The trillion-dollar club is certainly rarified air, and one that every company aspires to join. Investors, of course, aspire to find the next candidate with the potential to join such elite company. Odds are, every company that has reached that milestone has made many fortunes for investors along the way.
Case in point: Artificial intelligence (AI) chip star Nvidia (NVDA +0.87%) began 2023 at just a $350 billion market cap, but has more than tripled since, reaching a market cap today exceeding $1.2 trillion!
So, where to find the next big winner? It may not be surprising that I think the AI revolution could mint even more trillion-dollar companies, like this key Nvidia rival.
NASDAQ: AMD
Key Data Points
Can Advanced Micro Devices catch up?
Given the explosive growth in AI computing we are seeing and are likely to see through this decade, those tailwinds could potentially propel Advanced Micro Devices (AMD +3.36%) to new heights.
That is, of course, only true if the company executes. But AMD's execution has been excellent ever since Lisa Su took over as CEO in October 2014. Back then, AMD had just a $2.4 billion market cap -- yes, you read that right. Fast forward to today, and AMD's market cap has gone up nearly one hundred times under her tenure, to $223 billion.
AMD Market Cap data by YCharts
How did AMD do it? As is usually the case, a lot of skill, planning, and a little bit of luck.
First, Su made the choice to diversify away from PCs, where AMD got 90% of its revenue at the time, and into more high-end applications such as gaming, embedded chips, and data center processors. Second, Su used the 2017 transition to FinFet transistors to attempt to catch up to market leader Intel in process technology.
At the time, Intel had a near-monopoly on the processor market and leading technology. AMD practically existed just so many customers would have a secondary option for cheap processors and keep Intel from having 100% of the market. But as chipmaking became more complex, with FinFets and the 7-nanometer process node, Intel stumbled.
Meanwhile, AMD's foundry partner Taiwan Semiconductor Manufacturing (TSM 0.83%) excelled and pulled ahead of Intel on leading-edge chipmaking. That left an opportunity for AMD to actually leap ahead of Intel in terms of power and performance, enabling it to gobble up market share from a near-zero starting point.
Can AMD pull off the same feat in artificial intelligence chips?
Given that AMD was able to meet and overtake a large and dominant rival in PC processors, can the company do it again with AI accelerators?
While Nvidia has a big lead in that area today, AMD introduced its challenger MI300 line of AI accelerators in June 2023. The MI300, unlike the all-in-one die that hosts Nvidia's H100, is made up of a "chiplet" architecture consisting of 12 chiplets. That architecture allows for some advantages, including a whopping 153 billion transistors, with the ability to handle 192 gigabytes of HBM3 memory. That compares with the H100's "mere" 80 billion transistors and just 80GB of HBM memory capacity.
In light of these specs, AMD certainly has a shot at getting some looks from AI customers, especially those with memory-hungry applications. No wonder Su has said she expects at least $2 billion in revenue for the MI300 accelerators in 2024, despite just ramping production recently.
Can AMD compete with Nvidia in AI? Image source: AMD.
But Nvidia should still remain a leader for the foreseeable future
While the MI300 exceeds the H100 on certain specs, investors shouldn't expect AMD to match Nvidia any time soon. One important difference is that the MI300 doesn't have the transformer engine of the H100, which can be turned on to triple the H100's performance. That would, of course, shorten the time to train large language models, and could keep customers in the H100 for the most pressing applications.
Another advantage for Nvidia is that it will already be coming out with the H200 in the second quarter, and will move on to an entirely new architecture, the B100, next fall/winter. In October 2023, Nvidia announced that it would move from a two-year cadence to a one-year cadence for new AI chip architectures. Given that AMD is just ramping the MI300 now, it will actually be competing more with these newer chips than with the H100. And since Nvidia is now going into turbo-speed, it may be hard for AMD to eventually catch up.
This is especially true because Nvidia also uses TSMC as its foundry, so Nvidia will likely not stumble in the way Intel did five years ago. If TSMC stumbles, both Nvidia and AMD would suffer together.
In addition, Nvidia has been developing its CUDA software stack since 2006, and has achieved a nice network effect with developers. AMD is quickly developing its own quasi-open-source software stack called RocM, which it's piecing together from its own research and development combined with several recent acquisitions last year. Still, a 15-year-plus head start will be difficult to overcome quickly.
But $1 trillion is still in sight by 2030
Even if AMD forever remains a "next-best" option for AI chips, it's possible the AI market could be large enough that AMD will reach $1 trillion anyway. After all, at its December 2023 presentation, Su increased her projection for the AI chip market from $150 billion to $400 billion by 2027.
While some of that increase comes from chips for "AI PCs," the good news is that AMD will potentially have an even better chance at leading that market, considering its strong position in PCs today.
AMD hit $24 billion in revenue in 2022 before the PC downturn caused a slight dip this year. But even an incremental 10% of the $400 billion AI chip market could propel AMD's revenue to, say, $65 billion by 2027. Assuming a 30% net income margin, that could yield $20 billion in earnings. So a price-to-earnings multiple of 50 could get AMD to a $1 trillion market cap by that time.
That may seem like a high valuation, but it's a pretty good bet that the AI chip market will continue growing beyond 2027. It's also quite possible AMD could get more than that modest 10% share of the AI chip market. So by 2030, it's certainly a fair shot to get to that $1 trillion market cap, more than four times today's valuation.
