How you should have known to invest in NVIDIA
On the day that OpenAI’s ChatGPT 3 went public (November 30, 2022), NVIDIA’s stock opened at $156.97 per share. If you would have bought one share at that price, 19 months later (June 2024) you’d have over $1000 in profit. That ~600% growth smashes the ~35% growth of the S&P or the ~100% growth of AMD over the same period!
NVIDIA’s ascension is largely due to the insatiable appetite for GPUs to fuel more and bigger large language models (LLMs). Who could have guessed? Well, everyone could have, actually. Here’s why.
The Framework
None of this constitutes financial advice. But let’s consider how we might apply the theory of disruption to identify rocketship investments. Fundamentally, the answer comes down to finding a company that makes a highly-developed product that is useful across a large swath of disruptive applications.
In general, cost increases as any product becomes more developed. Consider wood being made into a mahogany bar. Each step, from lumber, rough-cut boards, to finished product adds value. The same occurs as sand is made into a computer chip. Typically, the farther along the production chain one is, the greater the profit margins, the more attractive to investors.
The greatest opportunity for profit often occurs at the application of the final product. If the mahogany bar is installed in a hotel, the hotel will make far more from drinks served across the bar than the company who made or installed it ever will. Unless, of course, the hotel fails. Or burns down. Or gets remodeled. The potential gains of applications come with greater risk.
To paraphrase Clay Christensen in The Innovator’s Dilemma, a precondition for a technology to be disruptive is that the applications of the technology aren’t yet known. Christensen gives the example of 3.5" hard-disk drives — which severely under performed the 5" drives being used in desktop computers — entering the market inside laptop computers and then taking over desktops and servers.
Investing in Toshiba or Dell would have turned out great! But what about Compaq or numerous other laptop manufactures that aren’t around any more? Not so good. The ideal spot would actually have been to invest in the 3.5" disk manufactures themselves.
The uncertainty of the success of any one application success is part of why Meta, Google, and Microsoft may not be good investment options; we don’t know how they will apply LLMs to their product and if they will make money doing so.
In many cases, you can’t invest in disruptive applications anyway because the start-ups aren’t publicly traded. Neither OpenAI nor Anthropic, jockeying for the best LLM available to the public, are for sale on the stock market.
So, as a casual investor, applications of disruptive technology are out. Instead, we have to find the (publicly traded) common dependency amongst the disruptive applications. In the case of LLMs, that common dependency is NVIDIA GPUs.
Specific Rules
Here are some more rules to help make that choice.
First, the higher up the value chain this dependency lies, the better for you. In this case, NVIDIA is preferable to TSMC — who does the actual manufacturing of the chips for NVIDIA — because NVIDIA is higher up the chain and takes a greater margin.
Second, the more unique of a capability the dependency is, the better for you. NVIDIA GPUs are the go-to for artificial intelligence because NVIDIA’s CUDA framework is what allows the software that trains the LLM to talk to the GPU. That software cannot talk to other GPU architectures (such as AMD) without extensive modification and testing. So in this case, NVIDIA is highly unique! In contrast, Dell sells servers into which these GPUs must be installed. But lots of companies make servers, and lots of companies don’t really care what server they have, as long as they can put an NVIDIA GPU into it.
Third, the more this dependency stands to grow the company as a whole, the better for you. Prior to the recent explosion, only 40% of NVIDIA’s revenue was from data center and AI applications. The bulk of the company’s revenue came from personal computers. But in 2024 a whopping 78% of revenue is data center and AI applications!
Fourth, the smaller the company is at the time you invest, the better for you. In November 2022, NVIDIA only had a market cap of $419B. Compare that to Alphabet’s $1.4T or Microsoft’s $1.9T. When a company is valued at over a trillion dollars it takes truly massive growth to see high-percentage stock returns.
Conclusion
When taken altogether, NVIDIA was the perfect investment back in November of 2022. The launch of ChatGPT 3 promised an LLM boom that is dependent on GPUs that only NVIDIA can produce, while making huge margins, with nothing but room to grow!
Where do we go next? NVIDIA GPUs are primarily used on the training side of the house, which happens before the model is served to the user. You can also use GPUs in the cloud to do inference (respond to a user), but this is high-latency and expensive.
It seems there is massive opportunity for inference at the edge; for example, running a LLM on your phone or laptop. In fact, this is the explicit goal of Microsoft’s Copilot+PC as well as Apple’s On-Device Models.
Both of those are applications from two ginormous companies, so don’t fit our criteria above. But it’s not a far shot to see where there might be some opportunity…
Happy disruption and investing!
