Musk wants one million AI satellites in orbit — compute is leaving the grid

overhypedComputeFinance

Thursday, June 11, 2026

Conviction

Medium

Time horizon

This year

Risk

Capex modeled on undelivered fabs

The News

SpaceX CEO Elon Musk outlined more details for his company's planned data centers in space ahead of a widely anticipated IPO on Thursday (June 11) expected to make him a trillionaire . Musk laid out the first detailed design of SpaceX's AI1 satellite in a 30-minute video, the opening generation of an orbital craft SpaceX wants to build by the million; the spacecraft carries a 150 kW peak compute payload across a 70-meter deployed wingspan with an interchangeable hardware design that lets different chipmakers supply the processors . SpaceX has filed plans with the FCC for up to one million 'orbital data center' satellites, saying it would operate a constellation with unprecedented computing capacity to power advanced AI models . Musk confirmed the new 11-million-square-foot Bastrop facility is expected to begin producing complete "AI satellites" by 2027, targeting 1 gigawatt of orbital AI compute capacity by the end of next year, with plans to scale by an order of magnitude annually thereafter . (Space.com)

The Read

The verdict is OVERHYPED, but not in the lazy sense — the engineering disclosures are real, and orbital compute is a defensible answer to a power-and-water-constrained terrestrial buildout. What earns the tag is the IPO calendar. Musk dropped the AI1 design three days before pricing, attached it to a Terafab plan that depends on a chip foundry that doesn't exist yet, and committed to scaling by an order of magnitude annually for three years — a sequence whose only test before subscribers commit is whether the prospectus walks any of it back. The Helix story (03) is the relevant comparison: $10B of disciplined PE capital just bought ~1-2 GW of terrestrial capacity. SpaceX is asking public markets to underwrite 100 GW orbital by 2030 against an unbuilt fab. The narrow, governed-infrastructure thesis says one of these gets built on time. The Musk reflex says price the other one accordingly.

Counterview

SpaceX is the only operator with experience flying 10,000+ satellite constellations, and terrestrial AI power is genuinely capped. If even 10% of the constellation ships, it reshapes hyperscaler cost curves enough to matter.

Watch For

SpaceX's S-1 disclosures on Friday for hard milestones — first AI1 launch date, Terafab first-silicon date, and any contracted customer for orbital compute. If two of three are missing or hedged, the satellite story is IPO narrative, not capex plan. If two of three are dated and contracted, it's the first credible non-hyperscaler compute source in the market.

For Founder & Operator