Part Two of The Valuation Trap Series
Part One documented the AI valuation trap — token bloat, circular revenue, infrastructure constraints, depreciation stretch, and retail concentration risk. This report applies that framework to the most dangerous IPO in the pipeline.
SpaceX is not an AI company. It is being valued like one.
At $1.75 trillion, SpaceX carries a valuation that exceeds the combined market capitalization of Boeing, Lockheed Martin, Northrop Grumman, and Raytheon. Bloomberg reported SpaceX has sought a valuation above $2 trillion. Morningstar values it at $780 billion — roughly 55% below the $1.75 trillion target.
That gap is not noise. That is an entire mega-cap company made of expectation.
The justification is Starlink — a satellite broadband business with genuine revenue — and an AI and data infrastructure buildout that is burning capital at an accelerating rate.
Starlink is real. Morningstar estimated Starlink generated approximately $10.6 billion of revenue and $5.8 billion of EBITDA in 2025. At 14x EBITDA, Starlink is worth roughly $81 billion. At 20x EBITDA, roughly $116 billion. At 25x EBITDA, roughly $145 billion. That is a powerful business. It does not explain a $1.75 trillion valuation.
The gap between Starlink’s defensible valuation and SpaceX’s total valuation is filled with three things: the launch business, the Mars narrative, and the AI infrastructure ambition. The launch business is profitable but capacity-constrained and faces intensifying competition from ULA, Rocket Lab, and emerging Chinese launch providers. The Mars narrative is not a business. The AI infrastructure buildout is early-stage, capital-intensive, and competing directly against Microsoft, Google, and Amazon on their home terrain.
Elon Musk retains more than 82% of the voting power through the dual-class structure. Public shareholders cannot influence capital allocation in a business that requires tens of billions in ongoing investment. That governance concentration limits external discipline on a company burning capital across four speculative frontiers simultaneously.
SpaceX’s pre-IPO equity is already being accepted as currency. San Francisco real estate listings are seeking SpaceX equity as payment. Pre-IPO equity treated as liquid currency — before public markets have tested the valuation underneath it — is a late-cycle signal. It happened with WeWork in 2018. It happened with Uber in 2019. It is happening with SpaceX now.
If Morningstar’s $780 billion fair value is right, the IPO target is about 124% above fair value. If the IPO prices at $1.75 trillion and the stock later trades to $780 billion, public buyers face a roughly 55% drawdown. If a more generous sum-of-the-parts model marks fair value at $1.25 trillion, the IPO still carries roughly 40% downside to that level.
The retail investor buying SpaceX at IPO is not buying Starlink. They are buying the Musk premium, the AI infrastructure narrative, and the Mars option — none of which are auditable, none of which have precedent, and all of which are priced at multiples that assume simultaneous success.
Valuations peak before business economics mature.
Railroads, autos, commercial aviation, the internet, and shale all followed the pattern: real technology, real capital formation, real overvaluation. The technology mattered. The long-term impact was enormous. The first wave of valuations still overshot the economics.
AI now shows the same late-growth behavior: exploding revenue, record capex, extreme valuations, immature margins, political bottlenecks, accounting stretch, and retail distribution before durable economics are visible.
The market is not asking whether AI matters. AI matters. The question is whether today’s valuations already discount more value than the companies can produce.
The current IPO wave asks public investors to capitalize stories that private investors already marked up.
SpaceX, OpenAI, Anthropic, and related AI names are not coming public as undiscovered assets. They are coming public after massive private-market revaluations. The offering price already includes years of expected growth, margin expansion, capital access, political execution, power availability, and strategic dominance.
The sell-side will emphasize scale, optionality, market leadership, and addressable market size. It will bury token bloat, circular funding, permitting risk, infrastructure bottlenecks, depreciation mismatch, customer churn, pricing pressure, and the difference between gross usage and economic value.
The banks that profit from the IPOs distribute the hype. The insiders cash out at peak valuations. Retail investors inherit the risk.
The bull case requires everything to go right at once.
Token prices must fall far enough to sustain enterprise adoption without collapsing revenue. Usage must rise faster than prices fall. Agentic systems must produce measurable labor savings that justify the 5-30x token multiplier. Enterprises must tolerate the higher AI budgets those systems demand rather than throttling spend the way Microsoft and Uber already have.
The infrastructure must keep pace. Data-center projects must survive local opposition that has already blocked $64 billion in planned capacity. Power and grid capacity must arrive on an 18-36 month timeline when permitting runs 5-10 years. GPU useful lives must support the 4-6 year depreciation schedules that currently inflate reported margins. If any of those assumptions break, the capex story breaks with them.
The financial architecture must hold. Pure-play labs must convert subsidized usage into durable cash demand. Hyperscaler partnerships must prove real customer revenue, not circular capital flows. Gross margins must look like software, not compute resale. Retail ETF flows must keep absorbing concentration risk without triggering correlated selling.
For SpaceX specifically: Starlink must reach 200 million subscribers. AI infrastructure must generate $50 billion in annual revenue by 2030. Launch must maintain monopoly-like pricing power. The Mars program must generate commercial returns within a decade. Every condition is speculative. The valuation prices them as certain.
Public investors must believe all of it before the economics are fully visible.
The story breaks when one of the load-bearing assumptions fails. It does not need to be a catastrophe. A single miss is enough.
Enterprise renewals weaken as token shock spreads beyond Microsoft and Uber. Token prices fall faster than paid usage grows — and the Goldman table in Part One shows how narrow that margin is. Customers cap agentic workflows after discovering that 52% failure rates at full token price are not a productivity gain. Local and open-weight models take share from frontier providers by delivering 80% of the capability at 10% of the cost.
The physical constraints compound the financial ones. Data-center projects stall under permitting, power, water, or ratepayer opposition. Depreciation schedules collide with GPU obsolescence, and the write-downs arrive before the revenue justifies the investment. S-1 filings reveal related-party revenue, customer concentration, or weak cash conversion that the private-market narrative buried.
For SpaceX: the AI infrastructure buildout produces utility margins rather than software margins. The broader AI repricing reduces the hype premium embedded in $1.75 trillion. Competition from ULA, Rocket Lab, and Chinese providers erodes launch pricing power. Starlink subscriber growth slows or ARPU compresses. The downside scenario is not a 20 percent correction. It is a return to Starlink’s defensible standalone valuation — $80 to $145 billion depending on the multiple — with everything else written down to zero.
The break does not require AI to fail. It only requires the economics to miss the multiple.
At $1.75 trillion, the valuation prices Starlink dominance, AI infrastructure success, launch monopoly, Mars commercialization, and governance concentration as simultaneous certainties. Morningstar values the entire company at $780 billion. The gap is $970 billion of pure expectation.
The same forces documented in Part One — token bloat, circular revenue, infrastructure constraints, depreciation stretch, ETF concentration, and the sell-side psyop — apply directly to this IPO. SpaceX is not immune to the AI valuation trap. It is embedded in it. The Starlink cash flows subsidize the AI bet. The AI bet inflates the multiple. The multiple justifies the IPO price. The IPO distributes the risk to retail.
Do not buy this IPO at the offering price. Better risk-adjusted entry points will come after the initial distribution clears and price adjusts to reflect economic reality rather than pre-IPO narrative momentum.
That is not a crash. That is a repricing to reality.
The same forces apply to the upcoming IPOs of Anthropic and OpenAI. The current valuations capitalize temporary inefficiency as if it were permanent demand. When the bloat is cut and real prices take hold, the growth assumptions collapse.
If prices fall, revenue compresses. If prices stay high, adoption stalls. Both paths lead to multiple compression.
Token bloat is the root. Circular hyperscaler revenue is the amplifier. Agentic failure is the adoption brake. Price deflation is the trigger. Multiple compression is the outcome.
Customers will stop paying for noise. The valuation multiples follow.
