SpaceX closed Monday with a market capitalization near $1.87 trillion, down only about 1.3 percent from Friday and roughly flat against a six-trading-day window that saw Nvidia's succession drama and a six-week NVDA losing streak, while Nvidia's September 30 13F showed a roughly 6 percent add to its SpaceX stake even as it trimmed Synopsys and telecom turnarounds—so allocators are testing whether SPCX can absorb whatever Tuesday's fiscal Q3 print says because Starlink still funds the margin, Super Intelligence still owns the revenue headline, Terafab and Tesla-adjacent silicon still sketch a fab path, Cursor and Grok still sell software on top of Colossus watts, and launch still moves the company's own freight—not because rockets beat GPUs in every scenario, but because the industrial system prices a different binding constraint than a single semiconductor guidance line.
Tuesday's Nvidia call is a test of who still speaks for the CUDA stack in Washington. SpaceX investors are running a parallel exam: whether a connectivity cash engine, a 1.7-gigawatt Super Intelligence build, a cursor-shaped software layer, and a Terafab-shaped silicon bet can keep holding the tape when the AI sleeve re-rates on governance—not on physics.
Hawthorne, Nov. 17 — Nvidia reports fiscal third-quarter results tonight in the first earnings cycle without Jensen Huang on the microphone. NVDA enters the call on a six-week losing streak, options implying a one-session move near 7%, and Washington treating succession as an export-control question. The AI complex is priced for governance beta, not a disappearance of demand.
SpaceX, trading as SPCX, is living in a different part of the same trade. Public snapshots put the company’s market capitalization near $1.87 trillion on Monday—down about 1.3% from Friday and only modestly below the ~$1.94 trillion print from Nov. 10. That is not immunity; it is decoupling within the sleeve: the rocket-and-compute conglomerate is holding more of its value this week than the default spokesman for merchant AI silicon.
The social version of the thesis—you listed it in one breath: chips, fab, industry ties, large language models, cars, robotics—maps onto something real inside the filing stack, but not as a guarantee. It maps onto a PLATFORM bet that SpaceX can keep financing and deploying physical AI infrastructure even when CAPITAL re-rates the GPU vendor on key-person and routing risk. Whether that holds through Tuesday night is an evidence question, not a slogan.
What “win no matter what” would require
A clean “heads NVDA wins, tails SPCX wins” story needs four conditions to stay true simultaneously:
- Recurring cash that does not depend on CUDA sentiment (Starlink subscriptions and enterprise/government lines).
- Compute rent recognized on contracts already ramping, not only on the next Blackwell allocation (Colossus hosting and cloud services agreements).
- Software distribution that monetizes model demand without owning every training flop (Cursor enterprise seats plus Grok adoption inside the SI segment).
- Optional silicon that reduces—but does not yet replace—merchant GPU purchases (Terafab planning with Tesla-scale demand as the anchor tenant narrative).
SpaceX’s Q3 print showed the first three operating at scale and the fourth still in the prospectus layer. Revenue hit $14.6 billion; Super Intelligence passed Connectivity on the top line at $7.56 billion versus $5.25 billion, while Connectivity still delivered $2.37 billion of operating income against $431 million in SI. Nameplate compute reached 1.7 gigawatts. Cursor consolidated after the August close. That is the industrial stack with a P&L, not a slide.
What it is not is independence from Nvidia’s ecosystem. Hyperscaler and neocloud capex still flow through Blackwell-class silicon for most of what SpaceX rents and trains. A brutal NVDA guide haircut would not cancel Starlink invoices; it could still tighten the financing air pocket around the neocloud tier that buys Colossus capacity and conditions how aggressively SPCX marks its own SI backlog.
The stack, item by item
Chips and fab. SpaceX and Tesla have publicly advanced Terafab, a Texas complex aimed at logic, memory, and packaging for internal compute needs, with Intel involvement reported and TSMC talks still informal in public coverage. That is a multi-year PLATFORM hedge against supplier concentration—not a current-quarter substitute for GPUs rolling into Colossus aisles. Culled’s industrial valuation frame treated Terafab as the closure of the loop from rockets to custom silicon; execution risk remains the binding constraint.
Industry connections. The same Q3 filing that booked $19.6 billion of quarterly capex also disclosed hosting economics: SpaceX monetizes watts it already poured into concrete. CFO commentary on the first public quarter described incremental EBITDA on activated cloud deals—a landlord story that connects SpaceX to frontier labs without requiring those labs to cheer NVDA’s succession drama.
LLM models. The xAI merger folded Grok into the listed Super Intelligence segment. Revenue mixes now include advertising and model-driven services, not only raw infrastructure rent. That is strategic differentiation from pure neocloud hosts whose fates move tick-for-tick with Nvidia guidance.
Cars and robotics. Tesla is not on SpaceX’s income statement, but markets still bundle the Musk industrial complex. Terafab demand narratives explicitly pair automotive and orbital compute loads; humanoid and factory-automation capex at Tesla informs how investors underwrite long-dated PLATFORM spend next to SPCX. Decoupling from NVDA is partially decoupling from “single-ticker AI beta,” not from Elon-linked governance headlines.
Launch and connectivity. None of the above works without the freight layer. Starship payload proof and Starlink’s 13.6 million subscribers (per Q3) keep the system physically coupled: the company lifts its own satellites, bills its own bandwidth, and powers its own data halls.

This week’s tape versus Nvidia’s
The price action is the user’s observation made precise. Between Nov. 10 and Nov. 17, SPCX’s public market-cap series shows high-single-digit percent swings at worst within the week, finishing near $1.87 trillion after a shallow Monday dip—not a flight-to-safety bid, but also not the same drawdown pattern as NVDA’s pre-earnings streak.
Meanwhile Nvidia’s September 30 13F snapshot—parsed ahead of tonight’s call—showed a ~6% add to the company’s SpaceX stake even as it trimmed Synopsys, Intel, and Nokia. That is CAPITAL speaking in portfolio weights: rebalancing toward private (now public) orbital compute exposure while telecom turnarounds bleed marks. It is not proof that NVDA earnings cannot hurt SPCX; it is proof that one of the largest holders of AI equity still wants the vertical stack on the balance sheet into the Huang vacuum.
Where the thesis breaks
SpaceX does not win in every Nvidia scenario. It wins in scenarios where AI demand persists but who captures the narrative and compliance premium rotates. It loses leverage in scenarios that combine:
- Financing contraction — wider spreads on AI-duration debt that freeze Colossus expansion timelines even if GPUs exist.
- Governance contagion — Bessent-era pressure on Nvidia becoming political pressure on any Musk-linked compute landlord. See succession scenario weights for how STATE can compress multiples without killing shipments.
- SI margin reality — Cursor purchase accounting and $16.4 billion of quarterly SI capex mean profits still lean on Connectivity; a Starlink ARPU fade matters even if Grok usage spikes.
- Supplier choke — export or remote-access rules that hit training clusters industry-wide, not only NVDA’s press release language.
Orbital compute remains an economics problem first and a physics problem second. Terafab remains a permitting and yield problem before it is a moat.
Verdict for holders tonight
Does SpaceX stand to win no matter what Nvidia says? No. CUDA still trains Grok; debt markets still gate SI expansion; SPCX still trades a conglomerate multiple with lock-up and index flows layered on top.
Does SpaceX stand to hold relative value through a NVDA governance scare that keeps hyperscaler demand intact? The week’s tape and the 13F add say that is the bet allocators are already making—because Starlink prints margin, SI prints growth, software prints distribution, Terafab prints optionality, and launch prints freight independence. That is a wider win condition than “beat on EPS,” and narrower than “immune to AI selloffs.”
Listen tonight for Nvidia’s operating truth. Listen just as closely for whether CAPITAL treats the AI build as a single-vendor story or an industrial-system story. SpaceX priced the second narrative into $1.87 trillion before the call; the first narrative still has the microphone.
Continue reading
Sources
SPCX market-cap snapshots for Nov. 10–17, 2026; Nvidia Nov. 16 13F coverage; SpaceX Q3 2026 SEC filing; Culled reporting on industrial valuation, Huang succession, and neocloud governance beta.