In the fast-paced world of technology investment, one question looms large: Can Nvidia sustain its incredible growth and high margins? Over the past two quarters, Nvidia (NASDAQ: NVDA) has displayed remarkable performance, but doubts persist about its sustainability. If artificial intelligence (AI) accelerators continue to grow at a 50% annualized rate over the next five years, Nvidia’s stock could still be considered a bargain. However, if rival forces begin to erode its AI dominance, its sky-high price-to-earnings (P/E) ratio might not be justified.
At a recent industry conference, Lisa Su, the CEO of Advanced Micro Devices (NASDAQ: AMD), a key competitor to Nvidia, expressed skepticism about the concept of moats in the rapidly evolving tech landscape. She stated, “I’m not a believer in moats when the market is moving as fast as this.”
This sentiment suggests that Nvidia’s current lead in the dynamic AI space may not be as secure as it seems, despite its multiyear head start in AI accelerator hardware and software development. But what’s happening beyond mere rhetoric? How are tech giants like AMD, Intel, and FAANG companies (Facebook, Apple, Amazon, Netflix, Google) planning to challenge Nvidia’s supremacy?
Nvidia’s CUDA moat: Real or perceived?
Many investors believe that Nvidia’s dominance in AI is not just due to its hardware innovations but also its CUDA software stack. CUDA was created to enable the programming of graphics chips for the parallel processing of regular data, making AI training and inference possible.
Software moats can be formidable, as seen with Microsoft’s Office suite, which includes PowerPoint, Excel, and Word. Once it became the standard for business operations, it became challenging for competitors to introduce a competitive product. This phenomenon is known as the network effect.
However, Nvidia’s CUDA might be more vulnerable to disruption than Microsoft Office. The prohibitive cost of Nvidia’s GPUs, which currently sell for $30,000 or more per chip, creates a strong incentive for large cloud platforms and AI customers to seek competitive alternatives. In contrast, Microsoft Office is relatively affordable for enterprises.
Moreover, AMD and Intel, along with tech giants like Meta Platforms, Alphabet, and Microsoft, are actively contributing to open-source alternatives. These massive companies possess significant developer resources and are well-positioned to create viable multichip platform alternatives for the AI era.
We are still in the early stages of the AI boom, which began in earnest just a year ago with the introduction of OpenAI’s ChatGPT. If these competitors move swiftly, a robust open-source competitive platform could emerge before Nvidia’s moat solidifies further.
RocM and SYCL: Competing with CUDA
Both Intel and AMD have presented their CUDA alternatives at recent AI and data center chip presentations. They emphasize the benefits of open platforms that allow their in-house software to be ported to different GPUs while integrating with existing open-source AI software.
Prominent open-source platforms like Pytorch (Meta), Tensorflow (Alphabet), Deepspeed (Microsoft), and Hugging Face (AI startup) are prime examples of this approach.
What makes AMD’s and Intel’s software stack intriguing is their portability features. These features enable software developers to migrate programming code written in CUDA to their platforms with minimal recoding.
– AMD’s software stack, ROCm, is “mostly open” and optimized for Pytorch and Hugging Face. Importantly, it includes a porting feature for code from other GPUs, likely including Nvidia and CUDA.
– Intel promotes an open-source AI programming platform called SYCL, developed by the Khronos Group. SYCL, a higher-level open-source C++ software, allows developers to write code for any accelerator.
– Intel also released SYCLomatic, a tool that facilitates porting over 90% of CUDA code to SYCL with only minor tweaks needed.
If there’s no moat, It’s a hardware battle
While Nvidia has a substantial lead in AI chips, AMD recently unveiled its MI300, featuring a “chiplet” architecture with significant capabilities. Intel’s Gaudi line of AI accelerators has also gained traction, attracting high-profile generative AI startup Stability AI. These competitors will undoubtedly invest heavily in the AI accelerator market, given its rapid growth.
The case for Nvidia maintaining its lead in the AI market hinges on the network effects of CUDA. Hardware superiority can be fleeting, as Intel experienced when it lost its lead in advanced chips around five years ago. Therefore, the AI market could potentially accommodate all three companies.
Investors, particularly Nvidia shareholders, must closely monitor the AI software competition, as it could determine whether the company continues its dominant growth and high margins or experiences more industry-standard margins in the 20%-30% range historically seen in leading-edge processors.
As the tech industry grapples with the rapidly evolving AI landscape, Nvidia faces challenges from formidable competitors like AMD and Intel, backed by FAANG giants. While Nvidia’s CUDA software stack has provided a significant advantage, it is not impervious to disruption.
The emergence of open-source alternatives like ROCm and SYCL, coupled with portability features, signals a concerted effort to challenge Nvidia’s AI dominance. While Nvidia’s hardware lead is evident, the battle for AI supremacy may ultimately depend on the adaptability of software and the ability to win over developers.
In this fast-paced arena, where technology evolves by the minute, Nvidia, AMD, and Intel will continue to vie for a share of the burgeoning AI market. Investors must remain vigilant, as the outcome of this competition will have a profound impact on the future of AI technology and the companies driving it.
