Nvidia (NVDA +2.59%) has had a well-deserved rally over the past year as the outright leader in artificial intelligence chips.
How does Nvidia continue to maintain a near-stranglehold on the AI chip market? A big reason has to do with CUDA, its full-stack software library, with software defined kits and proprietary applications that enable developers to program graphics processing units (GPUs) for parallel computing.
Having developed CUDA in 2006, Nvidia has a multiyear head in AI chip programming, having developed a nice network effect over that time. Developers have learned to program GPUs only through CUDA for 15 years, and companies have invested vast sums building their own AI software stacks on top of CUDA.
That 17-year lead would be pretty tough to crack for any potential challenger.
But eight challengers, working together? One such group of tech leaders is taking direct aim at CUDA's moat and expects to begin breaching it by the end of this year.
NASDAQ: INTC
Key Data Points
UXL: An Nvidia-killing consortium?
The new open-source consortium is called the Unified Acceleration Foundation, or UXL, and consists of eight of the world's largest AI chip and software companies.
The consortium is led technologically by Intel (INTC +1.07%) but also contains a slew of other heavyweights as steering members, including Alphabet (GOOG 0.20%) (GOOGL 0.41%), Samsung, Qualcomm (QCOM +0.24%), Broadcom (AVGO +4.69%)-owned VMware, Arm Holdings (ARM 0.68%), Fujitsu (FJTSY 1.80%), and privately held chip licensing company Imagination Technologies.
The group is focused on developing an open-source software that can work with any type of AI chipset, which would obviously be beneficial to Nvidia challengers like Intel. But it would also be highly beneficial to cloud platforms like Alphabet, which have to buy those very expensive and high-margin Nvidia chips. After all, Nvidia made a ridiculous 58% adjusted (non-GAAP) net margin in its recent quarter. Who's paying those exorbitant prices? Basically, everyone. Hence, the concerted effort to break Nvidia's near-monopoly on AI chips and lower prices across the industry.
The latest from UXL: We'll have a CUDA disruptor by 2025
Both Reuters and The Register recently interviewed members of UXL, which claims it will have a mature CUDA rival ready for release later this year. According to UXL Steering Committee Chair Rod Burns in an interview with The Register:
[T]he specification has been under development for a few years and has been released regularly over that time. This means we already have a mature specification for many of the fundamentals needed. ... This work continues through 2024 through refinement and is led by the Specification Working Group within the foundation; we will be aiming to ratify a spec release in Q4.
If that seems like a very quick time-to-market given that Nvidia's CUDA began in 2006, it should be noted this initiative has really been in the works for several years, mostly under Intel's homegrown open-source software initiative called OneAPI.
Intel announced OneAPI in 2018, and the 1.0 incarnation of this open-source software stack was released in 2020. Today, UXL is essentially an extension of OpenAI, only with more collaborators outside Intel contributing to ramp up the speed of development.
Image source: Getty Images.
The OneAPI initiative grew out of a need to unify all of Intel's different kinds of chips under a simple, easy-to-use API layer, especially after Intel's 2015 Altera acquisition. Unifying Intel's sprawling portfolio of chips under one open-source software stack made a lot of sense at the time. Like OneAPI, the UXL open source software stack aims to create a programming language not only for all third-party GPUs, including Nvidia GPUs, but also for CPUs, field programmable gate arrays, and other accelerators.
Of course, OneAPI's democratizing vision makes a lot more sense today as an industrywide phenomenon, especially as Intel has fallen behind Nvidia in the AI transition.
But can Nvidia be caught?
At first, a 15-year lead combined with Nvidia's lighting-fast hardware innovation may seem like an insurmountable obstacle for any would-be challenger to overcome. After all, around 4 million developers have built applications around Nvidia's CUDA for 15 years or so. That won't be easy to break.
However, if eight of the top tech companies in the world team up in a concerted push for a number of years, with significant financial resources dedicated to the venture? It's risky, but it's possible that could one day yield a viable CUDA alternative. And that day may come in late 2024.
That's not to say Nvidia's competitive advantage would go away entirely, even if UXL finds success. After all, UXL would still program Nvidia GPUs along with others, and Nvidia currently has the best AI hardware engine on the market today, on top of its software.
But should UXL show success by the end of this year, it's possible Nvidia's market share may go from "dominant" to maybe just being a "strong leader," and margin may come down from "insanely high" to just "high."
Yet with Nvidia's valuation soaring to begin the year and a lot of good news already priced into the stock, UXL's progress is definitely a risk worth monitoring for Nvidia shareholders throughout 2024.
