Key points
- Casey Handmer, Terraform Industries CEO, argues Starship will cut launch costs enough that AI compute will converge on cheap solar-powered hardware in space.
- Handmer says permitting delays make California infrastructure so slow and costly that the people needed to build it lose interest, citing the California Aqueduct's lost capacity.
- Handmer says Starlink works because it sells information rather than power, and he doubts Starship will reach 10,000 flights a year within five years given Starbase's roughly 250-ship ceiling.
Cheap launch and cheap solar push AI compute toward space
Handmer argues that Musk's edge comes from building physical things fast in a system that makes building slow. Cheap launch and cheap solar are pushing AI compute toward space, while permitting, zoning and taxes hold back similar progress on the ground. The conversation keeps returning to how much regulation costs the people who would build.
Takeaways
- Cheap Starship launches could move AI compute to space. Handmer says Starship will cut launch costs while power, permitting and local politics keep tightening limits on surface data centres. Musk reportedly said space compute would be cheaper to launch than build on the ground within three years.
- One solar array can power fuel and compute. Terraform's 1.8 megawatt array runs a synthetic fuel plant and a data centre rack together. Each solar module makes about a kilowatt and each GPU uses about a kilowatt, which more than doubles the output of the fixed solar infrastructure.
- Permitting makes big infrastructure slow and costly. Handmer says permitting makes projects so slow and costly that the people needed to run them lose interest. The California Aqueduct has lost capacity to subsidence, and each year of delay equals about 1,000 statistical lives. Zoning blocks Terraform in Burbank.
- Regulation and taxes drag on American builders. Handmer puts US federal regulation at 12m pages and proposes sunsetting agencies after ten years, with DARPA as the model. He estimates government takes about 70 percent of the net surplus passing from capital to workers.
- An AI token may be worth far more than a bit. Handmer says Starlink works because it sells information rather than power, and an AI token may be worth a thousand times a Starlink bit. One experiment had an AI analyse 200 topics for about $260 of tokens.
- Musk wins on technical intuition and decisive firing. Handmer credits Musk's intuition on technical matters. Starmind progressed once SpaceX's three best engineers were put on it, after Musk fired a team that was making progress but not shipping.
- Ten thousand Starship flights a year is distant. Handmer doubts 10,000 flights a year arrive within five years, since the Starbase factory is set to max out near 250 ships a year. Even at Musk's speed, approvals for the Louisiana launch facility may take long.
Doubts and limits
- Handmer doubts a lunar Mass Driver would beat building more Starships, launch pads and factories on Earth, which he says is thousands of times cheaper.
- Handmer argues the optimum level of income inequality is not zero and that the US middle class has been lifted, not lost. He says the worry is whether the bottom decile can climb the economic ladder.