On the day after Nvidia posted blow out earnings (again) and the stock has popped about $100 to >$1K/share, I figured this would be a convenient time to finally publish this piece that I’ve been writing since March. “But how high can it go?” is what everyone said about NVDA at stock prices of $300, $450, $600, and now over $1,000 (where it currently stands at a market cap of $2.5T). Here’s how Nvidia can go to $4,500/share(*) - which would value the company at more than $10T.
First, recall that today there are about 400M CPUs and 40M GPUs sold per year (for 10% attach). Given the seemingly insatiable demand for AI chips (and that NVDA has something slightly shy of 100% market share in the most important segment for these chips), it should be obvious that the 10% attach will grow considerably. But to get a 5-10x return at these levels, that attach rate would need to near 100%. Likely? Meh, but also not impossible.
Second, recall that I wrote in February about the modern software stack collapsing to a single logic layer (from the front end + logic + back end 3 layered model in place everywhere today) and that the only place that stack could sit is on GPU or GPU like hardware.
Third, aside from the big Blackwell buzz, Nvidia announced a new product at the March 2024 GTC called “NIMs” - Nvidia Inference Microservices. In effect, Nvidia is betting that companies will put the inference microservice in front of all of their (internal) databases so that querying databases and making sense of the content in these databases will be “LLM-like.”
See what’s happening? This isn’t just the ChatGPT-like AI chatbot / copilot for kids to “write” their term papers and self-proclaimed futurologist clowns to create silly images. Rather, this is embedding AI / LLM-like architecture deep into devops and code. If this trend continues, it is much less difficult to see a world where all software is based on this type of single (logic) layer architecture sitting on GPU or GPU-like infrastructure and for that GPU:CPU attach rate to sprint closer to 100% and for whoever owns the market for this technology to grow their market cap closer to $10T.
^^”Me.”
(*) I used the term “bull” here as a double entendre; you can figure out the alternative meaning.
(**) It is likely inconceivable that Nvidia will maintain ~100% market share here, but….
(***) All share prices here are pre the forthcoming NVDA split.
(****) I still own NVDA shares, though, as always, not enough.