The 5.4x AI Re-Rating Buried in SpaceX's $300 Target

A $300 price target crossed my terminal this week, and every aggregator ran the same headline: SpaceX could double. I don't trade equity narratives for a living. I audit them. So I did what I do to every token launch with a glossy deck — I tore the segment math apart and rebuilt it myself. What I found is the most aggressive assumption in the entire note, and almost nobody repeating it has noticed. The analyst's own sum-of-the-parts — rockets and Starlink at $127 per share, AI at $32 per share — lands at $159. The stock trades at $158.96. To the cent. Which means the entire $300 target, all $141 of upside per share, is one bet: that the AI business re-rates from $32 to roughly $173 a share. A 5.4x multiple on the AI segment alone. Nobody wrote that number down. I will. The setup matters. SpaceX listed in June 2026, and this week Morgan Stanley's Adam Jonas — relayed through a Walter Bloomberg tweet — floated $300 against a $158.96 close. Thirty-three analysts average $235.10; the low is $142. Jonas is the high-end outlier, and outliers always tell you where the story wants to go. Two structural anomalies sit underneath the headline, and the note never addresses either. First: Grok is xAI's product, not SpaceX's. The report folds "Grok chatbot plus compute rental" into a single "SpaceX AI business." That only holds if xAI has been merged or deeply integrated into SpaceX — a possibility Musk has openly floated. If it hasn't, the valuation rests on a factual error. Second: it prices a model-type business (Grok, software margins, subscription curves) and an asset-type business (compute rental, depreciation curves, utilization risk) as one line item. Those two assets behave nothing alike. Mixing them distorts everything downstream. I spent years learning this the hard way — back in 2017, auditing ICO whitepapers with a cybersecurity lens, I watched founders bundle a protocol, a token, and a "platform" into one valuation slide, and the market priced all three as if the best one were the whole company. Code is law, but audits are mercy. Here is the core. The AI division lost $1.26 billion last quarter and consumed 86% of SpaceX's capital expenditure. Jonas himself admits the current buyers are paying for rockets and Starlink, and paying almost nothing for AI. His short-term compute trades clear at $30 to $50 per watt; his own model assumes $17.60. That gap is the whole story in miniature — the spot market says compute is scarce and precious, the model says it reverts to a boring long-run equilibrium. The note never states whether $30 to $50 is a one-time capacity sale or an annualized rent. That single ambiguity is worth an order of magnitude in valuation, and it goes unexamined. And this is where the crypto reader should lean in. Compute is becoming a tradeable asset class. Pricing by the watt is precisely what happens when a resource stops being a service and becomes a commodity — the same transition that turned bandwidth into DePIN tokens and GPUs into on-chain yield. If $30 to $50 per watt holds as an industry anchor, every entity sitting on self-built compute gets marked up, asset-heavy players win, and pure-model players get compressed. That is the real reason SpaceX's AI segment carries a $4,220 billion mark. It is not the model. It is the megawatts. The pool remembers what the ticker forgets. I ran the same logic in 2020, reverse-engineering Uniswap V2's bonding curve for two weeks and arguing that centralized venues were structurally obsolete because MEV was being extracted in plain sight. Everyone called it premature. The point wasn't the price — it was that the mechanism, not the marketing, determines who gets paid. The same discipline applies here. If watt-pricing is real, then the physical constraint stops being silicon and becomes electricity and cooling. That reframes the entire AI race: whoever locks the cheapest megawatts wins, regardless of whose model benchmarks higher. For a company that owns launch cadence, orbital bandwidth, and now a Memphis-scale data footprint, that is a genuine moat — just not the one the note is selling. Now the contrarian read. The note's central bullish lever — SpaceX AI trades "about 40% below large AI peers" — has no stated basis. EV/Sales? EBITDA? $/W? Without a denominator, "40% cheap" is a constructed anchor built to justify a Buy rating, not a market-verified discount. Meanwhile Grok is a chaser, not a definer: investors list "can Grok keep up with competitors" as their single biggest worry, and the note offers no benchmark, no user count, no API volume to rebut it. A chaser earning a leader's multiple is how narratives die. Speculation is just data with a heartbeat — and the heartbeat here is faint. Then the supply picture. On September 24, roughly 328 million shares unlocked. President Gwynne Shotwell sold $52.5 million. And Jonas, canvassing forty clients, asked who holds the AI story. Nobody raised a hand. Read that as latent demand if you're bullish; read it as smart money leaving while retail is invited in if you're not. I know which reading has better evidence — insiders selling into a Buy rating is the oldest tell in the book, and it reads the same whether the ticker ends in a dollar sign or a token symbol. Zoom out, and this connects to the thesis I've been running since last year: that autonomous agents, not humans, will drive the majority of on-chain volume by 2027. Agents don't care about brand. They route to whoever offers the cheapest, most verifiable compute, and they settle in stablecoins at machine speed. That makes watt-denominated capacity a natural primitive — an asset class that agents can price, rent, and hedge on-chain without a human in the loop. The SpaceX note is, accidentally, the first mainstream document to treat compute this way. It just hasn't admitted it yet. So the doubling thesis is not "SpaceX is 89% undervalued." It is "everything except AI is already fully priced, so AI must do all the work — and it must do 5.4x of it." That is the most concentrated bet in the note, dressed as diversification. The bullish case requires three negative facts to reverse simultaneously: a chaser model must catch up, a bleeding division must turn profitable, and a buyer base that currently pays for nothing must start paying for everything. What to watch. Starship Flight 15 landing the returning ship is the strongest near-term catalyst, landing late October into early November. Third-quarter earnings will show whether the AI loss is narrowing and whether compute revenue is ever disclosed as a real line. Grok's next release and third-party benchmarks will confirm or kill the competitive worry — set a trigger: if Grok doesn't crack the first tier, the re-rating math fails. Most important, watch the next compute trades and their $/W. If $30 to $50 becomes an industry reference, the re-rating goes systemic, and on-chain compute markets become the real price-discovery layer for AI's physical footprint. If it was scarcity froth, the $4,220 billion mark is the first thing to crack, and it will crack faster than any equity desk expects. Rewriting the rules before the bug writes them is the only edge that compounds. Entropy increases until someone audits it — so keep auditing. The truth is hidden in the gas fees.