Investors have seen a tectonic shift in the semiconductor industry over the past few years, with Nvidia (NVDA 3.37%) rising in stature and Intel (INTC -0.58%) falling behind rivals for the first time in decades. Nvidia's leadership in GPUs enabled it to surpass Intel in terms of market cap back in 2020. But it wasn't until the recently completed second quarter that Nvidia surpassed Intel in revenue for the first time:
NVDA Revenue (Quarterly) data by YCharts
2023 has seen a major shift to artificial intelligence (AI) accelerators for training large language models, and away from personal computers and traditional servers where Intel had traditionally dominated. Furthermore, Intel fell behind Taiwan Semiconductor Manufacturing in terms of process technology a couple of years ago.
However, new Intel CEO Pat Gelsinger, who took over in 2021, has a turnaround plan as well as an agenda to get Intel its own piece of the AI pie and regain process leadership. With Intel down so much from its highs and with a market cap nearly one-tenth of Nvidia's, which is the better buy today?
Nvidia has a big lead and a software moat
Nvidia has risen to the top in the AI races because it's been innovating toward this goal for a long time. A leader in graphics processing units (GPUs), Nvidia was the first company to figure out graphics chips could be programmed to execute parallel processing of regular data needed for artificial intelligence.
Thus, Nvidia built a programming language and API library called CUDA, which enabled software developers to program GPUs to execute general purpose processing. Having begun development of CUDA nearly 20 years ago, the ecosystem consisting of both leading GPUs as well as programming software has given Nvidia a sizable moat in chips needed for AI.
Hence why Nvidia's revenue was up 101% from a year ago, its data center revenue up an even larger 171%, and why net margins have expanded to a whopping 50%.
Needless to say, Nvidia is in a great position to capture the massive growth of AI investment. Numerous industry leaders have now pointed to an AI accelerator market that will grow at a 50% annualized growth rate, from about $30 billion this year to $150 billion by 2027.
So while Nvidia trades at 47 times this year's earnings estimates and 29 times next year's estimates, if its growth continues at the rate of the industry until 2027, the stock may not be so expensive after all when looking to 2026 and 2027.
But Intel is aggressively pivoting to a turnaround
Of course, just because a company is doing better today doesn't necessarily mean its stock is the better buy. After all, Intel only trades at one-tenth the market cap of Nvidia, and its stock is still over 50% below its all-time high. Moreover, Intel's $147 billion market cap is only around 7 times its peak earnings reached back in 2019.
Therefore, if new CEO Pat Gelsinger, who took over in 2021, can turn Intel's ship around, it's possible Intel may actually be the better pick.
That's a big question though -- or rather, several questions.
First, will the AI revolution be incremental for CPU demand, or will GPUs displace more CPUs across the global data centers? Second, will Intel be able to make inroads with its own GPU and accelerator platforms? And third, can Intel get back its process technology lead, or at least pull even, with Taiwan Semiconductor Manufacturing, enabling Intel's new foundry business to attract third-party customers?
These are all unanswered questions, but Intel has offered some encouraging commentary on all these fronts in recent months.
In late March, Intel held its data center event, claiming that its product roadmap was back on track, with its latest 4th Gen Xeon processor receiving strong adoption. Then on the recent earnings conference call with analysts, Gelsinger noted 25% of 4th Gen Xeon shipments were for AI workloads. Even though AI training typically centers around GPUs, CPUs are still playing a part in certain parts of the AI process, including data preparation, head nodes that run the AI program, and even inference for smaller AI models. Management maintained that while there is currently a big shift to GPUs in this lean economic environment, AI should still be a market expander for CPUs overall.
Moreover, Intel showed confidence that its "five nodes in four years" roadmap is on track, with its 5th Gen Xeon due out later this year, two more chips on its next generation process on track for 2024, then a chip called Clearwater Forest in 2025. That chip will be made on Intel's 18A process, which is when management believes it will catch up to TSMC's leading-edge nodes.
Image source: Getty Images.
Of course, Intel is also developing its own AI accelerators and GPUs, which it hopes will make inroads against Nvidia. Intel has not one but two AI chip platforms: the Gaudi line of AI accelerators, which it acquired via the acquisition of Israeli-based Habana Labs in 2019, and its in-house designed GPU Max chips. Sometime in 2025, Intel is looking to combine both technologies into a unified platform called Falcon Shores.
On the latest conference call, Gelsinger noted Gaudi had over $1 billion of pipeline opportunities through 2024, and is already in instances on Amazon Web Services. That's far, far behind Nvidia, which had over $10 billion in data center revenue last quarter alone, but that figure for Gaudi was up 6x over the prior quarter for Intel, showing upward momentum as customers look for Nvidia alternatives.
Intel is also attempting to develop an open software system that can disrupt CUDA's moat, called SYCL. In its March presentation, Intel noted one of its APIs was able to migrate 90% of a CUDA workload to SYCL seamlessly. While details around that particular migration are scarce, if the SYCL API succeeds in easily porting over CUDA code to an open platform, that means Nvidia's CUDA would have less of a "lock in" effect in the future.
Finally, Intel could also benefit from AI through its foundry system that it's opening to third parties. Intel could theoretically produce some of the alternative AI accelerators designed in-house by the large cloud platforms, and Intel's back-end advanced packaging could garner lots of business for "chiplets," in which parts of an accelerator processor are made separately and then stitched together in an optimal way. Of note, Advanced Micro Devices' new MI300 is built on a chiplet architecture.
On Thursday, Aug. 31, Gelsinger also noted Intel had received a large customer pre-payment to build out the 18A fab, indicating a commitment to Intel's third-party manufacturing for leading-edge nodes. That vote of confidence was encouraging for the prospects of Intel catching up to TSMC.
Both stocks could also succeed
Nvidia has a "clean" story in AI, with a leading product and closed-source software ecosystem, as well as terrific near-term performance. Likely, it will be able to hang on to a large portion of its market share, and its stock can still move higher from here.
However, investors shouldn't discount Intel, either. Its story is multifaceted and therefore "messier," but it also has several ways to benefit from AI, even if one particular part of its portfolio doesn't pan out.
The AI revolution will greatly grow the market for computing solutions, so there may be a big enough pie to go around. Even a modest slice of that pie could work out well for Intel stock, as its depressed valuation isn't factoring in much good news.
Investors should monitor rapidly evolving news in this space -- especially for indications Intel's open-source AI software is beginning to make inroads against CUDA's dominance.
