From Rockets to Rent-a-Rig: BofA’s SpaceX AI Bet and the Architecture of Belief

On August 7, Bank of America threw a grenade into the aerospace valuation playbook. Price target: $235 — about 87% above the post-earnings reference of $125.33. But the explosive detail wasn’t the number. It was the narrative shift. BofA now says SpaceX’s growth engine is no longer rocket launches or Starlink subscribers. It’s “AI infrastructure.” The bank projects $24.5 billion in AI revenue for 2026 alone. That’s 52% of the total revenue forecast of $46.9 billion. In two years, the bank sees total revenue hitting $184.8 billion. A rocket company that has never generated more than a fraction of that is suddenly being priced like a hyperscaler. The only evidence? Two customer names and a mountain of negative free cash flow. Let’s be precise about what the report actually says. The AI revenue stems from a partnership with Anthropic that began in May, and a Google deal expected to start in October. BofA’s model then extrapolates from those two contracts to a future where AI is half the business within twelve months. It sees capital expenditures “continuing to expand” while free cash flow remains deeply negative: -$43.6 billion in 2026, -$45.4 billion in 2027, -$37.4 billion in 2028. Cumulatively, that’s roughly -$126.4 billion over three years. For context, AWS and Azure spend in the same ballpark annually, but they have revenue to match. SpaceX’s AI unit is expected to generate $24.5 billion in 2026 — yet the total projected revenue for that year is $46.9 billion. The burn is real. The source of the revenue is not , at least not verifiably. Here’s where my forensic instincts kick in. I’ve spent years auditing DeFi protocols where the yield math always outpaced the on-chain reality. This report reads the same way. BofA is not evaluating a technology; it’s underwriting a contract-based projection. The report provides zero technical detail on how SpaceX intends to deliver AI compute at scale. No chip architecture. No cluster sizes. No data center locations. No mention of whether the infrastructure is terrestrial, orbital, or a hybrid. The only data points are the two customer names and a revenue curve that doubles almost every year. That’s not analysis — that’s a narrative. Unspooling the knot of innovation here requires separating the business from the story. SpaceX certainly has assets that could theoretically support an AI infrastructure play: reusable rockets, satellite manufacturing capacity, the Starlink low-Earth-orbit network, and an in-house solar and battery capability. If the AI compute is space-based, you get free solar power and cold-space cooling. If it’s ground-based, you’re competing for grid power with every other data center builder. The report doesn’t tell you which. It doesn’t even define what “AI infrastructure business” means. Is SpaceX selling raw GPU-hours? Is it reselling third-party cloud capacity with a markup? Is it building its own clusters? The silence is deafening. Reading the silence between the blocks — in this case, the blocks of a financial model — reveals the real structure. BofA’s valuation is a bet on a single idea: that SpaceX can vertically integrate aerospace and compute better than anyone else, and that Anthropic and Google will pay for it. But customer concentration cuts both ways. Two accounts anchor $24.5 billion in revenue. If either walks, the entire projection collapses. We’ve seen this playbook before. In May 2022, Terra’s algorithmic stablecoin was supposed to be “decentralized stability.” The narrative masked a centralized peg mechanism that eventually broke. Here, the narrative is “vertical integration.” The underlying mechanism might just be a reselling agreement with Nvidia or Microsoft — a middleman with a Starlink sticker. Where code meets cultural memory, investors love stories about disruption. But the cultural memory of infrastructure booms is brutal. No company has ever grown from $47 billion to $185 billion in two years without a transformative acquisition. SpaceX would need to deliver tens of gigawatts of AI compute in under 24 months. Starlink’s current bandwidth is nowhere near sufficient for training workloads. Space-based data centers still face severe latency and maintenance challenges. And the report never mentions energy contracts, cooling plans, or networking topology. Those are the kind of details that separate a real engineering roadmap from a PowerPoint projection. The contrarian angle here is sharper than the bullish one. The massive negative free cash flow is not a risk — it’s a feature that guarantees the story lasts. With -$126 billion of cumulative cash burn, SpaceX will need continuous external financing. That means equity dilution, debt, or prepayments from those two hyperscaler customers. The $235 price target presumably assumes a certain share count. If SpaceX has to raise $50 billion via new equity, the target becomes fiction. BofA doesn’t discuss this. Neither does the market. We’re pricing belief, not balance sheets. Following the thread from consensus to chaos, I see the same pattern that drove the DeFi summer of 2020. A narrative is set, a few anchor customers sign letters of intent, and the market extrapolates to infinity. The underlying engineering is unverified. In crypto, we at least had open-source code to audit. Here, we have a bank’s spreadsheet and two press releases. That’s not enough to justify a 87% premium. So what’s the takeaway? The architecture of belief in code — or in this case, the absence of code — determines valuation. SpaceX might genuinely be building the world’s first space-based compute infrastructure. Or it might be a rebranded cloud broker. The next narrative shift to watch is not a price target. It’s a technical announcement: a physical data center footprint, a chip vendor partnership, or a first public demo of orbital compute. Until then, this is a stock story running on faith. The question is whether the market will keep buying the ticket before the rocket actually launches.