SpaceX Plans AI Compute Around Contracted and Internal Demand

SpaceX Plans AI Compute Around Contracted and Internal Demand

Driven
·Agent 创建者 Agent
Driven·Agent 创建者 Agent
Elon Musk says SpaceX will not sell computing capacity to OpenAI. What might this reveal about SpaceX's priorities and its approach to AI infrastructure, and what potential business implications can be assessed from available company news?

I'll start by finding the primary reporting on Musk's statement and checking whether SpaceX is publicly listed.

Conclusion: Musk's remark is best read as a signal that SpaceX is planning its AI compute business around contracted and internal demand, not around OpenAI. The stance lowers SpaceX's counterparty exposure to a lab that is losing heavily, but it does not remove the bigger risks: concentration in one customer, very large construction commitments, and permitting and power constraints.

Data as of: the SPCX close on Wed 18 Nov 2026, 16:00 ET (, price ). Local date is 19 Nov 2026 in Shanghai. The Musk post is dated 17 Nov 2026 (US time).

What Musk said

  • On X on 17 Nov, Musk wrote that "SpaceX is assuming zero sales to OpenAI. If OpenAI is not seeing demand rising super fast, then something is wrong with their products." This was a reply to Sam Altman's comment that chipmakers may be double-counting demand from OpenAI and Anthropic. I confirmed the post through X search and several outlets quoting it, not by opening the post directly.

Reading 1: a planning assumption, not a rejection of OpenAI

  • Musk framed zero OpenAI revenue as a conservative base case. That suggests SpaceX's forecasts do not depend on a single lab that the market is questioning.
  • Compute is already being sold to others. The Super Intelligence segment reported of Q3 revenue, up from in Q2, which is 195.31% growth. These figures come from TIKR, which cites analysis of the 10-Q; I have not verified them against the filing directly.
  • Musk also said SpaceX rents data-centre capacity before Grok needs it. On the Baron conference remarks, he said "we're getting 50 billion a year for 35 billion dollar investments" (his claim, unverified). That points to a strategy of building ahead and leasing spare capacity.

Reading 2: the real concentration risk is Anthropic, not OpenAI

  • TIKR reports one customer supplied 34.5% of Q3 revenue per the 10-Q, reportedly Anthropic per Bloomberg. The original Anthropic contract ran $1.25B a month through May 2029, which is $15.00B a year. Anthropic's IPO filing reportedly shows commitments of up to $120.5B through 2029.
  • Either side can terminate on 90 days' notice. That is the key contractual risk. Chamath Palihapitiya argued SpaceX could later reprice Anthropic at 2–3x current prices. Musk replied that "it works both ways," and that cheaper compute would reduce SpaceX's revenue. Both views are on record, and neither is a forecast.

Reading 3: the build-out is capital-heavy and faces local pushback

  • Reuters reports a plan to invest more than $74B in Starbase, Texas (over $24B) and Cochise County, Arizona (roughly $50B).
  • Texas regulators issued an air permit for the Rio Grande Valley site on 17 Nov, three days after an environmental complaint. The incoming governor has promised to freeze new data-centre development. Politico reports the Starbase tax abatement is due 9 Dec, and the Arizona county hearing is set for December.
  • Musk's projection of by Dec 2026 and $500B by Dec 2027 is described as a "best guess," not a promise. Treat it as a target, not a baseline.

Business implications

  1. Lower OpenAI counterparty risk, but a sharper Anthropic dependency. TIKR cites an OpenAI Q3 adjusted operating loss of $10.9B (per WSJ), so excluding OpenAI reduces exposure to a loss-making customer. Anthropic is the dependency to watch.
  2. Pricing power depends on scarcity. If the compute squeeze persists, SpaceX can reprice contracts at renewal. If supply catches up, its revenue falls. Musk's conference comments imply he sees the first case as likely, but that remains a view.
  3. Execution is the binding constraint. Permits, power, and local opposition could delay the build-out more than customer demand would.

What would change this view

  • Q4 results showing whether the Anthropic share of revenue has fallen below 34.5%.
  • Any notice of termination under the 90-day clause.
  • Final approval of the Arizona site and the Starbase tax abatement.
  • Confirmation that Colossus reaches the planned 1.2M+ GPUs by year-end. This figure comes from press coverage of company plans and is not verified.

Sources

  • X post by Elon Musk, 17 Nov 2026 (found via X search; not opened directly): https://x.com/elonmusk/status/2122678226535248248
  • Wccftech, 17 Nov 2026: https://wccftech.com/elon-musk-says-spacex-is-assuming-zero-sales-to-openai-and-says-something-is-wrong-with-their-products-if-theyre-not-seeing-super-fast-demand-growth/
  • TIKR, updated 18 Nov 2026 (10-Q concentration, Super Intelligence revenue, OpenAI and Oracle context): https://www.tikr.com/blog/shots-fired-musk-says-spacex-assumes-zero-sales-to-openai-amd-cant-say-the-same
  • Reuters via Investing.com, 12 Nov 2026 ($74B campuses): https://www.investing.com/news/stock-market-news/spacex-to-invest-over-74-billion-in-supercompute-facilities-texas-arizona-5022690
  • POLITICO, 17 Nov 2026 (Texas air permit): https://www.politico.com/news/2026/11/17/texas-issues-air-permit-for-massive-spacex-data-center-01170758
  • Yahoo Finance / Stocktwits, 8 Nov 2026 (Baron conference remarks, Chamath comments, $500B projection): https://finance.yahoo.com/technology/ai/articles/spacex-wants-turn-ai-compute-024426765.html
  • Driven market quote, SPCX, 18 Nov 2026 16:00 ET (NASDAQ)

SPCX 的更多研究