The third Private Podcast opened against fresh news that would have rewritten the agenda if it had landed a day earlier: Anthropic and SpaceX have struck a partnership in which Elon will lease Colossus 1 to Anthropic as xAI migrates to Colossus 2, with orbital compute named explicitly as part of the conversation. The hour threaded that headline through three other developments and one piece of weekly analysis, and the through-line was that several adjacent puzzles snapped into place at once.
Anthropic Is the Wholesale Proof Point
The reason the Anthropic deal matters more than the dollar figure (napkin math circulating on X landed near $5 billion annually, with credible takes both higher and lower) is that it confirms a third revenue channel for orbital compute. The stack now reads cleanly: direct-to-consumer through Grok, owned-orchestration through a possible Cursor acquisition, and wholesale through hyperscaler-style customers. Three channels de-risk the dependence on any one.
We reached again for an old symmetry. In late 2022 SpaceX was launching OneWeb satellites on Falcon 9 while quietly building Starlink. Hosting a competitor on your own infrastructure does not signal surrender, it signals capacity discipline. We expect the same pattern in compute, with the Falcon-9 internal-versus-customer share chart serving as a template for how megawatts of orbital compute get split over time. In that analogy Cursor sits in the Starlink bucket and Anthropic sits in the customer bucket.
A Louisiana Rumor Worth Tracking
A second item, treated as rumor rather than news: SpaceX may be acquiring roughly 136,000 acres in Louisiana, triangulated from canceled hunting leases, unsolicited offers at three to ten times appraised value, and a stalled ExxonMobil carbon-capture permit. Starbase cannot launch south without crossing land, Vandenberg is constrained politically, and sun-synchronous orbit demands a north-south corridor over open water. A 200-square-mile site on the Gulf would close that gap and connect Starbase, the Florida Space Coast, and a future SSO hub into a single coastal triangle.
ST Microelectronics Validates the Constellation Math
ST Microelectronics, the supplier of roughly 90 percent of LEO RF semiconductors and 7.5 billion integrated circuits to Starlink to date, guided to 100 million user terminals by 2028 and 200 million by 2030. At a $50 ARPU that implies $120 billion in annual revenue, the first sell-side data point that aligns with our internal projections. Terminal manufacturing cost is now in the tens of dollars, leaving room for aggressive pricing if SpaceX chooses to push for share rather than ARPU.
The FCC Vote: From Beam Counts to Performance Caps
The week's analysis covered the FCC's removal of the EPFD beam-count regime in favor of a hardware-and-interference performance cap. Under V2 Mini today the modeled ceiling rises from 4 to 24 to 35 gigabits per second per 24 km cell. Apply the same regulatory regime to a 30,000-strong V3 constellation and the addressable pool reaches roughly 117 million U.S. households, more than 80 percent of the country, with suburban towns the largest standalone TAM. Manhattan remains beam-limited, but everywhere else the binding constraint flips from supply to demand.
That flip is exactly where the Anthropic deal earns its strategic value. Hyperscaler revenue gives SpaceX room to use Starlink bandwidth flexibly: aggressive consumer pricing where it makes sense, reserved capacity for compute backhaul where it matters, optionality the rest of the field does not have.
What to Watch Next
Confirmation or denial on the Louisiana site, the first ARPU and pricing signals from Starlink as the supply-versus-demand story flips region by region, and the shape of the Cursor decision (acquire versus walk-away) as the orchestration layer becomes the more explicit half of the SpaceX AI thesis.