← sources

SpaceX at $1.8 trillion: is the real investment case rockets, Starlink or AI compute?

source: reddit (reddit:stocks)published 2026-11-02 09:36ingested 2026-11-03 02:40

retained text

# SpaceX at $1.8 trillion: is the real investment case rockets, Starlink or AI compute?

I've been looking through SpaceX's third-quarter results, and one thing strikes me: many discussions still treat it primarily as a space-launch business. The actual financial mix has changed remarkably fast. In Q3 2026, the Space segment produced $1.77 billion in revenue, Connectivity (mostly Starlink) $5.25 billion, and Super Intelligence $7.56 billion. The last category now includes data-center capacity, models and software, and the recently acquired Cursor.

That changes the way I would evaluate the company. The case for SpaceX used to rest heavily on its ability to launch cheaper than competitors, then turn that launch advantage into a satellite broadband business. This part is becoming demonstrably real. Starlink had 13.6 million consumer subscribers at September's end, with Connectivity producing $2.37 billion in quarterly operating income. You can argue about the long-term margin, but this is not some revenue-less space venture.

The catch is that consumer ARPU has fallen from $83 to $69 over the past year. Expansion into lower-income markets means the number of subscribers is not the whole story. And Starlink faces competitors, regulators and limits imposed by physics. A great product isn't guaranteed an unlimited customer base at American pricing.

Meanwhile, AI makes the valuation much harder. SpaceX had 1.7 GW of nameplate computing capacity at quarter end, and SI revenue is growing extraordinarily quickly as new cloud agreements become active. The thesis is that intelligence demand may rise faster than the supply of energized data-center capacity, allowing the owners of ready-to-run compute to earn scarcity rents. SpaceX can also consume that compute in Grok or its applications rather than selling all of it to third parties. That flexibility has value.

But look at the cash flow statement: $48.1 billion spent buying property and equipment in the first nine months versus $13.1 billion in operating cash flow. That leaves roughly negative $35 billion on a simple operating-cash-minus-capex basis. Management also expects $35–40 billion in capex in Q4 and has announced a proposed $40 billion GPU financing lease. A lot of today's earnings could end up supporting asset replacement rather than shareholder distributions if AI compute pricing falls or chip replacement cycles remain short.

Orbital compute is the genuinely fascinating wildcard. SpaceX's published Starmind design talks about 175 kW of average compute per satellite. To reach one gigawatt you would need roughly 5,700 such satellites, even before accounting for real-world losses. That shows why radically cheaper, frequent Starship launches might matter beyond broadband. It also makes the engineering problems hard to ignore: thermal radiation, hardware reliability, servicing, launch costs and orbital congestion are all nontrivial. I don't think it belongs in a central valuation yet.

At roughly $1.83 trillion in market cap as of October 30, this is not remotely a cheap conventional investment. My rough five-year range is $0.9 trillion in a bear case, $4.2 trillion in a base case, and $9 trillion in a bull case, with share dilution reducing the return from those market-cap outcomes. The result is attractive only if you give meaningful probability to AI and connectivity becoming enormous profit pools, not simply large revenue lines.

I'm curious how others would underwrite this. Would you value SpaceX chiefly as a launch-enabled communications utility, as an AI infrastructure business, or as an entirely new industrial platform? And which single metric would make you change your mind fastest: Starlink ARPU, compute cash returns, or Starship reuse?

document fetched 2026-11-03 00:32 · canonical https://reddit.com/r/stocks/comments/1xlstwi/spacex_at_18_trillion_is_the_real_investment_case