The code does not lie; only the founders do. And Bank of America is not a founder. It is a propagandist for the narrative.
On August 7, an unremarkable date, an investment bank released a report that treats a private rocket company as if it were the next hyperscaler. Buy rating. Target price: $235. Reference price: $125.33. The target implies an 87.5% upside. The stated reason is no longer the Falcon program. It is not Starship's payload capacity. It is "AI infrastructure."
This is dangerous financial fiction. Partially true fiction. SpaceX has real rocket revenue. Starlink has real subscriber growth. But the report's AI revenue projection—$24.5 billion in 2026—is 52% of all projected revenue that year. Two customers. One that started paying four months ago. Another that allegedly goes live in October.
I am a security auditor. I do not trust narrative verification. I examine the space between whitepapers and bytecode. That is where the bodies are buried.
The Report, Such As It Is
The parsed material lacks source attribution. No publication, no author, no timestamp beyond the August 7 date. The year is inferred to be 2025. This alone should raise alarm. A report with this much market-moving potential and no attributable authorship sits at the same epistemic status as an anonymous screenshot of a Telegram group chat.
What does the report claim? It maintains a buy rating and the $235 price target. It projects total revenue to jump from $46.9 billion in 2026 to $100.7 billion in 2027, then to $184.8 billion in 2028. That is a two-year compound annual growth rate of approximately 99%. For reference, no large-scale infrastructure company in modern history has sustained such a growth curve without serial acquisitions.
It projects free cash flow to remain negative across all three years: -$43.6 billion in 2026, -$45.4 billion in 2027, -$37.4 billion in 2028. The cumulative cash burn sums to -$126.4 billion. The report frames this as evidence of a capital expenditures supercycle. I frame it as a liquidity trap wearing a construction helmet.
It identifies two AI customers. Anthropic, with revenue contribution beginning in May. Google, with compute cooperation expected to start in October. No third customer. No contract details. No pricing model. No duration.
That is the entire information base supporting the AI-infrastructure transition. As a forensic document review, this is a public offering of a security with no prospectus.
The Audit That Was Never Performed
When I audit a smart contract, I begin with the storage slots. I map the functions. I test the access controls. I simulate adversarial conditions. The BofA document cannot survive that process. Because there is no contract. There is only an assertion.
Start with the underlying asset. What does SpaceX's AI infrastructure actually consist of? The report does not answer. The possibilities are starkly different.
Ground-based data centers are a crowded market. AWS, Azure, Google Cloud, and CoreWeave are buying the same NVIDIA GPUs at the same Taiwan Semiconductor prices. The unit economics are well understood—and they are brutal at the margin. A late player entering this market must either accept lower utilization or negotiate anomaly pricing on power. Neither is a structural moat.
Orbital compute is a different animal. In low Earth orbit, the sun illuminates the panels roughly 60% of the time. No air cooling is needed in vacuum. The space is, in a sense, free. But orbital mechanics impose eclipse cycles. Radiation causes single-event upsets in the most qualified chips. Space debris can puncture the payload. The maintenance latency is not hours—it is days or weeks. A single failed reaction wheel can make the entire node nonfunctional.
The report does not say which model it is pricing. An auditor would flag this as "unsubstantiated claim." A salesman would call it "optionality."
Then examine the revenue timeline. Anthropic has been paying since May. Google starts in October. The gap between these events and the report's 2026 projection of $24.5 billion in AI revenue is less than a year. In that time, SpaceX must scale from two customers to a revenue base of $24.5 billion—meaning compute capacity measured in thousands of high-end accelerators. The physical buildout of such capacity, regardless of whether it is terrestrial or orbital, typically takes 12 to 24 months from design freeze to production inference.
Unless the customers are prepaying for capacity that is not yet built, and revenue is recognized on a milestone basis. If so, the report is not projecting earned revenue. It is projecting future cash receivable from construction contracts. The accounting treatment matters enormously. It should have been disclosed.
Worse, the customer concentration risk is worse than a two-validator oracle. Anthropic is a player whose revenue model is itself facing precarious dependency on venture funding rounds. Google is a giant with its own TPU supply chain. If Google decides to build more in-house capacity, the SpaceX contract becomes optional. The termination clause is the single most important line in the deal. It is invisible to the public.
The Negative Free Cash Flow Abyss
Let me spend more time on the numbers. The report projects FCF of -$126.4 billion across 2026-2028. Even for a company with SpaceX's access to capital, this is a seismic number. To put it in perspective, this level of annual negative cash flow exceeds the entire GDP of a mid-tier sovereign. I am not comparing the company to a country. I am comparing the magnitude of the burn to the size of a real economy.
How does a private company finance that? Four paths.
Equity issuance. This dilutes existing shareholders. The $235 target price must be adjusted for the massively increased share count that would be required. The report does not address this.
Debt issuance. This adds fixed-interest obligations to a balance sheet that already has negative cash flow. In a rising rate environment, the cost of servicing $100 billion-plus of debt could exceed the AI gross profit.
Customer prepayments. If Anthropic and Google are prepaying for compute, those prepayments are liabilities until the compute is delivered. If SpaceX fails to deliver, it owes the money back—with penalties.
Strategic investment from sovereign wealth funds or national security entities. That comes with conditionality, surveillance, and policy strings.
The report presents none of these options. It just shows the negative FCF and the revenue growth side by side, as if one explains away the other. In my audit experience, this juxtaposition is exactly what a fraudster does: they show liquidity outflow as "investment in future growth" and hope the analyst does not interrogate who is paying for the invested amount.
The Yield Farming Equivalence
The BofA report's structure is identical to the liquidity-mining yield farms of DeFi Summer. The token price goes up because the protocol pays its own token as a reward for depositing liquidity. The TVL number goes up. The "revenue" is actually self-supplied subsidy. When the incentive stops, the users vanish.
SpaceX is a real company with real launch revenue. I am not equating it to a scam farm. I am equating the valuation methodology. The report is valuing the AI business at a premium based on two customers it does not name in detail, and it is ignoring the fact that a capital-intensive supercycle with negative $126 billion FCF requires the founders to keep appeasing capital providers—whether the AI revenue arrives or not.
The rug was pulled before the mint even finished. That sentence, written for a token-launch post-mortem, applies here with a twist: the market capitalization of SpaceX's future AI business is being minted before the compute is delivered. Investors are buying the promotional period.
The Missing Technical Specification
From 2020 to 2021, I stress-tested the Compound protocol's interest rate model. I found a rounding error in the borrow-rate calculation that could cause insolvency under high volatility. The core devs acknowledged it. They did not fix it immediately; they prioritized liquidity incentives. That decision taught me that financial incentives can outrank software security when the market is hot.
I see the same hierarchy in the BofA report. The financial incentives—the AI revenue projections, the price target—are placed above the technical constraints. The report does not disclose the chip procurement plan. It does not disclose how a multi-billion-dollar AI infrastructure buildout handles the simultaneous shortage of grid power, the packaging bottleneck at TSMC, and the export regulations around H100/A100-class accelerators. Those are reality constraints. The report treats them as footnotes that do not exist.
During the 2022 Terra collapse, I audited the Luna Classic mechanism post-mortem. My report showed that the algorithmic backstop was mathematically impossible to sustain. The oracle manipulation vectors were not the cause; they were accelerators. The cause was the absence of a stable supply floor. The same logic applies here: the $235 price target is the stablecoin peg. The AI revenue is the algorithmic anchor. It only holds if the anchor is verified.
Reentrancy is not a bug; it is a feature of trust. That is what I wrote after the MetaBeast incident. The owner function in MetaBeast lacked access controls. Anyone could mint infinite tokens. The project launched anyway. The market trusted because the NFT hype was overwhelming. Two weeks later, the rug was pulled. My short position on the governance token paid off because I did not trust the narrative—I trusted the code paths.
In the SpaceX case, there is no code. There is only a roadmap. The "smart contract" is the set of contractual commitments between SpaceX, Anthropic, and Google. If those contracts have adequate access controls—legally—then the revenue is real. But we cannot verify them.
What Would Change My Mind
I need to be precise. I am not bearish on SpaceX. I am bearish on the intelligence of accepting a revenue narrative without underlying data. To convince me, the report—or SpaceX itself—would need to release.
One: Customer contracts, redacted to the minimum. I need the commercial terms, minimum take-or-pay volumes, exclusivity windows, and termination conditions. The fact that these are not public is understandable, but the analyst should have described the order of magnitude of the prepayments.
Two: Capital expenditure allocation. Break down the $126 billion. How much goes into rocket manufacturing versus satellite production versus ground data centers versus power grid interconnection? The clarity of the allocation determines whether this is a sensible industrial plan or a scattershot reaction to AI hype.
Three: Unit economics. What is the price per FLOP? What is the expected utilization rate? What is the gross margin of the AI business? A company with negative free cash flow can still be a sound investment if the marginal economics are structurally superior to AWS. But the report does not even state the power cost per megawatt-hour.
Four: Technical architecture. Show me the compute. Not a render of a futuristic orbital server farm. Show me the rack density, the cooling approach, the network topology, the failure containment. I want the data equivalent of a verified smart contract address.
Five: Dilution plan. What is the share-count evolution over the next three years? If the company will issue 40% more equity to fund the capital expenditure, the $235 target fails against a constant-dollar basis.
None of this is unreasonable. It is the bare minimum you require in an audit before issuing a final report. The bank is issuing a buy recommendation with less disclosure than my non-binding review of a $200,000 token sale.
The DePIN Counter-Factual
This analysis is not a distraction for blockchain readers. The crypto ecosystem is actively building decentralized physical infrastructure networks—DePIN—to serve AI compute demand. Render pushes rendering workloads to idle GPUs. Akash rents out compute on a marketplace. Grass lets users share unused bandwidth. These projects are routinely dismissed by serious financial institutions as "unproven" or "permissionless bandwidth arbitrage."
Yet those same institutions are willing to accept a two-customer, zero-disclosure AI revenue projection for a company that has not yet proven it can operate a data center in orbit. The cognitive asymmetry is humorous. DePIN projects at least publish their smart contracts. They expose their incentive equations and their token emission schedules on-chain. The BofA report publishes nothing auditable.
In DePIN, you can check the utilization rate of a GPU node before you rent it. In the traditional market, you are told to trust the buy-side analysts who have been wrong about everything from Theranos to FTX. I am not saying every DePIN project is sound. I am saying the verification tools available globally are far better for distributed compute markets. The SpaceX AI narrative is not competing with better infrastructure. It is competing with better disclosure.
The Regulatory Elephant
The report ignores export controls and data sovereignty. AI infrastructure is a strategic asset. The US government will not allow SpaceX to sell orbital compute to any entity in contested jurisdictions. The client list will be restricted to NATO-aligned players. That is a constraint. It is also an opportunity. But the report's revenue projections are built as if the whole world is the addressable market. It is not.
MiCA-style frameworks across the Atlantic, and their equivalents elsewhere, are tightening the standards for digital infrastructure. Any company operating data centers in orbit is operating in every jurisdiction it passes over—and none at the same time. The legal vacuum is a risk to revenue recognition. The report's silence on this is another sign that the document is a marketing artifact rather than a technical assessment.
The Bulls' Stash
I promised a contrarian angle. Here it is. The bulls have a strong stash.
First, SpaceX's reusable rockets are not just cool. They are the only machines on Earth that have demonstrated rapid, low-cost orbital reusability in a meaningful sense. That is the base of new unit economics for orbital infrastructure.
Second, Google is not stupid. When Google signs a compute deal with SpaceX, it is probably not because Google has no data centers. It is because Google sees a strategic option in a certain orbital capacity that cannot be replicated by terrestrial expansion alone. The signal quality of Google's intent is higher than the signal quality of the analyst's report.
Third, the AI demand reality is unambiguous. Compute is the new printing press. Marginal useful compute—say, for frontier research and inference—is being consumed at a rate faster than any industry in history. Even a tiny share of that demand, secured through credible customers, can justify a significant investment in SpaceX's infrastructure arm.
Fourth, negative free cash flow is not always a fatal condition. Amazon burned enormous cash for years. Tesla did the same. The market rewarded both because the capex was linked to structural cost reductions. If SpaceX harnesses the capex to build lower-cost, higher-energy-efficiency compute, the FCF trajectory could reverse in later years. The report's flaw, again, is the absence of the unit economics—not the existence of the FCF deficit.
That is a fair reading. I would not call it the most probable outcome. But I would call it possible enough to keep the narrative alive.
The Bottom Line—And I Mean the Actual Bottom Line
The bank will not revise the target price based on this article. The market will not crash at 10am because a Warsaw auditor wrote a skeptical post. That is fine. I am writing for the people who read whitepapers and then look at the contract code. I am writing for the operators who ask "where is the revenue coming from" before they ask "what is the multiple."
Let me close the loop.
The code does not lie. Only the founders do. And the founders of this narrative are not in the cargo bay. They are on the sell side.
The rug was pulled before the mint even finished. In this case, the "mint" is the AI-infrastructure revenue recognition. It has not finished because it has not truly begun. The projection is the promise. The audit is incomplete. The exit liquidity—the investors who need to believe in $235 to feel comfortable holding at $125—is already in place.
I don't trust the audit; I trust the gas fees. And the gas fees of this report are negative $126 billion. That is the on-chain proof.
We are watching a multi-billion-dollar bet on a narrative that requires two counterparties, no public documentation, and a technical miracle. It might work. It might also be the biggest exit-liquidity event in aerospace history. In a sideways market, with capital rotating from the last cycle's losers to this cycle's winners, the responsible answer is the one that locks: verify the infrastructure. If the code arrives, I will audit it. If the contracts are disclosed, I will read them.
Until then, the only honest position on a $235 target with negative $126 billion of projected free cash flow is the observation that the code does not lie.
It just has not been written yet.