06-reference

stratechery moffett spacex bubbles

2026-10-22·reference·source: Stratechery·by Ben Thompson
stratecheryspacexstarlinkcapital-cycleai-bubbletelecom-historymoffettnathansoninvesting

An Interview with Craig Moffett About SpaceX and Bubbles (Stratechery, 2026-10-22)

Why this is in the vault

Craig Moffett (MoffettNathanson) gives Thompson a telecom-analyst's stress test for the AI buildout: a specific, falsifiable rule for distinguishing real demand from bubble froth, lifted directly from his own 1999-2000 fiber-bust coverage, plus a two-sided Starlink thesis (broadband vs. direct-to-device mobile) that reframes "fiber vs. satellite" as a live capital-allocation fight rather than a settled story.

Mapping against Ray Data Co

Direct input to the Markov capital-cycle build ([[project_investing_markov_capital_cycle]]): Moffett's dot-com lesson is not "bubble or not" punditry, it's a concrete solvency test — ignore all revenue exchanged between layers of the value chain (what Nvidia bills CoreWeave, what a neocloud bills a foundation-model lab) because that's intercompany noise; the only number that matters is whether end-demand revenue flowing into the terminal layer (enterprise and government dollars into Anthropic, OpenAI, etc.) will ultimately exceed the cumulative cost of building the entire stack beneath it. That is a cleaner operationalization of the Phase 2→Phase 3 transition the founder is tracking in the memory-cycle thesis ([[2026-05-18-memory-cycle-v1.1]]) than "capacity online → glut → margins roll over": it says the tracker should weight terminal-demand realization (price deflation risk from cheaper/open-weight models eating the high end, "thin client" local inference compressing the addressable spend) over midstream revenue figures like chip ASPs or neocloud bookings, which can look healthy right up until the terminal layer can't actually pay for what's been built. Moffett explicitly rejects the Gavin Baker counter-argument that "GPUs are being used and prices are rising" (unlike dark fiber in 2000) as sufficient reassurance — utilization isn't the test, terminal solvency is. Moffett's terminal-demand-vs-midstream-revenue framing is the clearest articulation yet of the pattern the vault already tracks as [[capex-revenue-deceleration-signal]] — he's independently supplying the solvency test that signal's capacity-vs-revenue gap needs in order to be read as more than an abstract risk. Separately, his telecom history answer to "did fiber ever pay back the 90s buildout" — "probably yes by now, but barely" and only after a wave of bankruptcies — is a sobering base rate for how long and how destructively a real-demand buildout can still take to clear, independent of whether the underlying technology turned out to matter.

⚠️ Sponsorship

"Presented by Stripe" appears in this issue's view-in-browser link and cover art, confirming the standing sponsorship announced one day earlier in [[2026-10-21-stratechery-presented-by-stripe-sponsorship]] is now live across issue types, including the Interview format — not a one-off placement. No Stripe-specific content in this issue; the interview itself carries no third-party promotional content beyond the masthead placement. Per Thompson's disclosed terms, this does not affect editorial independence on non-fintech topics like SpaceX/telecom.

The core argument

Moffett frames SpaceX (his third Stratechery interview after 2023's Charter/Disney and 2024's Apple/telecoms conversations) as sitting on two compounding "monopolies" he uses the word deliberately and provocatively: a dominant position in launch, and — built on top of it — a leading position in low-Earth-orbit satellites, both carrying real regulatory and geopolitical risk given Musk's political identification with one party. He draws a hard line between Starlink's two genuinely separate businesses: the Ku/Ka-band broadband service (dish-based, high-capacity, no wall penetration) and the direct-to-device mobile play (buying spectrum, including the old Nextel band, to use Starlink satellites as de facto mobile towers with "infinite backhaul"). On fiber, he argues the terrestrial buildout is not done but is being squeezed by its own success story: cost-per-mile is rising because data-center construction is competing with fiber-to-the-home for the same labor and materials, crowding the category just as Starlink pressures the edge from the other direction. The bubble discussion closes the piece: Moffett's dot-com-era rule is to disregard revenue that merely circulates between value-chain layers and ask only whether real end-demand will eventually cover the full cost of the stack — a question he says the market is currently "panicking" over precisely because it's genuinely unresolved, not because the technology is fake.

Related