34.5% of SpaceX's Revenue Could Disappear by Mid-2027

SpaceX reported third-quarter revenue of $14.6 billion, up 161% year over year and above the $12.7 billion analyst consensus; its AI segment generated $7.6 billion, versus $813 million a year earlier. The article highlights concentration and renewal risks: a customer identified by Bloomberg as Anthropic accounted for 34.5% of sales, under contracts that either party can cancel with 90 days’ notice beginning in early 2027. It also cites potential easing of compute shortages, customer shifts to alternative chips and high valuation as risks to sustaining AI revenue growth.

Analysis

The key risk is not simply customer concentration; it is duration and asset recovery. If a major tenant can exit on short notice while SpaceX has deployed costly, Nvidia-based capacity, revenue can fall faster than the infrastructure cost base, pressuring utilization and returns on capital. The reported customer identification is sourced indirectly, so verify the contract disclosures before treating that exposure as confirmed.

For the next 1–3 months, watch for evidence of contract duration, utilization, and incremental customer additions—not backlog headline value alone. In the 6–18 month view, new Alphabet TPU supply and customers’ own capacity buildouts could weaken SpaceX’s pricing power and make today’s third-party leases transitional. That is also a two-sided risk for Alphabet: it could replace leased compute with internal capacity, but its current role as a buyer leaves some near-term exposure to a compute-demand slowdown. NVIDIA faces substitution risk at the margin, though a shift in chip mix would not by itself establish a decline in aggregate AI compute demand.

Contrarian point: cancellation rights create downside asymmetry, but they also give SpaceX flexibility to redeploy capacity if other buyers are available. A short thesis therefore needs evidence of weaker utilization or pricing, not just cancellability. No valuation, share-price, or options data are provided to establish that the risk is mispriced today.

Key Decisions for Investors

• Keep SPCX underweight versus diversified cloud/platform exposure; do not add on revenue growth alone. Reassess if filings or company commentary show the largest customer’s share declining alongside stable utilization and replacement demand.
• Potential relative-value expression: short SPCX against a beta-adjusted basket of GOOG and AMZN, preferably with defined-risk options if premiums are reasonable. Entry trigger: confirmation of customer exit rights or falling utilization/pricing. The hedge is imperfect—both platforms have their own AI-capacity and demand exposures.
• Track three falsifiers before increasing the short: renewal or extension of major leases on firm terms, sustained high utilization after new capacity comes online, or evidence that alternative customers are absorbing capacity without price concessions. Conversely, a termination notice or weaker lease pricing would strengthen the downside case.
• Treat NVDA as a watch item rather than a direct short: seek evidence of actual TPU/custom-chip substitution in customer workloads before positioning, since aggregate compute demand could offset share loss.