It also solidified its success with CUDA, a platform and programming model for GPUs.
In contrast, competitors AMD and Intel have found it difficult to play catch-up because the GPUs they create do not have a set standard like CUDA that developers can use to run their software.
A more diverse ecosystem is probably good for competition, and it gives a lot more flexibility and opportunities and options for companies that are getting into AI to build their solutions.
Olivier BlanchardAnalyst, Futurum Group
"You've got all these companies making GPUs, and there isn't need in the marketplace for an open standard to use so that whoever is using any of these GPUs doesn't get tied in to a proprietary software to run on," Beccue said.
Due to this, Nvidia has been able to drive ahead in the AI market the way others have not.
With UXL, there is now a way to develop an open standard that will help Nvidia's competitors provide alternatives in the market while filling in some gaps.
"A more diverse ecosystem is probably good for competition, and it gives a lot more flexibility and opportunities and options for companies that are getting into AI to build their solutions," said Olivier Blanchard, an analyst at Futurum Group.
Open source and GenAI
Providing competition through an open source and open standard approach makes sense. However, UXL is certainly not the first group to champion this approach.
In 2008, the Khronos Group introduced OpenCL, a framework for writing programs across heterogeneous platforms that include CPUs, GPUs and digital signal processors. OpenCL failed to be as successful as CUDA, even though it was open source.
For UXL to choose an open source and open standardization route also speaks to the rise and growth of open source.
"Open source and standardization then put the onus back on the performance of that piece [of the architecture]," Beccue said. Users will not necessarily focus on the proprietary software, but rather on the performance of the entire architecture, he added.
"They're copying what other open source communities have done in the past," said Alvin Nguyen, a Forrester Research senior analyst. "They're following the footsteps because it has worked."
While it could take time to get the audience or the user base, open source has worked for many technology providers. Moreover, an open and more competitive market is good for the GenAI market and users -- particularly when there are supply constraints.
"There's just not enough general-purpose CPUs and GPUs to go around for people to determine what they can and cannot do," Nguyen said.
In addition, more competition in the market pushes innovation, Blanchard said.
"It's going to raise the bar both on quality and on accessibility," he said. "If it puts a little bit of pressure on Nvidia to innovate faster and better, that's also good."
Having more companies compete to come out with what's best or accessible is a net positive to the developer ecosystem, Blanchard added.
Esther Ajao is a TechTarget Editorial news writer and podcast host covering artificial intelligence software and systems.
