The market for artificial intelligence (AI) accelerators, which provide the incredible computational horsepower necessary to train and run powerful AI models, is dominated by Nvidia. A combination of best-in-class hardware and a software platform that locks in developers has made it next to impossible for competitors to make much headway.
Nvidia is almost certain to lose some of its dominance as competitors flood the market with alternatives, and as cloud computing giants design custom AI chips for their data centers. However, Nvidia's head start makes it unlikely the company will be dethroned in the near future.
That leaves rivals Intel (INTC -0.74%) and Advanced Micro Devices (AMD -1.52%) fighting for second place. AMD has made more progress than Intel so far. The company expects to generate around $4 billion in AI chip sales this year, driven by its recently launched Instinct MI300 series accelerators. Meanwhile, Intel expects its latest Gaudi 3 chips to produce just $500 million in revenue, although that number may not include revenue from older Gaudi 2 chips.
Betting on software
Despite Intel's lack of progress relative to AMD, the company is confident that it will emerge as the runner-up in the AI chip market. Speaking to Reuters, Intel CTO Greg Lavender recently talked up the company's software efforts and noted that Gaudi 3 chips were seeing "lots of demand." Lavender expects Gaudi 3 to help Intel claim second place behind Nvidia.
Lavender was previously at VMware, the software company that current Intel CEO Pat Gelsinger helmed for nearly a decade before returning to Intel to lead the company's turnaround. It's not surprising, then, that software has become a big focus for the company. Lavender is targeting $1 billion in software and developer cloud subscription revenue by 2027, partly by providing AI services.
On top of selling software and services to its customers, Intel is betting that open-source software will help it chip away at Nvidia's lead and make its own AI chips more attractive. Intel is part of the UXL Foundation, which is working to build an open software ecosystem for accelerators based on Intel's oneAPI technology.
The idea is to provide a software ecosystem that enables building software that can seamlessly run on accelerators of different types and from different vendors. Widespread adoption would benefit Intel, which, in addition to selling its Gaudi line of AI accelerators, provides data center GPUs and AI capabilities built into its Xeon server CPUs.
Intel, along with AMD and others, are also contributing to OpenAI's Triton initiative. OpenAI introduced Triton, an open-source programming language for GPUs aimed at making it easy to write efficient code, back in 2021. Intel's GPUs already support Triton and support is coming for its AI chips.
Another way to win
AMD is working on its own software initiatives, notably ROCm. ROCm is an open-source stack of software tools and libraries that's been around for years. While the effort is open source, the only officially supported GPUs come from AMD.
Whichever software effort ultimately gains enough traction to take off, the universal goal is to break Nvidia's stranglehold on the market. Once the playing field is leveled, the hardware itself will be the main selling point.
While Intel will face plenty of competition from AMD and others as it guns for second place in the AI chip market, the company can benefit even if it comes up short thanks to its foundry business. Intel is aiming to become the second-largest foundry by 2030, and the next major step in its plan is the launch of the Intel 18A process early next year. With Intel 18A, the company expects to have the most advanced process technology in the industry.
As cloud computing giants increasingly design in-house AI chips, Intel will have the opportunity to manufacture and package those chips. Intel has already signed a deal with Microsoft to use the Intel 18A process for the undisclosed custom chip. That may be an AI accelerator, a data center CPU, or something else entirely.
The silver medal in the AI chip race is very much up for grabs. While Intel is off to a slow start, its Gaudi 3 chips and future AI chips could eventually bring in billions of dollars in annual revenue. Add in additional foundry revenue from manufacturing third-party AI chips and incremental software revenue, and there's a lot to like about Intel's long-term AI prospects.
