May 18, 2026
SpaceX is heading to the public markets with the largest IPO in history — and the financial picture buried in its draft S-1 is the story founders should be paying attention to. The combined SpaceX-xAI entity generated $18.67 billion in revenue in 2025 and posted a $4.94 billion net loss. One year earlier, standalone SpaceX cleared roughly $791 million in profit. The single biggest reason for the swing is xAI, which burned an estimated $14 billion in cash against just $3.2 billion in revenue. Starlink, the satellite internet business, is carrying every other operation on its back.
Last updated: May 18, 2026
How much did SpaceX lose in 2025?
SpaceX posted a $4.94 billion net loss on $18.67 billion in revenue in 2025, according to draft S-1 financials reviewed by Reuters and Bloomberg ahead of the public prospectus filing expected this week. That is the first time SpaceX has disclosed audited financials, and it is the first window most investors have ever had into the combined entity created by the SpaceX-xAI merger.
The loss is a hard reversal from 2024, when standalone SpaceX cleared an estimated $791 million in profit on $14.02 billion in revenue. The headline numbers obscure what is actually happening underneath: the underlying SpaceX businesses are not bleeding. The bleeding is concentrated in one segment.
What is Starlink earning, and what is xAI burning?
Starlink is the only profitable segment in the combined entity. The satellite internet business generated $11.4 billion in revenue in 2025, up roughly 50% year over year, and produced $4.42 billion in operating income. As of February, Starlink served 10 million subscribers across 160 countries. The rocket launch business added another $4.1 billion in revenue.
xAI is the offset. The AI lab consumed approximately $14 billion in cash in 2025 against $3.2 billion in revenue — meaning xAI alone burned more cash than every other SpaceX operation generates. Total capex jumped from $4.2 billion in 2024 to $20.74 billion in 2025, with AI infrastructure spending alone rising from $5.6 billion to $12.7 billion. The story the financials tell is clean: Starlink is funding a separate, much larger bet on AI compute, and the rest of SpaceX is running at roughly break-even while that bet plays out.
When does SpaceX go public, and at what valuation?
The public S-1 prospectus is expected to hit SEC EDGAR between May 20 and May 22. The IPO roadshow is scheduled to begin June 4, with pricing on June 11 and trading debut June 12 on Nasdaq under ticker SPCX. SpaceX is targeting a $1.75–2 trillion valuation and a raise of up to $75 billion — the largest IPO in history by a wide margin.
A 5-for-1 stock split is processing this week, adjusting the fair market value from $526.59 to $105.32 per share. The split is designed to widen retail access ahead of the listing, and SpaceX is targeting up to 30% retail allocation — roughly three times the typical institutional-heavy IPO split. Musk has publicly committed not to sell any shares in the offering and will retain roughly 42% equity and 79% voting control through a dual-class share structure.
The cross-subsidy bet
What investors are about to underwrite at a $2 trillion valuation is one of the most ambitious cross-subsidy structures in modern corporate history. A single profitable segment — Starlink, at $4.4 billion in operating income — is the collateral on a separate $14 billion-a-year bet on AI compute. The launch business holds the door open. The AI business absorbs the cash.
That structure only works under two conditions. First, the profitable segment has to keep growing. A 50% year-over-year revenue jump at Starlink is the load-bearing assumption in the entire S-1, and any slowdown there changes the math on what SpaceX can keep funding. Second, the AI bet has to eventually produce returns commensurate with the burn. xAI is not yet a major commercial revenue line — $3.2 billion in 2025 is a fraction of what Anthropic and OpenAI are pulling in — and the path to justifying $14 billion in annual cash consumption runs through products that mostly do not exist yet.
Investors are not being asked to bet on whether the math works today. They are being asked to bet on whether Musk can engineer it to work over the next five years, with the same operational playbook that took SpaceX from near-bankruptcy to the dominant launch provider on the planet.
What founders should take from this
The Scaringe-style founder-credibility premium that drove Mind Robotics’s $3.4 billion valuation last week is the same dynamic at work here, just at a different scale. The S-1 is asking investors to underwrite a $14 billion-a-year bet on the basis of one founder’s track record, one profitable segment, and a structural assumption that the cross-subsidy holds.
For founders, the structural lesson is the part worth studying. There is a real playbook embedded in the SpaceX financials: build one segment that throws off durable cash, use it to underwrite a much larger second bet, and keep enough voting control that you can ride out the years where the second bet is unprofitable. That sequence — profitable wedge, then ambitious moonshot, with the wedge funding the moonshot — is more replicable than the Musk-specific brute-force version makes it look.
The control structure is the other half. A dual-class share structure that leaves the founder with 79% voting power after a $75 billion raise is a deliberate design choice, and it is what makes the cross-subsidy bet possible at all. A board accountable to public shareholders does not approve a $14 billion-a-year burn on an unproven business indefinitely. A founder who controls the vote does. How founders keep control after an IPO is the single most consequential governance question in the 2026 IPO class, and SpaceX is the clearest live example of the trade.
What this signals about the AI capex cycle
The capex line in the S-1 is the broader market signal. SpaceX’s AI infrastructure spend more than doubled in a single year, from $5.6 billion to $12.7 billion, and total capex hit $20.74 billion. That is alongside Anthropic’s $100 billion AWS commitment, OpenAI’s multi-year compute deals, and a wave of nuclear and geothermal power agreements from the same cohort. Capital is concentrating at the infrastructure layer faster than the application layer can absorb it.
For founders building outside the foundation model layer, the implication is that the cost of compute is going to fall — eventually — as this infrastructure comes online, but the timing is opaque. The companies building the infrastructure are doing it on the assumption that demand will catch up. The SpaceX S-1 puts a $14 billion-a-year price tag on that assumption.
The bottom line
SpaceX is taking $4.94 billion in losses to the public markets and asking $2 trillion for the privilege. The story underneath is simpler than the headline. One profitable segment — Starlink — is doing all the work, and the market is willing to value the combined entity at a record valuation because of the founder, the cross-subsidy structure, and the bet that the AI infrastructure spend pays back over a horizon longer than a quarterly earnings cycle.
This is the IPO that defines how 2026 capital is being allocated: founder credibility over current profitability, cross-subsidy over focus, voting control over governance, and a multi-year capex bet that demands investors look past the loss line. Whether it works will be the story for the rest of the decade.
