Austin Vernon is a chemical engineer and self-taught software developer who writes a widely-read blog on engineering, software, economics, and energy. He works pseudonymously, valuing the spread of his ideas over personal fame. The conversation spans his speculative writing on space weapons, software complexity, car manufacturing, his garage-based CO2 electrolysis project, and his paper on energy superabundance.
Starship as a Kinetic Weapon
Starship’s cost revolution: SpaceX has reduced launch costs from ~$20,000/kg (1990s) to ~$2,000/kg, with a path toward $200/kg. At that price, entirely new military applications become feasible.
Cheap mass on target: A Starship-derived “bomb truck” could deliver kinetic projectiles or guided munitions (like JDMs—Joint Direct Attack Munitions, ~$20,000 kits that turn dumb bombs into GPS-guided weapons with ~2m accuracy) at costs so low that missile defense becomes impractical through sheer volume.
Orbital basing: Keeping kinetic weapons in orbit reduces strike time from ~30 minutes (suborbital launch) to minutes. A two-stage “shroud” system would slow the weapon in the upper atmosphere before releasing it, preventing burn-up. Tungsten rods (the classic “rods from God” concept) deliver roughly the power of ~10 tons of TNT—significant but far from nuclear.
Defensive applications: Shrapnel packages could be released above ship radar ceilings to blind defenses, or heat-seeking packages could unfurl near aircraft. These are more plausible as area-denial tools than as city-destroying weapons.
Chips as a bottleneck: Russia’s advanced weapons depend on smuggled American chips and PLCs (programmable logic controllers) that are hard to replicate as integrated systems, not just individual components. This shortage cascades to countries like India that buy Russian weapons.
Broader lesson: When a technology gets 100-1,000x cheaper, it creates unpredictable new categories of use—just as cheap computing did.
Software and Complexity
Why software hasn’t produced runaway productivity growth: Digitizing complex processes requires capturing every corner case at the bit level. The easy automation goes first; what remains is exponentially more complex. This mirrors the “waterbed theory” of complexity—you can move complexity around but never eliminate it.
Toyota Production System as a model for software: Many software practices (Agile, small functions, APIs, type safety) are rediscoveries of principles from lean manufacturing. Toyota’s kanban cards reduce communication by encoding work instructions directly into the process, just as type systems encode constraints into code.
Local vs. global optimization: A core failure mode in both manufacturing and software is optimizing one part of a system while harming the whole. A factory that runs one machine at 100% utilization may just be building excess inventory. Tesla’s advantage over Toyota is partly that it can redesign the entire production process (e.g., giga casting that eliminates thousands of robots) rather than incrementally optimizing a mature system.
Vertical integration and firm size: Because complex processes require global optimization, firms that vertically integrate can coordinate more effectively than networks of vendors communicating through APIs. This implies software will increasingly be dominated by large firms—unless technologies like blockchain reduce inter-firm transaction costs enough to make decentralized coordination practical.
The Coasean framework: Firms exist because internal transaction costs are lower than market transaction costs. As software complexity grows, the coordination costs of working across firm boundaries favor larger organizations—unless regulation or technology changes the equation.
Car Manufacturing
Toyota’s no-layoff guarantee: After post-WWII labor unrest, Toyota agreed to lifetime employment in exchange for workforce flexibility. This aligns worker incentives with long-term company health and builds deep institutional knowledge, though it may be diminishing returns beyond ~5 years in a role.
“Metal manufacturing” (coined by Vernon): Tesla’s approach applies American-style creative production technology—inventing new alloys, using giga casting to eliminate parts—rather than Toyota’s conservative incremental optimization. Electric cars have fewer parts, making radical redesign easier.
Market cap vs. enterprise value: Tesla’s market cap ($1T+) vs. Toyota’s ($300B) is misleading because legacy automakers carry enormous debt. Enterprise value (market cap + debt) gives a more honest comparison.
Automation traps: Premature automation often creates local optimization and excess capacity. The Toyota approach is to simplify the process first, then automate—or eliminate the step entirely.
CO2 Electrolysis
The core idea: Use cheap electricity + water + CO2 + a catalyst to produce complex molecules like methane, ethylene, ethanol, or formic acid. This is a way to turn intermittent solar/wind energy into storable liquid fuels and chemical feedstocks.
Scale: Current global CO2 electrolyzer output is measured in kilograms per day; industrial demand is millions of tons per day. Massive scale-up is needed.
Vernon’s approach vs. Terraform Industries: Terraform makes hydrogen via electrolysis, then combines it with captured CO2 in a separate methanation step—conservative, using proven world-scale processes. Vernon wants to make the target molecule directly in the electrolyzer, which could be simpler and scale down better.
Why flat plates: Electrolyzers use flat-plate geometry (anode and cathode pressed together to minimize electrical resistance), which scales by simply making the plate larger—unlike traditional chemical engineering, which scales poorly to small sizes.
Energy cost crossover: At ~$10-20/MWh electricity, CO2-derived fuels become competitive with fossil fuels. Solar is already ~$25/MWh in Texas and falling. Vernon is building a prototype in his garage, supported by family, and is focused on reliability, product concentration, and manufacturability.
Commodity limits: Chemical commodities have thin margins and limited defensibility. Even with a great process, competition from a few rivals compresses returns. The window for small-scale entry may close as the technology matures.
Energy Superabundance
The thesis (from Vernon’s paper with Eli Dourado): If energy becomes extremely cheap (~$10/MWh), it unlocks transformative applications even if per-capita energy use has plateaued in rich countries since the 1970s.
Diminishing returns to efficiency: Past GDP growth has come partly from using energy more efficiently. Cheap energy flips this: you can use more energy to get more output (e.g., shifting freight from efficient trains to flexible trucks, or from ships to planes).
Transportation revolution: Electric planes with even 500-mile range could dramatically increase air freight (which has extreme elasticity of demand). This reduces working capital costs (goods spend days on planes instead of weeks on ships) and enables more specialization.
Batteries are not the binding constraint: Most energy-intensive processes (chemical production, crop growing, desalination) can be designed to run when energy is available, without storage. Batteries matter most for transportation and air conditioning—uses where people will pay a premium. Abundant chemistries (lithium iron phosphate, carbon-iron) can scale without hard material limits.
Cities and VTOLs: Cheap energy enables electric VTOL aircraft (vertical takeoff and landing), which would use small general aviation airports rather than large hubs, bypassing TSA bottlenecks. Tunnels (à la Boring Company) allow new surface roads without eminent domain. Cities become effectively larger as commute times shrink, but walkable neighborhoods may become more desirable even as people commute farther.
Carbon shortage: If CO2 becomes a valuable feedstock for fuels and plastics, and if we simultaneously decarbonize, we could face a carbon shortage by end of century. Companies like Charm Industrial are already sequestering bio-oil underground—a potential carbon reserve.
Nuclear: Vernon is skeptical of large nuclear (light water reactors can’t reach $10/MWh; alternative coolants like sodium, lead, or molten salt introduce serious engineering challenges) but excited about small/micro reactors. The NRC’s probabilistic risk assessment framework has actually made nuclear very safe (US plants run at ~90% capacity factor), but the licensing process is expensive and slow regardless of design safety. Small reactors like the “KRUSTY” project (1 kW, $20M, built in years) have potential in remote power, space, and military markets where they’re not substitutable.
Fusion: Vernon is pessimistic about fusion approaches that use steam turbines (too expensive, no learning curve left) but optimistic about direct energy conversion technologies.
Solar vs. subsidies: Vernon argues solar subsidies were unnecessary—Germany’s feed-in tariffs spent heavily to scale a technology that someone would have developed anyway. The learning curve would have been climbed regardless.
Finding Alpha in Efficient Markets
Vernon’s framework (based on Fama): Markets are efficient at pricing public information. Excess returns come from legally acquired private information, which includes information generated through labor.
Labor as alpha: Warren Buffett’s early returns came from deep operational research—visiting factories, reading 10Ks, interrogating management. This is labor-intensive private information. As Buffett’s capital grew, his returns fell because the marginal dollar is harder to deploy skillfully. Capital is fungible; skilled labor is not.
Contrast with Piketty: Piketty argues capital earns higher returns than labor grows. Vernon inverts this: the highest returns go to skilled labor (especially specific knowledge and brand), which then attracts capital. Capital is a complement to labor, not a substitute.
Brand as specific knowledge: Y Combinator earns excess returns because its brand attracts the best founders—built on specific knowledge of what founders want. This is a form of labor-generated alpha.
Blogging market: Blogging returns are hard to measure and probably inefficient. Vernon’s own blog led to connections that helped launch his CO2 electrolysis project—returns that are potentially enormous but unpredictable. Blog prizes may help but don’t fully address the measurement problem.
