Sarah Lin: Nathan, before we get into anything — does it feel to you like there's a grief story buried in a revenue report? Because that's where I ended up.
Dr. Nathan Hayes: A grief story. Walk me toward that.
Sarah Lin: SpaceX's Q3. $14.58 billion in revenue. And for the first time, the Super Intelligence segment — the AI data center side — is more than half of that. 52%. More revenue than Starlink. More revenue than rockets, than launch contracts, than anything with a trajectory toward Mars.
Dr. Nathan Hayes: And yet Starlink is still $2.37 billion in operating profit — the largest profit source in the company despite the smaller revenue share. The SI segment is running at an operating loss.
Sarah Lin: Which is... the part that makes me stop. Because the story being told publicly is diversification, transformation — Elon Musk calling it Yankees versus Little League. But if your biggest revenue line is losing money and the customer behind it, Anthropic, can exit in 90 days by contract — what is the actual story?
Dr. Nathan Hayes: The actual story is customer concentration risk at scale. That 90-day exit clause is not a footnote — it's the fragility at the center of the headline.
Sarah Lin: And SPCX shares moved four percent after the earnings. Four. Which — for a $14.58 billion quarter — that feels like the market sort of... holding its breath.
Dr. Nathan Hayes: Or pricing in exactly what that exit clause implies. Which is the question worth spending time on.
Sarah Lin: But the exit clause is almost — wait, I want to stay on what's inside the profit number first. Because I don't think I actually understand what happened there.
Dr. Nathan Hayes: Right — and this is the part that matters. Think of it like a landlord who reports record rental income, but when you actually read the books, half the profit came from selling bitcoin in the basement. The rent itself barely covered costs. That's structurally what happened in Q3. Net income swung from a $1.1 billion loss to $4.4 billion profit — which sounds extraordinary. But approximately $1 billion of that came from bitcoin price recovery and some unspecified related-party income. Neither of those is flagged prominently in SpaceX's press release.
Sarah Lin: Hold on. A billion dollars. From bitcoin.
Dr. Nathan Hayes: One-time accounting gain. Not operational. And if you strip that out — plus the related-party income — the net income story looks meaningfully different. Now, the SI segment figures compound this because, actually, I want to flag something I find genuinely odd: the revenue number for the Super Intelligence segment isn't consistent across sources. Depending on where you look, it's roughly $7.56 billion, or $6.47 billion, or closer to $6 billion. Those can't all be right, and they haven't been fully reconciled publicly.
Sarah Lin: So where is the money — like, the actual durable money — being made?
Dr. Nathan Hayes: Starlink. $5.25 billion in Q3 revenue, up 69% year-over-year, and $2.37 billion in operating profit. That is the largest operating profit source in the company. The SI segment, for all its headline size, is running at a loss on operations. So the profit floor of SpaceX is satellite internet, not GPU rentals.
Sarah Lin: And CFO Bret Johnsen is still pointing at a hundred billion ARR target by end of 2026.
Dr. Nathan Hayes: Internal expectation, not a confirmed outcome — that distinction is load-bearing. Deutsche Bank's Edison Yu estimated that signed AI compute customers represent only $54.5 billion of run-rate revenue. That's a significant gap to a hundred billion, and it means SpaceX is banking on deals that don't exist yet to validate the headline.
Sarah Lin: So someone reading that press release — someone who doesn't know to look for the bitcoin line, doesn't know about the $54.5 billion gap — they walk away thinking something structurally different from what's actually there.
Dr. Nathan Hayes: Worse than that. Because the thing you're describing — the misread — gets structurally stranger when you ask who the $7.56 billion is actually coming from. Picture a Thursday night — a researcher at Anthropic is running a training job on Claude. Grok engineers, three floors up in the same SpaceX data center, are benchmarking Grok against Claude on the exact same task. SpaceX is invoicing both.
Sarah Lin: Wait. SpaceX is literally subsidizing its own competition against itself?
Dr. Nathan Hayes: That's the mechanism. xAI — which SpaceX owns — is building Grok to beat Anthropic's Claude. And Anthropic is paying SpaceX, under an $84.5 billion compute arrangement, for the GPUs to fight back. The landlord is also the opposing team's coach.
Sarah Lin: And Anthropic can just... leave. Ninety days.
Dr. Nathan Hayes: Ninety days. One of the most well-capitalized AI labs in the world — and that exit term is actually what makes the Cursor acquisition land differently than it was reported. Sixty billion dollars for an AI coding tool, in Q3. Which is, I mean — actually, no, let me be precise about what that signals: SpaceX isn't just a landlord anymore. It's assembling its own AI product stack while renting infrastructure to the companies that product stack will directly compete with.
Sarah Lin: So Google is also in this building. Reflection AI too. But Anthropic is the one that makes the math work?
Dr. Nathan Hayes: Customer concentration risk. Google and Reflection AI are named customers, yes — but the $84.5 billion arrangement is Anthropic. That's the number. Musk calls this Yankees versus Little League. But the Little League team has a 90-day exit clause and every reason to use it once their own data centers are ready.
Sarah Lin: Mm. And the Yankees built a stadium assuming that team never leaves.
Dr. Nathan Hayes: Which is what makes the capacity build the part worth watching next — SpaceX tripled computing capacity to 1.7 gigawatts, and the sub-one-year payback projections embedded in current unit economics? Those collapse the moment Anthropic renegotiates or walks.
Sarah Lin: And that payback math — the sub-one-year number — that only holds if the gigawatts stay full. 0.6 to 1.7 gigawatts, tripled, in essentially one year. That's not an expansion. That's a... um, it's a bet the size of a country's power grid.
Dr. Nathan Hayes: Now here's the mechanism that breaks it: GPU pricing is not fixed. Google, Microsoft, Meta are all scaling capacity simultaneously. The moment spot pricing on H100s drops — which competition makes more likely, not less — that payback window extends. Could be two years. Could be three. The $7.56 billion quarterly figure was built on today's pricing.
Sarah Lin: So the whole model is interest-rate sensitive in a way that isn't being said out loud.
Dr. Nathan Hayes: Price sensitive, yes. And concentration sensitive. Both at once.
Sarah Lin: Meanwhile, Starship — the actual rocket, the thing the whole company was supposed to be — completed its first revenue-generating mission in Q3. Deployed 26 Starlink V3 satellites. Which is genuinely remarkable and somehow... sort of, it got buried. The traditional Space launch segment is what now, $1.4 billion? Against $7 billion in GPU rentals?
Dr. Nathan Hayes: Approximately $1.41 to $1.77 billion. Depending on the source reconciliation issue I mentioned. Either way — a fraction.
Sarah Lin: I keep thinking about whoever was on shift for that Starship mission. Like, a flight director — worked there fifteen years, believed in the Mars thing — watching telemetry as 26 satellites deploy. And that morning's board meeting is about Cursor. Sixty billion dollars for a coding tool. Same quarter.
Dr. Nathan Hayes: The Cursor acquisition at $60 billion is what tells you the SI segment isn't a landlord play anymore — it's vertical integration. SpaceX wants the model, the developer tool, the compute. Which means every dollar Anthropic pays for those 1.7 gigawatts is funding the stack that replaces them. And SPCX only moved four percent on earnings. The market read that tension.
Sarah Lin: Four percent. For a quarter that size. That's not celebration. That's... mm. That's the market going — we see what this costs if it goes wrong.
Dr. Nathan Hayes: And the grief story you started with — I think that's actually the right frame now. Not grief for Mars, exactly. Grief for certainty. SpaceX's durable asset is Starlink. Its fragile asset is Anthropic. And Bret Johnsen's hundred billion ARR target needs roughly forty-five billion in deals that aren't signed yet to close inside six months.
Sarah Lin: Edison Yu's $54.5 billion gap. That's — yeah. That's not an optimistic forecast, that's a structural distance that has to be crossed by customers who can leave in 90 days. Um. I came into this asking whether SpaceX was grieving its mission. I think the actual uneasy thing is — it might not know yet what it's built.
Dr. Nathan Hayes: Which is an honest place to be, actually. The next 90 days will tell more than the headline quarter did.
