Current Thesis
SpaceX has become a compute landlord funded by Starlink, and Grok is a laggard. Anthropic, Google and Reflection lease Colossus at about $2.3B a month on 90-day terms; Cursor is the only proven AI software revenue. At $2T the stock pays for a $100B AI run-rate and orbital compute that do not exist. The AI exposure is cyclical: it rises with frontier-lab capex and falls first when that capex turns, and the multiple falls with it.
Connectivity earns the only operating profit and funds $15.8B of quarterly AI capex. Contracted compute is real, with claimed sub-year payback, but every tenant is also building elsewhere and every lease is a capex line a lab can cut in a downturn. Comp multiples on Starlink, launch and contracted compute cover roughly half the market value.
We'd revise if
Revise up if AI run-rate approaches $100B by mid-2027 with tenants extended past 2029, or Grok wins revenue independent of Colossus. Revise down if hyperscaler capex guidance turns, a frontier-lab IPO prices below its last round, or a tenant exits early.