Ethan Thornton, founder and CEO of Mach Industries, describes building a defense company that vertically integrates critical components — solid rocket motors, jet engines, radars, warheads — to produce thousands of low-cost, high-performance autonomous systems per month, iterating through rapid block redesigns and scaling manufacturing before securing government production contracts.
Starting a solid rocket motor factory
Mach is industrializing solid rocket motor production at a 117,000 ft² facility near the former George Air Force Base in the California desert, with plans to open a second energetics factory twice that size within four months.
Solid rocket motors are essentially controlled explosives: millions of pounds of precursor materials must be stored, mixed, cast into machined casings, cured at precise temperatures, and then stored and shipped again — all while keeping humans out of the loop as much as possible.
The factory site provides rail access, an airport, and remote desert location for testing up to 60,000 lbf motors and hypergolic engines (one of two U.S. companies actively testing hypergolics at scale).
Mach acquired Squadum, a 20-year-old lab that designed cutting-edge solid rocket motors for the Air Force, NASA, DARPA, and SOCOM, to bypass 5-year licensing timelines and inherit deep IP across air-breathing, solid, liquid, and in-space propulsion.
Squadum’s founder’s saying — “jet engines are plumbing, solid rocket motors are baking” — captures the process: mixing a dough-like propellant, pouring it into casings, and curing it with exacting, arts-and-crafts precision that must be fully automated for rate production.
The goal is hundreds of thousands of motors per year within two years; Thornton believes solid rocket motors and warheads are the single biggest rate limiters on production of missile systems after jet engines.
Precursor supply chain is a major constraint: most U.S. precursor comes from one facility booked out three years; Mach is seeking partners to build dedicated precursor production on-site.
Smoothing out the boom-bust cycle in defense
Defense revenue is extremely lumpy: few multi-billion-dollar production contracts with binary win/loss outcomes, and the government controls the timing of production decisions.
Mach’s strategy: vertically integrate components needed for its own platforms, then sell those components (jet engines, solid rocket motors, radars) to other companies — about half of revenue is now B2B, not government-tied.
This creates a flywheel: platforms demand excellent components → vertical integration makes components excellent and cheap → selling components funds more vertical integration → platforms get better and cheaper → more component volume → lower unit costs.
Over a 10-year horizon, Thornton wants to push defense-developed technology back into commercial markets (as with rocketry, semiconductors, lithium batteries historically) not primarily for profit, but to generate capital to reinvest in defense and improve civilian life.
Creating tight iteration loops
Fast iteration is a balance, not a maximization: start with deep analysis, simulation, and design reviews to derisk every failure mode before flight; then redline the test machine once the simulation pipeline is mature.
Pike example: flight test today, rework vehicle overnight, fly again tomorrow — 10 flights in two weeks, no two identical, ~5% change per flight, building technical debt deliberately.
Technical debt must be periodically collapsed via block redesigns (complete new aircraft) — average cycle is 3–4 months; 3–4 blocks per vehicle before production.
You must be willing to blow up hardware; building production capacity first (e.g., one Pike per day) enables this tempo.
Block redesigns
Three block types: Block 1 — get something flying that resembles the platform (1% effort, learn ops, convince customer); Block 2 — manufacturable at hundreds to low thousands, optimize only for performance (40% effort, flexible processes like 3D printing); Block 3 — true rate optimization with deep DFM (e.g., 40,000 glide airframes/month at $7/airframe).
Each block cascades: learn → collapse → redesign; the goal is to compress the full whiteboard-to-production cycle from ~2 years to ~2 months.
Product iteration speed matters more than cost or performance in defense: asymmetry comes from fielding systems that create temporary advantage before adversaries copy or counter them.
Product iteration is more important than performance
Optimization target: dollar-for-dollar, CNC-for-CNC, technician-for-technician asymmetry against China (3x to 100x manufacturing scale advantage in consumer electronics).
Example: China outbuilds U.S. shipping tonnage 232:1; one container of Pikes (thousands/month production) can threaten ships — 1 Pike kills small ship, 3 kill large ship — creating asymmetry.
China will copy and counter; Mach is already building Dart to counter Pike-like threats, and the next product to bypass Dart — a continuous cat-and-mouse game where asymmetry windows shrink from years to months.
The machine must iterate products faster than adversaries while maintaining ability to retool factories within weeks for the next system.
Opex vs Capex
Mach prefers opex-heavy, capex-light manufacturing: capex must be pre-invested before contracts exist (chicken-and-egg), while opex scales only after winning; also gives government surge capacity without massive fixed cost.
Flexibility is the capex goal: retool factory in 2–3 weeks for a new aircraft type; slight unit cost increase is acceptable if DFM excellence offsets it.
Low automation initially: human labor stands up production fast, then automation follows; labor stays with company and moves to next product.
This approach is specific to defense’s uncertainty; for cars with predictable volume, Thornton would invert it.
Building a flexible organization
Functional (matrix) engineering org, not verticalized by product: same electrical, software, aero engineers work across products, specializing on one at a time.
Learning, hardware, and tooling generalize across products; each new product bet gets incrementally cheaper.
Enables rapid reallocation: engineers move from Pike to Viper without changing boss or tools; easier to spin down bets without organizational trauma.
Spikes in discipline workload (thermal, RF, structures) align across programs; impedance-match by keeping one vehicle in each development phase so spikes don’t overlap.
Investing in programs before there’s a contract
Most important skill for a defense company: deciding what to build and invest in before a contract exists, given long engineering lead times.
Process: start with geopolitics (Taiwan, Ukraine, Iran scenarios) → obsess logistics (launch, fuel, manufacturing, maintenance) → wargame against adversary capabilities and own future counters (e.g., Dart designed to shoot down Mach’s own products once copied) → anchor requirements → creative engineering pass to squeeze performance.
Closing the loop: wargame against yourself to accelerate the offense-defense cycle.
Asymmetric warfare
Asymmetry often comes from non-obvious missions: balloons (patented in high school, 2018) cost thousands at 80k ft but require million-dollar fighters/missiles to shoot down — unit economics favor the balloon; now proven in Ukraine and Chinese overflight incident.
Once a bet is proven, two things happen: you’re already positioned for the next layer (counter-balloon systems), and government trusts you with deeper problems.
Thornton’s edge: lifelong obsession with these problems (family dinner-table conversation), enabling earlier extrapolation than competitors.
Starting a new project from scratch
Functional org solves talent gaps: need one thermal lead (not one per program) who spikes at specific phases; same for RF, loads, reliability, etc.
40+ disciplines required; missing one fails the program (e.g., bad thermal design = 3-month slip on 6–12 month program).
Impedance-match workloads by staggering programs across phases so discipline spikes don’t collide.
Designing high output teams
Small teams (10–15 best people) outperform 300 top-1% engineers because aerospace is tightly coupled — communication overhead grows factorially with team size (SR-71 had ~15 design engineers).
Deep analysis (HIL, aero, every domain) enables speed; cutting prototypes fast requires vertical integration of prototyping + vendor relationships + willingness to pay for speed.
Excellent test organization is critical: no-fail test ops running 10–20 tests/week; safety lives here; component/subsystem testing must catch issues before flight — if you learn in air what ground test could have taught, you failed.
Manufacturing test excellence flows directly from component test excellence; manufacturing engineering faces 1,000 simultaneous breakages — good test data gives them orders of magnitude easier debug.
Build more vehicles than you think you’ll need
Overbuild test articles: cost of extra prototypes is negligible vs. 4-week wait for next batch; but guard against engineers flying them prematurely without ground-test understanding.
Prototypes should look distinctly different from production: choose billet machining over casting in proto for speed; DFM is the power-law lever (Glide Block 3→4: same specs, one process change cut operator hours from 50 to 0, cost from thousands to $7/airframe).
Stack rank for manufacturing excellence: 1) DFM, 2) Supply chain (invest, license, acquire, co-locate, or build — not just “buy stuff”), 3) Capital stack (opex/capex trades, product financial viability), 4) Manufacturing operations (99% of pain, but pain scales with how well first three were done).
From R&D to scaling manufacturing
Three stages: 1→10 (isolated, high risk, blow stuff up safely), 10→90 (deep customer coupling on connectors, launch boxes, fueling — revenue starts here), 90→100 (valley of death: government must commit to production; companies die here from overhead bloat if stuck too long).
Mach is in 10→90 across Pike, Glide, Stratus, Dart; Viper rearchitecting around Mach’s own jet engine; targeting production decisions in 6–9 months.
Unfair advantage: component factories (jet engines, solid rocket motors) already producing at rate — ex-Pentagon officials can visit a live line, not a slide deck.
Must not forget 1→10 while in 10→90; keep products at every stage to retain 1→10 engineers and culture.
Being a multi-product company from day 1
Thornton argues it’s harder to transition from single-product to multi-product than from zero to one: vertical integration, systematization, prototyping infrastructure all depend on multi-product intent.
Starting multi-product forces the right choices early; conventional wisdom (start single) is wrong for defense.
Vertical integration
Vertical integration is a tool, not a metric; Apple (partnerships, low capex, flexibility) and SpaceX (rotten supply chain, forced integration) are both excellent but opposite models.
For Mach: integrate where supply chain is rotten (solid rocket motors, micro-turbines, radars — 100–1000x cost inflation from cost-plus contracting and man-rated over-reliability) and where rate/surge control is existential.
Don’t integrate commodities (PCBs, visual cameras — Apple’s supply chain wins); do integrate defense-specific tech (IR/thermal cameras) via hybrid model (Mach design + partner production).
Core argument for Mach’s vertical integration: surge capacity — if Pentagon orders 1M units, Mach can deliver in 18 months because it owns the line; competitors share a supply chain capped at 5k split across Ukraine, Europe, legacy replenishment, and multiple primes.
Acquisitions
Strategy: acquire as little as possible; only when organic growth is impossible (specific IP, irreplaceable infrastructure like hypergolic test stands).
Tactics: merge cultures intentionally — take best of both, unify CAD/PLM/tools, avoid hubris; acquired company succeeded for a reason, Mach succeeded for a reason; end with homogeneous culture.
Squadum acquisition: 20-year IP bench + hypergolic test infrastructure + 10x faster SRM lead time quote → made sense.
Parallel pathing and sequencing
Programs have different lead times (software: weeks; jet engines, chips: years); stagger bets so they converge on a 5-year timeline.
Golden rule: don’t design aircraft and engine simultaneously (requirements orbit each other); anchor one, then the other (Viper anchored to off-the-shelf engine first, now redesigning around Mach jet).
Stack government sales lead time: engineering can be 3x slower and government buying is still the bottleneck — so run many programs in parallel to serialize the sales pipeline.
Started with balloons, fixed-wing drones, hydrogen (2018); hydrogen driven by: high-velocity guns (artillery range) and austere fuel generation (aluminum → hydrogen in field for artillery, fuel cells, drones).
Bet was directionally right (artillery surge, logistics vulnerability in Ukraine), but: engineering deceptively hard (leaks, embrittlement), unit economics failed (aluminum fuel cost never dropped despite salt ball-milling papers — likely fake), and $85M check enabled pivot to deterrence on 3-year timeline.
Spun down hydrogen at 100% of revenue, most engineering talent, investor opposition — novated contracts/IP to former Lincoln Lab lead Eric Limpacher, helped him stand up new company, retained float capital.
Mission-first framing made the painful decision easy: embarrassment vs. failing to deter WWIII / lose AI race.
Making painful decisions
Moved company from Austin to LA at 40 people after 3 months: 98% of talent pool in LA, hiring 1 engineer/month in Austin was fatal for 3-year deterrence timeline.
Failed leadership moment: tried consensus (“raise hand if you want to stay”) — everyone voted Austin; next day forced move anyway.
Most moved; two stayed. The move became a cultural filter: only mission-driven people uprooted families for a cash-poor startup — formed a distinctive, scalable culture.
Decision was objectively wrong in execution but right in outcome; burning boats (unintentionally) created cohesion.
Refounding Mach in LA
The Austin→LA move acted like “rehab + new environment” for a heroin addict: broke old patterns, forced cultural reinvention.
Institutions bloat and sclerotize; clean-slate reinvention works best when driven by a new hard challenge (e.g., Pike’s order-of-magnitude complexity jump after Viper).
Constant reinvention is necessary, but must be organic — not schemed.
Designing the company to scale 10x/year
Unscalable systems, deeply scalable people: hire executives who will lead in 10 years (top-down, once, painfully), then develop internal talent (metric: % of promoted leaders homegrown vs. hired).
Processes should always feel slightly broken — build for 3–6 months out, not 2 years (defense is too lumpy to predict); 20-year vision stays clear, 2-year is noise.
Exec team built over last year; lieutenants 80% done; remaining 20% are functions not yet stood up.
Leadership failings
Core theme: Thornton’s leadership reps lagged company scale (dorm room → factory in 15 months).
Rev A Viper disaster: top-down timeline pressure (“I think this should take 9 months” vs “for the mission this must take 9 months”) demoralized team, produced RC-plane-level product, cost 9 months.
Better approach: ask “why is this taking this long?” — aligns execs with engineers against the problem; admitting ignorance builds credibility; pushing only on things you’re confident in builds trust.
Job is to be dumbest in room on any given topic, but have the most developed world model of how pieces fit together and where they must be in years.
Bouncing back from failure
Company building is biologically easy (sitting, talking, eating) but emotionally terrifying and cognitively exhausting.
Discipline: separate fear/stress from decision-making — they are the worst decision makers; practice ignoring them until it becomes habitual.
Emotional inertia: don’t get too high on good weeks, bank emotional reserves for bad weeks; act objectively toward team/stakes/risk without personal fear — this is the best thing for all three.
