Three Companies in a Trench Coat

The S-1 hit Wednesday. $1.75 trillion target. $75B raise. June 12 debut under SPCX. Most of the coverage I read this week quoted one number, the consolidated 2025 revenue of $18.7B, as if it described a single company. It does not. It describes three businesses stapled together at the cap table and priced as one number.

I have been reading the filing on the commute and I keep coming back to one page in the segment disclosures.

The income statement is a magic trick.

Here's the deal. Standalone SpaceX did somewhere between $15B and $16B in 2025. The merged entity reports $18.7B because xAI and X got folded in via common-control accounting in February. xAI had already swallowed X in March 2025. The auditors recast every historical period to make the consolidation look clean, which it technically is. The unit you are analyzing changed.

Strip the three segments apart and the picture is unrecognizable.

Starlink (Connectivity) did $11.4B in revenue, $4.4B in operating income, $7.2B in segment-adjusted EBITDA. Launch (Space) did $4.1B in revenue and lost $657M on purpose, because the segment is mostly Starship R&D right now. Any quarter that does not lose money there is a quarter you should worry about. xAI (AI) did $3.2B in revenue, lost $6.4B in operations, and consumed $12.7B of capex. In Q1 2026 alone, the AI capex line annualizes north of $30B.

Consolidated 2024 net income: $791M profit. Consolidated 2025 net loss: $4.94B. The company did not deteriorate. The accounting did.

If you came up writing systems, you recognize the shape immediately. Three services in one repo, sharing a billing table, masquerading as a monolith. Ouch.

The Anthropic deal is on page 13.

Most of the AI-side coverage I read this week missed it. Tucked into the cloud services disclosures, SpaceX and Anthropic signed Cloud Services Agreements in May 2026 worth $1.25B per month through May 2029. Roughly $15B a year. Up to $45B over the term. Either side can walk on 90 days notice.

Sit with that for a second. Anthropic, the lab I wrote about two weeks ago when it raised at a $1T mark, is now renting compute from xAI's COLOSSUS and COLOSSUS II clusters. Andrej Karpathy, who joined Anthropic this month to run pre-training research, will be training Claude on Musk's silicon. Grok 5 trains on the same campus.

I underwrite the AI infrastructure layer differently after reading that clause. The clean story has been that the frontier labs own their compute. The actual story, in writing, is that one of the loudest competitors will rent his cluster to anyone who can write a $15B annual check.

Frankly, the 90-day termination clause is the part I cannot stop reading. xAI is monetizing spare capacity, not committing to a customer. The day Grok 5 needs the cluster back, Anthropic gets 90 days to source $15B of replacement compute. Good luck.

Connect your email, calls, product data and more, and Attio instantly builds your CRM with enriched data and complete context. Whether you’re running product-led growth or enterprise sales, Attio adapts to your unique GTM motion.

Then Ask Attio to plan your next move.

Run deep web research on prospects. Update your pipeline as you work. Find customers and draft outreach emails. Powered by Universal Context, Attio's intelligence layer, Attio searches, updates, and creates across your data to accelerate your workflow.

Ask more from your CRM.

Ask Attio

The TAM is a vision document with a spreadsheet on top.

SpaceX calls it "the largest actionable TAM in human history." $28.5 trillion. Space at $370B, Connectivity at $1.6T, AI at $26.5T. AI is 93% of the number. Enterprise AI applications alone, a category the company does not yet sell into at scale, is $22.7T. The filing also notes, almost as an aside, that the estimates exclude China and Russia.

The S-1 itself concedes that several of these markets, including in-orbit manufacturing, lunar energy production, and asteroid mining, do not yet exist. The public market investor is being asked to buy at $1.75T is being asked to buy that the TAM gets built.

I am not dismissing the number. Musk has been right about market sizes nobody believed before. The 2006 SpaceX deck looked equally detached from reality at the time. So did the internet-at-$1T claim in 1995. The bet is not that $28.5T is correct. It is that 90x revenue is closer to right than it looks.

The part engineers should sit with is that the businesses generating actual cash today are under 7% of the claimed addressable market. Yep, pretty much. You are buying a satellite ISP and getting paid back for everything else in optionality.

Starlink's ARPU is the line nobody is screaming about.

Here's the thing the headline missed. Starlink subscribers more than doubled, 5.0M to 10.3M, year over year. Beautiful chart. Then look at revenue per subscriber: $99 in 2023, $91 in 2024, $81 in 2025, $66 in Q1 2026. ARPU fell 23% year over year. Volume is the only thing driving the topline at this point.

For a connectivity business priced at a 90x-plus revenue multiple, the direction of the per-unit line matters more than the topline does. There is a subscriber count, somewhere ahead, at which the price drop starts outrunning the adds. The date is a function of how aggressively Starlink keeps pushing into lower-priced international markets. That is the same plan that produced the 23% drop in the first place.

I keep relearning this one, both as an engineer and now as somebody paid to underwrite. Aggregate growth covers a busted unit metric for exactly as long as the market lets it.

As for me?

I am going to keep reading the filing this weekend. The retail allocation is going through Schwab, Fidelity, Robinhood, SoFi, and ETRADE at the IPO price, which is an unusual choice for the largest IPO ever attempted and worth a separate post. Musk holds 85.1% of voting power on Class B at ten votes per share. On Monday a federal judge handed him a fresh loss in his suit against OpenAI. Same week, by the way.

I do not own ten-vote stock anywhere else in my book. I will probably not buy this one either.

What I am unsure about is whether the three-segment teardown matters to the price the market sets on June 12. Public markets are not in the habit of unbundling monoliths in their first three trading days. The financials are sitting in the filing in plain English. Whether the market reads them, or just buys the ticker, is a different question, and I do not have a clean answer. Most likely just buys the ticker.

I will read the pricing amendment when it lands. Probably on my phone, in line for coffee, like everybody else.

— SWEdonym