Bank of America's August 7 report on SpaceX is the kind of document that makes a macro person pause. The bank repeated its Buy rating and a $235 price target โ 87.5% above the reference price of $125.33 โ but the reason has changed. The bull case is no longer launch volume. It is AI infrastructure. According to the note, SpaceX will generate nearly $24.5 billion in AI revenue in 2026. That represents about 52% of the bank's projected $46.9 billion total revenue. For a company that built its reputation on reusable rockets and low-Earth-orbit internet, this is not a pivot. It is a takeover.
When the algo breaks, the axiom remains. The axiom is not thrust-to-weight ratio. It is the cash flow statement.
Context: From Space Highway to Compute Utility
To understand why BofA made this move, we have to stop thinking of SpaceX as a launch company. SpaceX is a vertically integrated physical network: it builds rockets, launches them, operates the Starlink constellation, and now it wants to sell computational power. The financial projections are extreme. 2026 revenue: $46.9 billion. 2027: $100.7 billion. 2028: $184.8 billion. That is a two-year compound annual growth rate of roughly 99%. By 2028, SpaceX would be approaching the revenue scale of the largest cloud providers on Earth.
The AI revenue line is even more aggressive. BofA assumes AI infrastructure will generate $24.5 billion in 2026, or more than half of total revenue. The bank points to two specific commercial relationships: Anthropic has been contributing revenue since May; a compute collaboration with Google is expected to begin in October. From whitepaper fantasy to ledger reality, this is what it looks like when a transportation monopoly tries to become a compute utility.
But the report says almost nothing about the technology. There is no disclosure of chips, cluster sizes, power sources, data center locations, or latency profiles. There is no mention of whether the infrastructure sits on Earth, in orbit, or somewhere in between. BofA is not selling engineering confidence. It is selling financial projections.
I have spent years in crypto stress-testing liquidity assumptions. The first rule: when the story changes, check the funding. The second rule: when the revenue forecast outruns the technical disclosure, assume the forecast is the dream and the disclosure is the reality check.
Core: What the Numbers Actually Say
The Missing Technology
Let me be clear about what this report is not. It is not a technical analysis of SpaceX's AI capabilities. There is no mention of NVIDIA or AMD, no mention of GPU cluster size, no mention of interconnection bandwidth. There is no mention of power procurement, water cooling, or grid interconnection. For an AI infrastructure business, those are the fundamentals. Without them, the revenue projection is floating on a balance sheet that looks like a black hole.
As someone who spent years stress-testing crypto protocols, I have learned to treat financial projections without technical operators the same way I treat whitepapers without working code: as hypotheses, not evidence. The same beauty trap killed a hundred DeFi projects in 2020. The protocol said 'we will generate yield from real economic activity,' but the only real activity was the token going up. Here, the analog is not a token. It is a $24.5 billion revenue line with no audited contract comps.
When I audit a protocol, I ask where value accrues. Is it in the token? In the treasury? In the protocol's ability to capture fees? With SpaceX, the value accrues to a private company that does not publish financial statements. That is not automatically a problem, but it means every input in BofA's model is an estimate. The model is not verified. It is believed.
Customer Concentration Is the Elephant in the Vacuum
The first thing I did after reading the projection was not to build a discounted cash flow model. I pulled up the customer list. It is exactly two names: Anthropic and Google. That is not a diversified enterprise book. That is a two-customer bet on the continuation of the AI capex supercycle.
Anthropic has been contributing revenue since May. Google is expected to start in October. There is no third name. There is no 'and several others under negotiation.' There is no 'blue-chip financial institution looking to diversify its AI compute.' There are two names. If either relationship stumbles, the 52% AI revenue mix becomes a miss.
In 2022, I watched funds blow up because they assumed a handful of stablecoin issuers would never break the peg. Concentration is not a risk metric until it is a realized loss. Skepticism is the highest form of due diligence. That is the rule I apply to every portfolio I manage. It is the same rule I apply to BofA's SpaceX note.
The revenue trajectory implies that SpaceX will add more than $130 billion of revenue in two years. The largest hyperscalers on the planet do that by building massive physical plants, acquiring rare chips, and locking power supply. SpaceX has not publicly demonstrated any of that. The company has a balance sheet that is hemorrhaging cash.
BofA projects negative free cash flow of $43.6 billion in 2026, $45.4 billion in 2027, and $37.4 billion in 2028. That is roughly $126.4 billion of cumulative negative free cash flow over three years. This is not a minor detail. It is the entire story.
The Funding Gap: Who Pays for the Supercycle?
Negative cash flow at that scale means one of three things. SpaceX is either raising massive amounts of equity, issuing debt, or extracting prepayments from its AI customers. Each path has consequences.
Equity raises dilute existing shareholders. Debt raises the cost of capital and introduces covenants. Customer prepayments can make revenue forecasts look real before the service is actually delivered, but they also shift bargaining power to the customers. If Google and Anthropic are prepaying for compute that has not yet been delivered, they are not just customers. They are co-investors with veto power over the roadmap.
The market doesn't price intentions; it prices the contractual right to receive cash. Until we see the contracts, the $235 price target is a bet on the bank's ability to model a company that is not required to publish financial statements.
This is where my training as a liquidity analyst kicks in. When a company with negative free cash flow asks the market to believe in a more than 90% CAGR, I run a liquidity stress test. The inputs are simple: revenue concentration, customer prepayment behavior, funding access, and execution milestones.
On concentration, SpaceX has two customers. On prepayments, there is no disclosed data. On funding access, SpaceX is one of the most valuable private companies in history, so equity and debt are available โ but not at zero cost. On execution, the only hard date on the calendar is the October Google partnership. If that slips, the entire 2026 AI revenue number becomes a leadership fantasy. If it lands on time, the bank's credibility rises and the next contract likely follows.
The Space Data Center Fantasy
Let's talk about the 'space data center' narrative. It is elegant: solar power in orbit, no atmospheric cooling costs, global coverage via Starlink, and the ability to deploy compute anywhere on Earth. The problem is that orbital data centers have three unsolved problems.
First, latency. AI training runs demand high-bandwidth, low-latency interconnects between thousands of GPUs. A satellite in low Earth orbit is not a good neighbor to an NVLink fabric. The round-trip to a terrestrial GPU cluster is already the bottleneck for distributed training. Adding orbital hops makes it worse.
Second, maintenance. A failed GPU in a terrestrial data center is replaced in minutes. A failed GPU in orbit is a year-long engineering mission. If the entire point of orbital compute is to avoid terrestrial constraints, the service vehicle for routine repairs becomes a project in itself.
Third, collision and debris risk. Every spacecraft has a limited service life. The same infrastructure that makes a space data center provocative makes it operationally fragile. None of these issues is addressed in the BofA report. Maybe the company has solved them. Maybe it has not. The market is being asked to buy the destination without seeing the trajectory.
Now consider the alternative: SpaceX is not building orbital data centers at all. It is building terrestrial AI compute facilities, powered by its reputation for extreme engineering and financed by its ability to raise capital at favorable rates. That is a much more credible path. It also means SpaceX is entering direct competition with AWS, Microsoft Azure, Google Cloud, and the entire GPU-cloud ecosystem.
If SpaceX is just another hyperscaler with a cooler brand, the 52% AI revenue mix is not a structural advantage. It is a demand forecast that relies on two customers continuing to increase orders indefinitely. The market is not valuing a rocket company anymore. It is valuing a startup inside a startup, and the startup is asking investors to accept $180 billion of revenue before anyone has seen a single benchmark test.
Macro Liquidity and Computational Liquidity
From a macro perspective, this is the most important signal I have seen in 2026. A major bank has decided that SpaceX's valuation should be built on AI compute, not launch services. That is not just a stock recommendation. It is a reallocation of investment identity.
In the same way that crypto moved from 'digital cash' to 'macro asset' to 'yield product' to 'AI token' over the last decade, SpaceX is being re-designed by sell-side narrative. The trajectory is not unique. Tesla was a car company before it became an energy-storage company. Amazon was a bookstore before it became a cloud company. Corporate identity is a lagging indicator of capital flows. SpaceX is the newest name to flip.
But the global liquidity environment matters here more than the story. Central banks spent 2024 and 2025 walking a tightrope between inflation and fiscal dominance. Liquidity is not infinite. Every dollar that goes into AI infrastructure is a dollar that does not go into another asset class. When the world's largest technology companies commit hundreds of billions to compute, they are effectively making a macro bet: the AI cost curve will keep falling fast enough to justify the capex.
That is the same bet SpaceX is making. But SpaceX is trying to do it with a private balance sheet, a national-security profile, and a customer base of two. That is not a Microsoft-scale diversified revenue stream. It is a concentrated trade on the belief that AI demand is the new oil. The oil era created Saudi Arabia. The AI era is creating something stranger: a rocket company that might be worth more as a data center operator than as a launch provider.
My own macro theory on 'Computational Liquidity' is simple: the next phase of asset pricing will be driven not by where money is stored, but by where computation is executed. AI models require transparent, verifiable training data. They require energy, chips, and physical security. Crypto protocols are trying to solve the transparency problem with ZK proofs. SpaceX is trying to solve the physical supply problem with rocket-grade engineering.
Neither has proven its model at scale. But the convergence is real. The same institutional investors who were forced to buy bitcoin as an inflation hedge are now being forced to think about AI compute as a strategic reserve. And if SpaceX becomes one of the few entities on Earth that can deliver massive compute capacity under a single contract, it will be treated like a nation-state with a datacenter.
What the Sell-Side Is Actually Selling
BofA's ratings are not independent market signals. They are the output of an institutional sales process. When a bank maintains a Buy rating with an 87.5% upside, it is not saying 'buy the stock.' It is saying 'the market's current price is wrong, and here is a model that justifies a different price.'
But the model is built on assumptions about AI contracts that have not been disclosed. For retail and institutional investors alike, the right response is not blind adoption. It is to ask what happens if the AI revenue assumption is cut by half.
If SpaceX's AI revenue in 2026 is $12 billion instead of $24.5 billion, total revenue drops to roughly $34.5 billion, and the 2028 revenue projection of $184.8 billion loses its anchor. That is not a minor scenario. It is a standard stress test.
What about the target price? BofA's $235 target is based on a private company's secondary shares, not a public market. There is no SEC filing to verify the share count, no audited income statement, no recognized revenue policy. The target is an opinion. It is not a fact.
We don't invest in companies we cannot audit. That is a rule I learned watching crypto natives lose everything to unaudited treasury models. SpaceX is not a scam. But the absence of audited financials, the absence of technical specifications, and the presence of an aggressive revenue forecast create an environment where narrative can masquerade as due diligence. That is exactly how the most painful losses of the last decade began.
There is also the regulatory question. SpaceX is a defense contractor as much as a commercial space company. Starlink is already embedded in critical national-security infrastructure. ITAR and EAR restrictions will determine who can buy SpaceX compute. If the AI infrastructure business is physically located in orbit, every customer relationship triggers a new set of space governance questions. If it is terrestrial, it triggers energy and data sovereignty questions. The report does not address any of these constraints.
Contrarian: The Decoupling That Isn't
Here comes the contrarian angle. The crowd will say this report is either a brilliant catch of an AI transformation or a sell-side fantasy. Both takes are too binary.
The real insight is that SpaceX is not decoupling from the rocket business; it is fusing the rocket business with the compute business. The same engineering discipline that made reusable rockets work is the only asset that can make AI infrastructure cheap enough to justify $184.8 billion in revenue. The rocket is not the story. The cost curve is.
The contrarian bet is not 'SpaceX will fail' or 'SpaceX will succeed.' It is that the market is still pricing SpaceX as either a rocket company or an AI company, when it is actually becoming a physical constraint clearinghouse: owners of energy, spectrum, orbital slots, and manufacturing speed. In a world where GPU supply is limited by power, cooling, and export controls, the entity that controls physical expansion owns the upside.
The blind spot is on the bear side. Bears will point to negative free cash flow and say the model is unsustainable. But negative free cash flow is exactly what a new hyperscaler looks like in its construction phase. Amazon posted negative operating cash flow for years. The question is not whether SpaceX burns cash; it is whether the cash is being converted into assets that generate contracted future returns.
The bear thesis can also be upside-down: if SpaceX's AI contracts include prepayments, the free cash flow numbers may understate the company's true liquidity position. The risk cuts both ways.
The true contrarian position is not 'short SpaceX.' It is to recognize that the market's obsession with either 'rocket company' or 'AI company' misses the only number that matters: the unit cost of compute at an inaccessible physical location. If SpaceX can build compute capacity in orbit or in remote terrestrial sites faster than AWS can build data centers in Virginia, the competitive landscape changes immediately. If not, the $24.5 billion is a rounding error in someone else's cloud bill.
Takeaway: Watch the Ledger, Not the Launch Webcast
So what does this mean for positioning?
First, watch the October Google milestone. If it ships, the 2026 revenue model becomes credible. If it slips, the AI revenue story loses its only hard date.
Second, read the 10-K filings of Anthropic and Alphabet. Prepayments, related-party transactions, and compute purchase obligations will appear before SpaceX ever publishes a number.
Third, treat the $235 target as a statement of belief, not a statement of fact.
The ledger is speaking. Is anyone listening?