GM. This is Milk Road Stocks, the newsletter thatβs the designated driver for your market decisions.
Today weβre taking a look at why 3 of our 5 PRO analysts are buying the SpaceX dip right now, and what catalysts/threats could lie ahead.
WHY OUR ANALYSTS KEEP BUYING THE SPACEX DIP π
Compute is scarce right now, and whoever owns it can charge more for it.
Gavin Baker (founder of Atreides Management) said the price to rent "essentially any kind of a GPU" went up every day in October.
Yet AI stocks have spent the past few weeks getting sold off anyway...
Kyle (our Head of Research) thinks the market has this backwards.
In his view, the biggest advantage a company can have right now is how much compute it owns or can access.
Google and Microsoft lead that race today, but Kyle thinks SpaceX can take the #1 spot within 5 quarters.
His case comes down to three things:
1. SpaceX rents at today's prices. It sells compute on short-term contracts at whatever the going rate is right now, which Kyle reckons works out to $50B+ for every gigawatt it brings online.
When rents climb, SpaceX's revenue climbs with them. Neoclouds (smaller companies that rent out AI chips) mostly lock customers into multi-year deals at fixed prices, so they don't get that boost.
2. SpaceX builds fast. It has ~2.2 gigawatts online today, and expects 2.5 to 3GW by the end of the year.
Elon is "highly confident" of ~13GW by the end of 2027, partly because SpaceX makes its own turbine blades (a part that takes 12 to 18 months to arrive for everyone else). π
Source: SpaceX Q3 2026 earnings call
3. The customers are already there. Q3 revenue hit $14.58B, up 161% from a year earlier, and AI was the biggest chunk at $7.56B.
Anthropic alone has agreed to pay SpaceX ~$3.1B a month once all its contracted compute is delivered.
You'd think all that would have the stock flying...
Instead, SpaceX is trading at ~$133 this morning, down ~17% over the past month.
Part of that is the wider AI selloff, while the rest comes from share unlocks - which free up ~40% of the company by December.
(Only 5% could trade at its June IPO.)
Kyle doesn't expect a wave of selling though, since most of those holders could already cash out at every funding round before the IPO and chose to hold.
That's why three of our five PRO analysts have bought more SpaceX since Friday:
- Kyle said he "will buy SpaceX under $140 all day," calling it probably the best risk/reward in the market for anyone holding a year or longer. It's now ~11% of his portfolio.
- Vincent sold some Tesla to pay for his, saying he'd "rather own the compute now."
- M0xt doesn't own any hyperscalers, so SpaceX is his way into the trade.
The catch here is the cost of it all...
Kyle estimates SpaceX needs to spend ~$470B over the next 5 quarters to reach 13GW. That's about $30B per GW, rising to $45B once Nvidia's next chips (called Vera Rubin) arrive.
Vincent thinks ~$340B of that has to come from outside investors, so he expects SpaceX to sell new shares within the next 12 months (more shares = a smaller slice for existing owners).
Melius Research raised the same worry this week, calling Big Tech's flood of new stock and bond sales the #1 risk to the AI trade.
Elon's response is that the market underestimates how cheaply SpaceX can build.
In short: buying SpaceX here is a bet that compute stays scarce.
If GPU rents keep climbing, SpaceX sells more compute at higher prices and the buildout pays for itself in months, like Kyle expects.
If rents flatten, that ~$470B expenditure will become a lot harder to justify.
