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 deployed, including $1.25 billion monthly from Anthropic and $920 million from Google. However, the AI unit spent $15.8 billion on infrastructure in Q2 and remains unprofitable, while 90-day contract termination clauses make the reported revenue backlog less durable. The article argues that sustained profitability across AI, Starlink, and launch operations will be required to support SpaceX's roughly $2 trillion valuation amid risks of AI-compute commoditization and capacity overbuild.

Analysis

The relevant valuation question is not contracted revenue but utilization-adjusted contribution margin after power, depreciation, financing, and customer concentration. Short cancellation windows convert nominal backlog into a rolling option held by customers; this warrants a lower revenue multiple than hyperscalers receive unless renewal rates, prepayment terms, and take-or-pay protections are independently disclosed. A large infrastructure build also creates a negative working-capital and depreciation burden that can leave reported revenue growing while free cash flow remains deeply negative for several quarters.

Near term, incremental compute commitments reinforce demand for Nvidia's accelerator ecosystem and data-center power equipment, but they are more ambiguous for pure-play GPU clouds such as CRWV. New capacity from a well-capitalized entrant reduces scarcity rents and raises the probability that GPU-hour pricing declines before smaller operators have amortized their fleets. GOOG is better insulated: external capacity can bridge peak demand, while its proprietary infrastructure and distribution preserve optionality to internalize workloads later.

The contrarian view is that the equity debate is prematurely focused on whether AI demand exists rather than who bears utilization risk. Customers retaining the right to resize capacity can arbitrage an eventual surplus across providers, causing margins to reset much faster than revenues; that would favor vertically integrated hyperscalers over leveraged neoclouds. The thesis is falsified if disclosed utilization stays above 85%, contract renewals extend beyond annual terms, and GPU pricing remains stable through the next two earnings cycles.

Key Decisions for Investors

• Maintain a 3-6 month relative-value bias: long GOOG / short CRWV. The trade expresses hyperscaler bargaining power versus merchant-cloud utilization and refinancing risk; target a 10-15% relative move, with a stop if CRWV reports sustained positive free cash flow and contracted utilization above 85%.
• Do not initiate a directional SPCX position solely on reported contract values. Require independently verifiable disclosures on cancellation economics, upfront deposits, power commitments, delivered capacity, and AI-segment EBITDA; treat confirmation of multi-year take-or-pay terms as a catalyst to reassess.
• Retain or add NVDA exposure on 1-3 month weakness rather than chase the headline. New deployments support accelerator shipments, but cap upside if customer capex guidance or lead-time commentary indicates capacity is being built ahead of funded demand; use a 8-10% downside risk limit from entry.
• Set an alert around CRWV debt spreads and GPU-cloud pricing disclosures over the next two quarters. A widening in funding costs alongside flat or declining realized revenue per GPU-hour would validate the short leg and could justify increasing the GOOG/CRWV spread.