Michael Burry Says SpaceX Isn't Worth $1 Trillion. Its AI Business Could Prove Him Wrong.

SpaceXAI has signed compute agreements that could generate $3.4 billion per month, or $40.8 billion annually, once fully ramped, including $1.25 billion monthly from Anthropic and $920 million from Google. However, the unit spent $15.8 billion on AI infrastructure in Q2, remains unprofitable, and its contracts allow termination on 90 days' notice. The deals strengthen the near-term case for SpaceX's AI business, but high capex, potential AI-compute commoditization, and overcapacity risk leave its ability to support a roughly $2 trillion valuation uncertain.

Analysis

The key valuation variable is not contracted run-rate but committed utilization after cancellation rights, power pass-throughs, and depreciation. A provider carrying front-loaded infrastructure spend can show rapid revenue growth while producing weak incremental free cash flow if customers retain the option to resize every quarter; this deserves a lower multiple than hyperscale cloud revenue and more resembles merchant capacity. SPCX needs disclosed utilization, contract take-or-pay terms, power costs, and debt/lease financing before its AI segment can support a durable sum-of-the-parts re-rating.

Competitive implications are more negative for CRWV than for GOOG. New large-scale third-party capacity expands buyer leverage for AI labs and hyperscalers, pressuring neocloud pricing and reducing the scarcity premium embedded in CRWV's equity; conversely, GOOG can use external capacity as a variable-cost buffer rather than overbuild internally. The second-order risk reaches NVDA and data-center supply chains in 6-18 months: an industry-wide buildout initially supports shipments, but a utilization reset would rapidly defer accelerator orders because compute infrastructure has a long deployment lead time and very low marginal operating cost once installed.

Consensus may over-credit headline revenue while underweighting the asymmetry of a 90-day churn event. In the next 1-3 months, any evidence of sustained customer drawdowns or financing terms could support SPCX sentiment, but the more important 6-18 month test is whether realized AI gross margin remains intact as capacity additions arrive. A broad AI-demand slowdown is not required for the thesis to fail; even modest price concessions or subscale utilization would materially impair returns on a large fixed asset base.

Key Decisions for Investors

• Maintain a relative-value bias: long GOOG / short CRWV over a 3-6 month horizon. GOOG gains capacity optionality and avoids some internal capex timing risk, while CRWV is more exposed to AI-compute price compression; cover if CRWV demonstrates sustained utilization and gross-margin expansion despite new supply.
• Do not initiate a directional SPCX long on announced contract value alone. Upgrade only after quarterly evidence of positive AI operating cash flow, utilization above management's breakeven threshold, and cancellation-adjusted backlog; a failure to disclose these metrics is itself a valuation-risk signal.
• Use CRWV put spreads dated 6-12 months as a defined-risk expression of an oversupply/utilization reset, rather than outright short exposure. The thesis is invalidated by accelerating contracted capacity, stable pricing, and improving free-cash-flow conversion through at least two reporting periods.
• Monitor GOOG capex guidance and external-compute expense at the next earnings release. A material increase in outsourced capacity without offsetting monetization would turn the current relative-long thesis into a watch item, since variable compute can still become a margin headwind if AI product revenue lags.