On their earnings call this week, Nvidia mentioned that they expect the Total Addressable Market (TAM) for data center semis to grow from $250 billion this year to $500 billion in five years. They contend that incremental $250 billion will largely go to things near and dear to Nvidia’s heart – which we are just going to call AI. A massive market will double and Nvidia expects to pick up much of the share in that growth. Just for kicks, we thought we would model that out to see how big Nvidia could grow.
Normally, we are confident that our readers would know to not take this kind of musing too seriously. However, if you are new to our site, if you were perhaps sent here by a Fintech Influencer of some sort, just know that you should absolutely not be investing on the basis of these numbers. All TAM estimates live just this side of the boundary between provable fact and wishful thinking. The number of unknowns and variables inherent in a five-year TAM forecast could easily be the basis for a science fiction trilogy.
Specifically, Nvidia said that they expect the installed base of data center hardware to grow from $1 Trillion now to $2 Trillion in five years. We pressed them a bit on the details, and they said that works out to $250 billion a year growing to $500 billion a year, which roughly makes sense if we assume that the useful life of data center hardware is 4 years, give or take a few accounting treatments. That includes total hardware spend – memory, power systems and the metal racks that hold all of it. By their math, Nvidia can address 60% of that spend – CPUs, GPUs, accelerators and networking. Again, this roughly squares with our intuition of the market. Note that only AMD can make that same 60% claim today. Intel can probably only tap into 40%-50% (if we are being very generous about Gaudi).
That gave us enough detail to hang our model on. We forecast out the base TAM, then added an “AI TAM” to reflect the growth component. We then calculated the amount of the TAM that Nvidia can actually service (the serviceable addressable market or SAM) which we pin at 60%.
Then we get to the stickiest part of the model – how much of the SAM Nvidia can actually capture. As we have noted, Nvidia captured 75% of the data center processor market last quarter, but this is not the same as the SAM. To get their current revenue to square with the data, we assumed that their share last year was 19%. This is probably too low, and if we were more diligent about this, we would rethink this figure a bit as it ends up being fairly important. So depending on how we want to measure the SAM, we estimated that Nvidia’s share grows to 45% in calendar 2024, and then stays there as the market grows but competitors get their products into the mix. To be clear, this is already a fairly bullish outlook for Nvidia, it assumes that they hold onto their share gains in recent years. We then assume the rest of Nvidia’s business continues to grow a bit and that they maintain their net income margins and share counts.
That model yields 2028 revenue of $150 billion and EPS of $25.52. For most companies that kind of growth sounds outlandish, but Nvidia doubled revenue this year, and we are not forecasting anywhere near that kind of growth going forward.
There are many big swing factors in this model. Market share and revenue growth are top of the list, but profit margins and share count stand out as well. So we created a sensitivity analysis that looks at 2028 EPS under varying net margin and market share (SAM share) levels.
Looked at on this basis, Nvidia’s stock looks fairly valued. If they can mostly hold on to their share gains and only give up a little bit of margin, then their stock price has largely priced in the gains we forecast here. Of course, this is all in the eye of the beholder. We can easily construct bearish scenarios here, as well as much, much more bullish ones. Echoing our note from yesterday, Nvidia looks to be in enviable shape as a company, but its stock is much more finely balanced.
Photo by Microsoft CoPilot
Leave a Reply