Casey Handmer on Why Elon Musk Outcompetes Everyone: Space Data Centers, Starship Economics, Permitting Paralysis and the Hidden Tax on Building

Casey Handmer, founder and CEO of Terraform Industries, sits down for the third installment of the “Why Elon Outcompetes Everyone” series. Recorded at Terraform’s shop in Burbank, the conversation runs from lunar mass drivers and orbital AI data centers to Starship launch math, why large projects have become effectively illegal in America, how Elon Musk surfaces talent, and why Handmer thinks AI will supersede the tax and spend state rather than reform it.

TLDW

Handmer argues that a lunar mass driver is physically elegant (a short electromagnetic track firing a few kilograms of moon rock per second) but economically hard to justify while Starship pushes launch costs toward a few hundred dollars per kilogram. The real prize is orbital compute: Starlink proved that modulated microwaves carrying information earn roughly a billion times more per watt than space based solar power, and AI tokens are worth perhaps a thousand times more per bit than Netflix streams, so GPUs sprinkled along sun-synchronous solar arrays become a business. On Earth the long run question is whether batteries beat launch. From there the talk turns to permitting: the half-capacity California Aqueduct, Altadena’s slow rebuild, Henry Kaiser building a Tahoe estate in 22 days to beat an injunction, and Terraform being pushed out of heavy-industrial Burbank. Handmer explains how secondary manufacturing creates most value (and how Australia lost it), why talent never saturates at the top, why Elon uses up or out escalation, why synthetic fuel is his “narrow way,” and why Elon’s edge is intuition rather than secret spreadsheets. The back half covers DOGE, old age dependency, Japan’s habit of rebuilding 98% of a city every 50 years, compressing 12 million pages of federal law into 4,000, DARPA style sunsets for agencies, the roughly 70% of value that taxes siphon between an investor and a worker’s groceries, Starbase versus Louisiana as regulatory arbitrage, the math on 10,000 Starship flights a year, Baumol’s cost disease, and why the optimum amount of inequality is not zero.

Thoughts

The sharpest piece of economic reasoning in the conversation is Handmer’s revenue per watt comparison. Space based solar power fails because it takes something expensive (power in orbit) and sells it where it is cheapest (the grid, at about ten cents a kilowatt hour). Starlink works because it modulates the same microwaves with information, which he estimates is worth a billion times more per watt. Orbital AI compute is the next step up that ladder: tokens from a frontier model are worth far more per bit than error-corrected video, so the physics that sank space solar becomes a margin engine. This framing is more useful than the usual “free power in space” pitch, and it also explains why he is skeptical of lunar industry. When a GPU is the expensive part and launch is a rounding error, you do not need moon rock. You need more Starships.

His permitting argument gets its force from putting a number on delay. Using the value of a statistical life, he estimates the economic loss from California’s inability to repair a subsided section of the California Aqueduct at roughly 10,000 lives, about a thousand a year while the project sits in review. The fix is “bulldozer, concrete, dump” in a corridor that has already been disturbed. Set that beside Henry Kaiser finishing a lakeside estate in 22 days before an injunction could be filed, and Handmer’s point becomes clear: the U.S. still has the capacity to build fast, but the process now drives away the people who know how. It is also why he calls fast infrastructure “super progressive”: rich people do not run out of water.

The middle of the interview has the most transferable management lesson. Handmer rejects the idea that talent saturates a few levels above wherever you happen to be standing. The best people separate from the merely good, which is why, in his telling, Starlink stalled with a fully staffed team of respected engineers until Elon replaced them and three people made it work. His model of Elon’s system is simple: hire many talented people, hand them steadily harder problems, and let up or out reveal who can handle the next level, because no exam can. Add his point that a problem escalated to the boss should be “messy,” and you get a coherent theory of why Elon’s companies move faster than their size would predict.

The segment most likely to stick is late in the conversation, when Handmer traces a dollar from the Collisons’ Stripe fortune, through Terraform’s roughly $350,000 to $400,000 monthly payroll, to what his employees spend on rent and food. His estimate is that about 70% of the value is taken in taxes along the way. You can argue with the exact figure, but the exercise is the point: follow capital to the physical goods workers need, and ask why a Burbank neighborhood that once built 100,000 aircraft now hosts vacant studios and one of the last anodizing shops in California. It pairs well with his proposal that non-constitutional agencies sunset automatically after ten years, the way DARPA limits staff to fixed terms, so institutions die before they calcify.

Near the end, Handmer’s analysis of the Louisiana Starbase is one of the least discussed ideas in the interview. He reads it less as a capacity play than as regulatory arbitrage: when the Rio Grande Valley authorities got comfortable with their monopoly on Starship, Elon built an escape hatch that costs something but not everything. “You can shear the sheep but you can’t skin them.” Handmer then runs the launch math himself: about 250 Starships a year from the factory, times ten reuses, gives 2,500 flights, well short of Elon’s 10,000 per year target in five years. His view that the target is a good orienting number rather than a forecast is a useful way to read almost every Elon goal.

Key Takeaways

A lunar mass driver could be less than 1,000 feet long if acceleration is unconstrained, with a shuttle firing payloads of a few kilograms roughly once per second, which adds up to about 100 tons a day.
The mass driver could be built and tested on Earth, then shipped to the Moon for minimal assembly, far easier than a 20 mile precision track.
Handmer has no timeline for it and doubts the economics, because building more Starships and launch pads on Earth, where everything is thousands of times cheaper, will likely win.
Starship should push launch to orbital constellations down to a few hundred dollars per kilogram, small next to the cost of a high performance GPU.
Most of the embodied cost of an orbital compute satellite is the GPU. Radiators, solar arrays and comms are “spiritually two dimensional” and can be folded or made very thin.
Elon correctly foresaw that terrestrial data centers would hit marginal constraints (permits, local demands, power) that raise their cost.
Local jurisdictions are in a “price discovery phase” for what data center builders will pay to get approvals.
Terraform’s 1.8 megawatt solar array runs its synthetic fuel plant by day and can feed a battery-backed AI compute rack continuously, more than doubling the output of fixed solar infrastructure.
Handmer expects AI compute to converge on two forms: a giant Starlink-like satellite in space, or the same thing on the ground with the radio swapped for a battery. The long run question is whether batteries are cheaper than launch.
Land is not the constraint. Nevada alone is roughly 65 million acres, much of it federal land that was once too marginal to give away.
A small-government U.S. funded by tariffs would, today, be building data centers on federal land in Nevada and leasing them to the highest bidder.
Over the last 50 years, large projects have become effectively illegal, mostly through regulation rather than statute, and the best builders will not wait 15 years for permission.
The California Aqueduct runs at about half its design capacity because a couple of hundred miles subsided. Handmer estimates the lost value at roughly 10,000 statistical lives.
Speed makes infrastructure cheaper. World War II aircraft plants cut labor hours per unit by 10x and raised output by about 50x.
Henry Kaiser built his Lake Tahoe estate in 22 days, working 24 hours a day, to finish before an injunction could be filed.
Altadena lost about 10,000 structures, and at the current pace it will still be rebuilding when the next fire comes.
Terafab in Texas may move faster because chip onshoring is a national security imperative, and Elon knows “the numbers to call.”
Terraform cannot expand even in heavy-industrial Burbank. Landlords rebuilt as light industrial, and metal refining is effectively banned in California.
Terraform’s neighbor is, as far as Handmer knows, the only certified aircraft hinge maker in NATO, a remnant of the supplier network that served Lockheed’s Burbank plant.
Secondary manufacturing is the single biggest value multiplier. Australia lost its middle, keeping only primary extraction and tertiary services, and its economy is flat.
Starlink earns roughly a billion times more revenue per watt of transmitted microwave power than space based solar power would.
Sun-synchronous orbits keep satellites almost always in sunlight, which cuts battery needs. Laser links route data through Starlink.
An AI token bit is probably at least 1,000 times more valuable than an average streaming bit. Handmer spent about $260 on tokens for research he says beat a year of expert analyst work.
Spreading GPUs along a solar array (one kilowatt module per one kilowatt GPU) removes the thermal box problem and makes dead GPUs irrelevant. You de-orbit the satellite when enough have failed.
Rivals have struggled to match Starlink because it requires best in class satellites plus preferred access to cheap launch.
The Louisiana launch site touches the coast, the Intracoastal Waterway, the Mississippi system and wildlife refuges, so federal and state approvals could take a long time even at Elon speed.
Citing Anthony Colangelo of Main Engine Cutoff, Handmer notes that SpaceX and Tesla are arguably the preeminent large scale real estate developers in the country.
Elon’s new companies often take six to eight years to “bake” before shipping a strong product. Terraform is four years in.
Talent does not saturate at the top. Starlink shipped only after Elon replaced a fully staffed team, and three top people made it possible.
Surfacing talent requires graded, increasingly difficult problems with up or out. No test at the end of school can do it.
Managers should boil messages down to a single bit, and teams should be able to model what the leader would say without asking.
Pick a startup area where you have a persistent differential advantage and capitalism is not already allocating capital well. Handmer passed on drones for that reason.
Profitable synthetic fuel has never been done. The science is a century old, but making it manufacturable like an appliance is “ridiculously impossibly difficult.”
Elon’s edge is mostly correct intuition, not secret spreadsheet derivations.
Elon would never win a popularity contest, and DOGE showed he is too easy to misunderstand. Handmer thinks society is better off with Elon outgrowing the spending state than pruning it.
Old age dependency will break political systems unless growth radically increases or longevity is radically extended. Germany’s median age is 48.
War no longer clears out regulation the way it did in 1941. Competitive pressure between countries is the more likely forcing function.
Japan normalizes construction and aims to replace about 98% of a city’s fabric every 50 years, so buildings average 25 years old and builders stay in practice.
Federal code and regulations run to about 12 million pages. Handmer had an AI compress American law to 4,000 pages and thinks the law should fit in an overhead bin.
Non-constitutional agencies should sunset automatically after 10 years, as DARPA does with fixed term staff, to prevent capture and sclerosis.
Handmer estimates about 70% of value is lost to taxes between investor capital and what workers need to live.
AI will supersede the income tax model built for 20th century payroll labor, the way the motor car ended the horse manure crisis rather than taxing horses.
The Louisiana Starbase is regulatory arbitrage against Texas: “you can shear the sheep but you can’t skin them.”
At 250 Starships a year with 10 reuses each, the math gives about 2,500 flights, so 10,000 a year within five years is unlikely, but it is a good orienting target.
Baumol’s cost disease may dilute AI’s real economic impact. Universal high income is unlikely because of the hedonic treadmill.
The optimum level of income inequality is not zero. The real test is whether the bottom decile can get on the ladder, and “the United States has the richest poor people on Earth.”

Detailed Summary

The Lunar Mass Driver and Why Earth Launch Probably Wins

Handmer, who worked on Hyperloop’s magnetically levitated systems, explains that low lunar gravity and no atmosphere make it possible to launch mass into space without a rocket. A shuttle in a short electromagnetic track, like the shuttle in a loom, accelerates, releases its payload halfway, decelerates, reloads and fires again, perhaps once a second. That turns 100 tons a day into 86,400 small shots. He points out that the device could be built and tested on Earth and needs only minimal assembly on the Moon. But he is blunt about the business case: the mass would supply an orbital AI constellation, and it is hard to see why that beats building more Starships and factories on Earth. The binding constraint may be social tolerance for thousands of launches a day rather than cost.

Orbital Compute, Batteries Versus Launch, and Federal Land

Discussing Elon’s claim, on a podcast with Dwarkesh Patel and John Collison, that space compute will soon be cheaper than ground compute, Handmer says Starship makes launch both cheaper and far more available, while terrestrial data centers keep running into new costs as localities demand their share. Terraform is testing the ground version: a 1.8 MW solar array that powers synthetic fuel production during the day and a battery-backed compute rack around the clock. He expects AI compute to converge on either a Starlink-like satellite with a GPU or a ground unit where a battery replaces the radio. Land is abundant, and he imagines a lean, tariff-funded federal government leasing Nevada data center land for revenue.

Why Large Projects Became Effectively Illegal

Handmer contrasts the justice system’s tolerance of imperfection with the presumption that any large public project carries unbounded environmental downside. The result is a process so slow and costly that capable builders will not engage. His main example is the California Aqueduct, running at about 50% of its 4.5 million acre-feet design capacity because sections subsided from groundwater pumping. The fix is ordinary road crew work, yet approval takes a decade and billions. He contrasts that with World War II shipbuilding and aircraft production, Henry Kaiser’s 22-day Tahoe estate, and Altadena, where very few of 10,000 lost structures have been rebuilt nearly two years after the fire.

Terafab, Burbank and the Lost Middle of the Economy

Asked how Elon can plan Terafab, possibly the world’s largest building, Handmer points to repeated success and the national security leverage of chip production. Most growth, though, comes from thousands of small businesses, and Terraform’s own plan to expand across its Burbank neighborhood collapsed under zoning and landlord restrictions. Burbank built around 100,000 aircraft in World War II. Lockheed’s plant is now an Amazon fulfillment center, and aging specialist suppliers remain, including what Handmer believes is NATO’s only certified aircraft hinge maker. Secondary manufacturing captures most of the value between raw materials and services, and Handmer warns that Australia, despite every natural advantage, lost it and now has a flat economy.

Star Mind: Tokens From Orbit

Handmer says he dislikes the name and was initially skeptical of data centers in space, since power there is neither infinite nor free. The Starlink model changed his mind. Instead of relaying Netflix at $100 a month, sun-synchronous satellites could relay tokens from a hyperintelligent model that improves with software updates. He describes a weekend project that used about $260 of tokens to scrape and categorize data on 200 topics, which he says beat a year of expert analysis. Rather than a box of racks with a cooling problem, GPUs can be distributed along the solar array, so a failed GPU just means 99 of 100 keep working until the satellite is de-orbited. Scaling fast comes down to maturing Starship and building out production and pads in Texas, Florida and Louisiana.

How Elon Builds Teams and Surfaces Talent

Handmer notes that Elon’s companies often incubate for six to eight years, and SpaceX and Tesla are now elite real estate developers as well as engineering firms. Elon seems less filtered and faster now, though Handmer says Elon also shot from the hip in the PayPal days, with less exposure. Talent keeps differentiating at the top, as sport makes obvious, and up or out on steadily harder problems is how you find the one in a million project manager. Elon’s rare skill is working effectively at every organizational size, and that requires transmitting a clear message so teams can model the leader in their heads.

Choosing a Hard Problem: Synthetic Fuel

Responding to Warren Buffett’s idea of stepping over one-foot bars instead of jumping ten-foot ones, Handmer recommends Poor Charlie's Almanack and explains his rule for founders: find persistent differential advantage in an area capitalism is not already solving. Drones failed that test. Synthetic fuel passed it, but it is “ridiculously” difficult: not as hard as orbital rockets, yet no one has made it profitable, and the field is littered with failed carbon capture and hydrogen ventures. Paraphrasing a line from Dune, he says there is a narrow way and Terraform has found it, even if it is as hard to explain as reusable rockets were to Arianespace’s leadership. He concludes that Elon’s advantage is mostly correct intuition.

DOGE, Old Age Dependency and Competitive Pressure

Handmer says Elon is too unguarded to win elections. DOGE compiled a list of cuts that Congress voted down, much of its modernization work continued without him, and his time is worth millions per minute. The deeper problem is old age dependency: rich countries age until the median voter, 48 in Germany, can no longer vote for a future they will be part of. Modern war destroys wealth rather than clearing regulation, so competitive pressure is the likelier catalyst. One Anglophone democracy defecting to aggressive deregulation would be enough. He points to Japan’s culture of constant rebuilding, and imagines turning the Salton Sea into a lakeside Phoenix, since an environmental catastrophe can only improve.

12 Million Pages of Law, DARPA Sunsets and the 70% Siphon

Handmer had Claude compress American law and regulation to about 4,000 pages, roughly a McMaster-Carr catalog, and notes that the first compilation after the Civil War came to around 1,700 pages. Today it is about 12 million. Companies act as a laboratory of governance because anyone can leave and failures die quickly, while agencies never go out of business, so risk-tolerant staff leave and those who remain grow more risk averse. DARPA’s fixed terms are his model for automatic ten-year sunsets. He then traces Stripe-generated wealth through Terraform’s payroll, where almost half goes to payroll and income tax, and on into rent and groceries, estimating that roughly 70% disappears into government.

Starship, Louisiana and Setting Objectives

Handmer frames the Louisiana launch complex as regulatory arbitrage: if the Rio Grande Valley becomes too extractive, Elon will do to Texas what he did to California. He likes 10,000 flights a year as a power-of-ten north star but runs the math: a factory capacity of about 250 Starships a year and ten reuses each gives 2,500, so the target is unlikely within five years. Ships could launch from one site and land at another like airliners. On AI takeoff, he takes Baumol’s cost disease seriously and doubts universal high income because of the hedonic treadmill.

AI Superseding the Tax State, and Why Inequality Is the Wrong Metric

Handmer sees AI as swamping demographics, tax rates and drone warfare as variables. Income tax fit a 20th century economy of payroll labor, while future productivity will be “machines emitting streams of ones and zeros in low Earth orbit.” Most consumption is already nearly equal across incomes, with differences mainly in energy use and privacy. He does not envy the super rich, whose lives are more complicated and less private. He argues the optimum amount of inequality is not zero, and that the middle class shrank mostly because people rose out of it.

Notable Quotes

“If you want to do 100 tons a day, you don’t need to do like one Starship a day. You do it in 86,400 bite-sized pieces of a few kilograms at a time.”
Casey Handmer, on how a lunar mass driver would work

“The economic value that’s lost as a result of this kind of clot just with this one example is roughly equivalent to 10,000 lives.”
Casey Handmer, on the unrepaired California Aqueduct

“If you run the math on like what is the revenue per watt of transmitted microwave power for Starlink, it’s a billion times higher than space-based solar power.”
Casey Handmer, on why orbital data beats orbital power

“In the real world there’s no upper bound. You’re 10 times better than the next person. Good. Why aren’t you 100 times better?”
Casey Handmer, on why talent never saturates

“A lot of Elon’s advantage here is that actually he is very intuitive and his intuition is correct, rather than him sitting down in some secret chamber full of spreadsheets.”
Casey Handmer, on the source of Elon’s edge

“As a society, we’re better off with Elon on the outside trying to outgrow the spending state than on the inside trying to prune it.”
Casey Handmer, on Elon leaving DOGE

“I should be able to fit the laws of the land in an overhead locker.”
Casey Handmer, on 12 million pages of federal code and regulations

“What fraction of that value gets siphoned off by the government? It’s like 70%. It is unbelievable how much just disappears.”
Casey Handmer, tracing investor capital through payroll to workers’ groceries

“You can shear the sheep but you can’t skin them. If you shear the sheep, it makes more wool. If you skin it, it dies.”
Casey Handmer, on the Louisiana Starbase as leverage over Texas

“The middle class has disappeared in the United States because people have left it through the ceiling.”
Casey Handmer, on the inequality debate

Watch the full conversation with Casey Handmer here.