SpaceX's $24.5B AI Revenue Pile: The Re-Rating Nobody Is Shorting
Bank of America just dropped a number that rewrites SpaceX's valuation framework. Price target: $235. Reference price: $125.33. Implied upside: 87.5 percent. And none of it is about rockets.
The bank's bullish thesis, per the August 7 report, is no longer anchored to launch services. The new engine is AI infrastructure. Revenue from the Anthropic partnership began in May. Google's compute collaboration is expected to go live in October. BofA projects AI-related revenue of $24.5 billion in 2026 โ roughly 52 percent of the expected $46.9 billion total. By 2028, total revenue is projected at $184.8 billion.
Read that again. A company the market still files under aerospace is expected to generate hyperscaler-level revenue within three years. That is not an upgrade. That is a sector reclassification, and the market has not fully priced the mechanics of what it takes to deliver those numbers.
Here is the structural problem hiding in plain sight. SpaceX's free cash flow projections are violently negative. Negative $43.6 billion in 2026. Negative $45.4 billion in 2027. Negative $37.4 billion in 2028. Cumulative three-year burn: approximately $126 billion. The bank frames this as a capital expenditure supercycle. I frame it as a financing event waiting for a clearing price.
Context first. SpaceX is vertically integrated in ways no cloud provider can replicate: rockets, satellite manufacturing, the Starlink low-latency network, solar, energy storage. That vertical stack is the bull case for AI infrastructure โ compute plus connectivity deployed where terrestrial providers cannot reach. The bear case is less discussed. None of it is real until contracts are signed, chips are delivered, and power is secured. The market's current pricing, built on launch dominance and Starlink subscriber growth, already embeds a substantial premium. What it does not embed is the AI re-rating. That discrepancy is the trade.
The AI compute market, meanwhile, is already crowded. AWS, Azure, and Google have allocated hundreds of billions to capex. CoreWeave is building GPU clouds on debt. Oracle is buying clusters by the gigawatt. Every one of these players has a head start in chip supply chains, data center operations, and enterprise sales. BofA's model assumes SpaceX carves out a meaningful share without pricing the competitive response. That is the key omission.
Now the core analysis, and I am applying the same framework I used during the 2020 DeFi yield farming window: find the unit economics, count the counterparties, stress-test the timing.
First, unit economics. $24.5 billion of AI revenue requires a specific compute volume. At prevailing market rates, that means hundreds of megawatts of deployed capacity at very high utilization. It requires GPU allocations from NVIDIA or AMD, power purchase agreements, and a network fabric capable of sustaining training runs without bottlenecking. None of this appears in the report, because BofA is underwriting the contract, not the physics. In the DeFi summer of 2020, I deployed $500,000 into a Uniswap V2 and Curve rebalancing strategy and captured $85,000 in two weeks. The edge came from understanding the mechanics of liquidity, not from believing the yield was free money. Same discipline applies. Revenue projections without cost curves are marketing.
Second, counterparties. The counterparty structure is where the model gets fragile. Anthropic is paying SpaceX for compute. Google is paying SpaceX for compute. Two anchor customers. Google builds its own TPUs. Anthropic already rents capacity from AWS and Google Cloud. Why would both turn to a rocket manufacturer? Three plausible explanations: terrestrial compute is genuinely scarce in specific regions, SpaceX is offering lower effective pricing through vertically integrated energy and launch costs, or the strategic alignment includes equity, prepayments, or exclusivity clauses that are not public. If it is the third, the cash flow picture is better than it appears. If it is the first two, the model is durable but vulnerable to hyperscaler price wars. Google is a particularly strange ally: it is simultaneously a customer and a direct competitor in AI compute. That is a frenemy structure with built-in opacity. It also means the single most important contract in this model โ the Google deal โ could be priced more for strategic positioning than for genuine capacity need.
In my 2024 institutional hedging work, I ran a delta-neutral collar on a $10 million ETF exposure. The counterparty quality mattered more than the strike selection. Same principle here. If Anthropic is prepaying for future capacity, SpaceX's negative free cash flow is partially funded by customer capital โ a non-dilutive source. If the contracts are pay-on-delivery, the entire capex burden lands on equity and debt markets. At current funding costs, that is a structural disadvantage against hyperscalers financing at lower rates.
Third, timing. The Google partnership is "expected to begin in October." That is optionality, not certainty. The Anthropic revenue has a live pilot behind it. But $24.5 billion is not a pilot. It is a declaration that AI compute demand remains tight through 2026 and that SpaceX can execute on a deployment schedule that has tripped up every other entrant. Hyperscaler builds are running 12 to 18 months behind schedule. The delivery timeline is the single largest technical risk in this model, and the market is treating it as a footnote.
Now the contrarian angle. The crowd reads "SpaceX wins AI." Institutional desks read "SpaceX needs $126 billion of external capital at the peak of the AI capex cycle." Those are not the same trade.
The floor didn't hold for BAYC in 2022 when the narrative cooled and liquidity thinned. The same mechanics govern narrative-driven equity valuations. When the story rotates from "rocket company with AI optionality" to "capital-hungry infrastructure provider with two customers," the multiple compresses before fundamentals catch up. In 2022, I sidestepped a 60 percent NFT floor crash by executing structured OTC block sales instead of market panic dumps. The discipline was simple: liquidity management decides who survives, not conviction. Apply that same lens to SpaceX.
There is another layer the report conveniently skips: the engineering risk of space-based data centers. Free solar power and cryogenic cooling in orbit sound elegant. But inter-satellite optical links have latency and bandwidth constraints. Orbital maintenance, collision avoidance, and space debris liability are cost lines that do not exist in terrestrial cloud accounting. If SpaceX instead executes a ground-based buildout, it competes directly with hyperscalers on power, land, and chips โ where it has no structural edge. The market is pricing infrastructure that may not physically exist yet.
Institutional reality check: BofA is sell-side. It is structurally incentivized to present the bull case. The bank's report landed August 7, right after quarterly numbers confirmed launch strength but left the AI engine unproven. Sell-side re-rates fast. Cash flows do not. The report also ignores export controls, data sovereignty, and the renegotiation risk embedded in two-customer concentration. Each anchor customer has alternatives. In this story, SpaceX does not.
The arbitrage embedded here is real. The market will eventually reassess SpaceX from an aerospace multiple to an AI infrastructure multiple if even half of the AI revenue materializes. But chasing that re-rating today means accepting a company that burns cash at a rate requiring constant capital market access. That works in a bull market. It becomes existential in a liquidity crunch.
The takeaway is mechanical. Watch the capital structure, not the price target. If SpaceX announces an equity raise or convertible notes, the $235 target is under pressure. If it signs a third and fourth AI customer, the model begins to self-fund. If neither happens, negative free cash flow compounds into dilution.
The question is not whether AI infrastructure is SpaceX's future. It is whether the market is paying for contracted revenue or for hope. I have personally traded both setups. Only one of them survives a sharp drawdown.