1+1=Fish? A short note on SpaceX IPO

23 May, 2026

¹In 2021H2, I wired $225K into an SPV to buy SpaceX shares at a valuation of ~$100B. It turned out to be a fraud. I was fortunate to get the principal back. SpaceX then looked like a monopoly: ~80% of mass to orbit, pricing power on launch, and a connectivity business that no other company can build, and I wanted a piece of the pie.

The S-1 dropped this week. We finally got to see the numbers, and they're bad.

Ignore EBITDA

Before getting into segments, every number in this filing has an adjusted EBITDA. For a satellite business, that's the wrong line. Starlink satellites have a rough 5-year orbital life before they deorbit and burn up. The depreciation on those satellites is a real, recurring, cash cost of staying in business. It is not a non-cash accounting artifact you get to add back. Same for the AI compute clusters, GPUs are not buildings. They're 3-6 year assets in a market where the next generation makes the previous one economically obsolete quickly.

So I'm going to use operating income throughout. Adjusted EBITDA is for selling the deal, not for valuing the business.

Segment Numbers

Space (launch): 2025 revenue $4.1B, operating loss $(657)M. The loss is entirely Starship R&D, $3.0B in 2025 plus another $930M in Q1 2026. The underlying Falcon business generates roughly $2.3B of operating income on $4B of revenue. The monopoly is real. The reported number is bad because they are funding the next vehicle out of the current one's cash flow.

Connectivity (Starlink): 2025 revenue $11.4B, operating income $4.4B. Op margin ~39%. YoY growth of 50% on revenue, 120% on operating income, not bad. Q1 2026 op income $1.2B on $3.3B revenue, margin 36%. 10.3M subscribers across 164 countries. This is the one good business in the filing. And the depreciation is already in that 39% margin.

AI (xAI, consolidated February 2026): 2025 revenue $3.2B, operating loss $(6.4)B. Q1 2026: $818M revenue, $(2.5)B operating loss. AI capex was $7.7B in Q1 alone, or $31B annualized, against $3B of revenue. The losses are accelerating faster than the revenue. This is a money furnace with a satellite business strapped to it.

And remember, the Q1 2026 $(2.5)B operating loss reflects only a fraction of the depreciation from $7.7B of Q1 capex. As those GPUs roll onto the depreciation schedule over the next 6 years, the operating loss gets worse before the revenue catches up. If it ever catches up.

Starship V3

The first V3 flight test happened tonight. My guesstimate is 22 tons of dummy payload, apogee 195 km, suborbital. The vehicle has not demonstrated enough delta-v to reach orbit so far.

This is one flight and the program will iterate. But the S-1 specs V3 at 100 tons to orbit fully reusable, expects payload delivery to orbit in H2 2026, six months from now, and bases its forward case on a single Starship deploying 60 V3 satellites per launch. This is highly unlikely.

The Starship optionality embedded in any valuation above $1T just got more expensive to underwrite.

The TAM is comedy

The S-1 claims a $28.5T total addressable market. Of which $22.7T is "enterprise AI applications." That is roughly 22% of global GDP. The filing also helpfully excludes China and Russia from the comp set. Why not make the number even larger? I am not going to insult the reader.

What is this worth

Sum-of-the-parts on operating income, at multiples I'd consider generous:

• Starlink: $4.4B 2025 op income, call it $6B 2026E if growth holds. At 50x op income (rich for any infra business, but Starlink deserves a premium) → $300B. Push it to $500B if you believe V3 + Starlink Mobile scales the way they say. After today, the V3 capacity ramp is on a longer fuse.
• Space (Falcon + Starship optionality): ~$2.3B of underlying Falcon op income at 50x gets you ~$120B. Add some Elon premium, call it $150B all-in.
• AI segment: this is destroying $6B+ of operating value per year on a run-rate basis. At a generous read, assume xAI is worth what the last private round said it was, maybe $100–150B. At an honest read, it's probably CoreWeave's value + Twitter's value, roughly $70-80B.

Sum: $525–575B base, ~$800B if every assumption goes right. The secondary market is asking $1.5–2T. The entire delta is paying for the orbital data center story, the $28.5T TAM, and the assumption that AI capex keeps compounding from here.

The AI read-through

The most interesting part of this filing to me is AI.

Most frontier AI economics are hidden inside private companies, cloud credits, compute partnerships, strategic investments, and hyperscaler capex. SpaceX gives us a clearer window. The scene looks ugly.

Revenue is growing, but spending is growing faster. xAI is losing roughly $10 billion a year on a Q1 run-rate basis, before the full depreciation cycle of the current capex wave hits the income statement. Google's Gemini 3.5 Flash also launched at a much higher price than prior Flash-family models, with current listed pricing at $1.50 per million input tokens and $9.00 per million output tokens. Closing in on 3.1 Pro pricing ($2/$12).

If inference were already a clean, high-margin software business, we would not see price hikes, tighter quotas, and massive infrastructure spending all at the same time.

My guess is that the AI capex cycle is closer to exhaustion than the market thinks. Within a year, I expect at least one hyperscaler to slow down or guide down the growth rate of AI capex.

So

I might buy SpaceX under $1T. $1T is already a generous bid, and after today's V3 flight, it's a stretch. $1.5T requires the AI segment to be worth more than the rest of the business combined. $2T is just ludicrous.