THE CAP TABLE - APR 22, 2026
"No one has any compute = AI data centers are a great investment (in our opinion)"
Link to watch video =
youtube.com/watch?v=hArtSgwr…
Aaron Ross and I started this podcast a few weeks back. Mr. Ross is a super sharp guy and I highly recommend folks reach out to him if you're interested in his pre-IPO stock research or care to learn more about the pre-IPO stock asset class. You can contact Aaron at aaron.ross@rosspreipo.com.
Our discussion this week focused on AI data centers. The single most underappreciated dynamic in markets today ... in my opinion ... is the compute shortage powering artificial intelligence. Roughly 30% to 50% of US data centers are currently running behind schedule, while demand continues to compound at a pace few investors fully appreciate. Heavy users of Claude, ChatGPT, and Grok are routinely hitting daily usage caps before noon. Software developers have begun launching autonomous coding agents that run uninterrupted for 13 days at a time, then immediately queuing the next job behind it. Inside major enterprises, executives are starting to ration tokens across departments the way they once allocated headcount. Within the next 24 months, we expect virtually every white collar worker to have automated agents executing tasks in the background of their day, and at that point the global compute demand curve becomes staggering very quickly. Compute is on track to become one of the most strategically important commodities on the planet, and we would not be surprised to see a tradable futures market for AI compute "tokens" emerge within the next several years.
For investors, this points directly at the AI infrastructure layer: semiconductors, data centers, and electricity production. These are the picks and shovels of the buildout, and on a risk adjusted basis we view the setup as one of the most compelling in the market today. Compute access is also reshaping competitive dynamics among the model providers themselves. Anthropic currently leads on model quality but remains constrained by its reliance on hyperscaler capacity. OpenAI has absorbed meaningful dilution to fund its own buildout, though that proprietary compute may translate directly into faster revenue growth from here. xAI's Colossus cluster in Tennessee, now home to 555,000 GPUs, gives Elon Musk's team more proprietary compute than Anthropic and OpenAI combined, and the company has already begun selling capacity to third parties such as Cursor (...although it was later announced this week that SpaceX/xAI is going to buy Cursor for $60 billion). On the private side, the names we are watching most closely include the Neocloud operators (Lambda Labs, Together AI, Crusoe, Nscale, Vast, Fermi), the inference specialists (GroqCloud, Cerebras), and the orchestration layer (Baseten, Fireworks AI). Our screening framework is straightforward: we want AI infrastructure companies with a strategic partnership with Nvidia and Nvidia on the cap table as an investor. Check those two boxes, and the setup gets very interesting.