A group of major technology companies is trying to loosen one of Nvidia's strongest advantages in AI: not the chips themselves, but the software layer that developers use to make those chips useful.
The effort centers on UXL, an open-source platform backed by companies including Google, Intel, Qualcomm and Arm. Its stated aim is to give developers a practical route away from Nvidia CUDA while still supporting demanding work in artificial intelligence and high-performance computing.
Why CUDA matters so much
Nvidia is widely associated with AI hardware, but the source of its strength is also software. CUDA, the API introduced by Nvidia CEO Jensen Huang in 2007, has been the standard for more than a decade.
That history matters because software habits become infrastructure. More than four million developers worldwide use CUDA to build AI and other applications. For any rival chipmaker, competing with Nvidia is therefore not only a hardware challenge. It also means competing with a large, established developer ecosystem.
The source article points to the core problem: even a company with comparable or better AI chips would face a steep climb without massive investment in the surrounding software environment. In AI training and inference, the tools around the chip can shape what developers choose, how quickly they work, and which hardware they can realistically adopt.
That is why Nvidia manager Manuvir Das recently described Nvidia as "80 percent a software company". The phrase captures the strategic point: Nvidia's position depends on making its hardware accessible and productive through software, not simply on selling processors.
What UXL is trying to change
The UXL coalition wants to create an alternative path. Its members include Arm, Qualcomm, Google and Intel, and the work is being organized through the UXL Foundation.
The first focus is on urgent computing needs, including the latest applications in artificial intelligence and high-performance computing. Rather than asking developers to stay inside one proprietary platform, the coalition wants open-source tools that can address AI accelerators from different manufacturers.
Vinesh Sukumar, Qualcomm's head of AI and machine learning, described the practical goal in an interview with Reuters: "We're actually showing developers how you migrate out from an Nvidia platform."
That migration point is central. The challenge is not only to create software in the abstract, but to make the move away from CUDA understandable enough for developers who already depend on Nvidia's platform. If UXL can make that transition clearer, it could reduce one of the largest barriers facing non-Nvidia AI chips.
Intel OneAPI is the starting point
The project begins with Intel's OneAPI technology. Building from that base, the participating companies plan to develop a set of open source software tools that work across AI accelerators from different manufacturers.
The goal is an open ecosystem built around productivity and hardware choice. In plain terms, the coalition wants developers to be less locked into one vendor's software stack when choosing where to run AI workloads.
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 intended result is a stable foundation that can handle code from different companies and run on any hardware.
That ambition is broad. A common software base would need to support contributions from multiple companies while remaining useful across different hardware designs. The source article does not say how quickly developers will adopt it, but it makes clear that the coalition is aiming at the software layer where Nvidia has built a long-term lead.
Who else is involved
Beyond the founding members, the effort is expected to involve cloud providers such as Amazon and Microsoft, along with other chipmakers. That matters because cloud providers are part of how many developers access AI hardware in practice.
UXL was launched in September 2023. It will also have long-term support for Nvidia hardware and software, which suggests the project is not positioned only as a clean break from Nvidia. Instead, it is being framed as an open foundation that can include Nvidia compatibility while expanding choices beyond Nvidia's proprietary CUDA platform.
The companies behind UXL are therefore pursuing two related goals:
- Reduce dependence on CUDA: provide tools that help developers move out from an Nvidia-centered workflow.
- Support hardware choice: make it easier for AI accelerators from different manufacturers to be used through open source software.
- Focus on demanding workloads: prioritize artificial intelligence and high-performance computing, where software support is especially important.
The larger AI chip stakes
The push by Google, Intel, Qualcomm, Arm and others shows that AI chip competition is not just a race to build faster hardware. It is also a contest over the developer platform that sits above that hardware.
CUDA's advantage is that it has been used for years, has millions of developers, and has become a familiar path for building AI and other applications. UXL's challenge is to create an open-source alternative that is useful enough to make hardware choice feel realistic rather than theoretical.
If the coalition succeeds, developers could have a clearer route to AI accelerators from multiple manufacturers. If it struggles, Nvidia's software position may remain one of the hardest parts of its AI dominance to challenge.
For now, the most concrete next step is the technical specification work planned by the UXL Foundation's Technical Steering Group. The coalition wants those specifications defined in the first half of the year and ready by the end of the year. That timetable will shape whether UXL can become a practical foundation for open AI chip software, or whether CUDA continues to define the default path for developers.