YT YouTube2Text

YouTube2Text → Transcripts → How SpaceX Works

How SpaceX Works — Transcript

by Founders Podcast · 6,687 words · 1,162 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00A few years ago I started working on a
0:01book called SpaceX Foundation, a
0:03historical account of SpaceX's first
0:05decade told through first-hand sources.
0:08First-hand sources like Elon's company
0:10updates, launch dispatches, internal
0:12memos. This is the real-time record of a
0:14company that almost died three times and
0:17then became the most dominant launch
0:19provider on Earth. The gap between
0:22SpaceX and everyone else is enormous and
0:25widening. Yet most of what's been
0:26written focuses on Elon himself, not on
0:29the specific methods, cultures, and
0:31decisions that actually built the
0:32company. That is what the book is about.
0:36While the book is still in progress,
0:38I've been writing an introduction essay
0:40as a way to work through the central
0:42question. Why did SpaceX succeed in ways
0:45no one else has been able to replicate?
0:47And more importantly, is any of it
0:50learnable? The practices that made
0:52SpaceX dominant aren't unique to
0:53rockets. They're a blueprint for
0:55building anything hard. That's the
0:57introduction to this introduction essay
1:00of this book. So, the introduction essay
1:01is called Atoms are Cheap, Process is
1:04Pricey, What SpaceX Teaches Us About
1:06Building Hard Things. It is written by
1:08Max Olson, who is writing that book
1:10called SpaceX Foundation. I've read this
1:12essay three times. I think it's really
1:13good, so I want to go through some of
1:15the main ideas with you. And so, the
1:16essay starts like this. SpaceX has been
1:18remarkably open about how they operate.
1:20They've been succeeding in public for
1:21more than 15 years now, and yet no one
1:23has replicated the results. Competitors
1:26know their strategy. The engineering
1:28philosophy gets explained in interviews,
1:30tweets, and factory tours. Many of the
1:32ideas aren't even new. Lockheed's Skunk
1:34Works ran similar approaches 60 years
1:36ago. Founder Kelly Johnson's 14 rules
1:40read like a SpaceX operations manual.
1:43The performance gap just keeps getting
1:44bigger. In 2025, SpaceX launched more
1:47mass to orbit than every other provider
1:50on Earth combined. Much more. This is
1:53crazy. Every payload from China, Russia,
1:57Europe, and all American launchers
1:59wasn't even a fifth of what SpaceX put
2:02into orbit. They're the only company
2:04producing rockets at an industrial
2:06scale. A Falcon 9 goes up every two to
2:09three days. Competitors manage
2:11single-digit launches per year. The same
2:14boosters have been reused 20 times each.
2:18The company has sent astronauts to the
2:19International Space Station, the first
2:21private company to do so. Starlink,
2:23their satellite internet constellation,
2:25now has over 9,000 satellites in orbit,
2:28the largest in history.
2:30Both built and launched by the same
2:33company.
2:34SpaceX is now the most valuable private
2:37company on the planet. Yet the skeptics
2:39were confident it couldn't happen.
2:41Apollo astronauts Neil Armstrong and
2:43Gene Cernan testified before Congress
2:45against commercial spaceflight. They
2:47said that they thought reusability is a
2:48dream, and even if it did work, the
2:50market was too small to support the
2:52hundreds of launches needed to make
2:54reusability worth it. Elon was described
2:57as a software guy playing with expensive
2:59toys. The early failures seemed to
3:00confirm them. Three Falcon 1 explosions
3:03between 2006 and 2008. By September
3:052008, SpaceX had funds for exactly one
3:07more attempt, and Tesla was weeks away
3:09from bankruptcy. Elon was borrowing
3:11money for rent. Then it worked. Flight 4
3:15succeeded, and NASA's $1.6 billion cargo
3:18contract followed 6 weeks later.
3:20Then came Falcon 9, Dragon, ISS docking,
3:23boosters exploding on the pad, boosters
3:26landing, crewed flights, and eventually
3:28Starship. So is any of this outlier
3:30performance repeatable?
3:32This is the puzzle.
3:34If the strategy is known and the
3:35principles are public, what's actually
3:37hard to copy?
3:39Obvious factors explain some of this,
3:41but not enough.
3:42And so he goes over some of the factors.
3:44The Space Shuttle retired, creating a
3:46gap. This was really good timing for
3:47NASA to become SpaceX's biggest
3:49customer. But Blue Origin was founded 2
3:51years earlier, and Boeing and Lockheed
3:53saw the same opportunity. The grand
3:55vision of boots on Mars attracted
3:57missionaries, but ambitious visions are
3:59cheap, and plenty of founders have them.
4:02Elon putting in a hundred million
4:03dollars bought early runway, but Bezos
4:05poured much more into Blue Origin, and
4:07legacy primes had multiples of this
4:09amount. Technology was also getting
4:11better. 3D printing, simulation,
4:12advanced materials, all commercially
4:14available to competitors. These factors
4:16are real, and none are sufficient. If
4:18they explained it, others could have
4:20caught up easily, but they're not even
4:21close. SpaceX is a hotbed of case study
4:24material. Engineering, product, finance,
4:27strategy, manufacturing, product
4:29management, etc. If you're interested in
4:32the company, these are all important,
4:33but I'm more curious about which are
4:35repeatable. The question isn't why did
4:37SpaceX succeed? That's too vague to be
4:39useful. The sharper question, what can
4:41someone building hard things actually
4:43take away? And so then Max breaks it
4:45down into a bunch of subheadings. The
4:47first one is the strategy.
4:49What SpaceX has done more than anything
4:51is minimize the cost of getting things
4:53to space. The vision is humanity
4:56expanding across our solar system, but
4:57the lever is the cost of moving mass
5:00from Earth's surface to orbit and
5:02beyond. Everything else, the launches,
5:05the landings, the reuse, serves that
5:07goal. When you study how companies hold
5:09advantages over time, consistently being
5:12the low cost provider might be the
5:14hardest to maintain. The reason is that
5:16it has to be baked into everything you
5:18do. It cannot be an initiative or an
5:21afterthought. It has to shape how you
5:23design products, structure the company,
5:25and choose what to build. And as you'll
5:27see in the book, it all started from the
5:29earliest days. Before starting SpaceX,
5:32Elon wanted to get to Mars, but he
5:33didn't set out to build a rocket
5:35manufacturer. In 2001, he tried buying
5:38Russian ICBMs to get there, but the
5:41Russians quoted him ridiculous prices.
5:44So he famously reframed the question
5:46from first principles. What is a rocket
5:48made of? Aerospace grade aluminum alloys
5:51plus some titanium, copper, and carbon
5:53fiber. And then I asked, what is the
5:55value of those materials on the
5:56commodity market? It turned out that the
5:58materials cost of a rocket was around 2%
6:00of the typical price, which is a crazy
6:02ratio for a large mechanical product.
6:052%. Your car's raw materials are maybe
6:0820 to 30% of the sticker price.
6:11Consumer electronics are similar, but
6:13rockets 98 cents of every dollar was
6:17going somewhere other than what it was
6:18made of. Where was it going? Three
6:21places, it seems. Supplier markups
6:23stacking through contract layers, each
6:25tier adding 15 to 30% margin. Custom
6:28designs that couldn't achieve
6:30manufacturing scale, and expendable
6:33hardware thrown away after every flight.
6:35None of these are the laws of physics.
6:38Traditional aerospace treated high costs
6:40as fixed constraints, but what if you
6:42treated them as variables? How do you
6:44actually capture that 98%?
6:47I'm reading an early copy of this book
6:49called The Book of Elon. It's written by
6:50my friend Eric Jorgenson. It'll be out
6:51in a few weeks. It's about 200 pages of
6:53just Elon in his own words, and he was
6:55actually talking about another way to
6:56think about this. This is what Elon says
6:58from this book. How could the Russians
6:59build low-cost rockets? It's not like we
7:01drive Russian cars, fly Russian planes,
7:04or have Russian kitchen appliances. The
7:06US is a pretty competitive place, and we
7:08should be able to build a cost-efficient
7:10launch vehicle. So, back to the essay.
7:13Another subheading, rethink from first
7:14principles. Start with the actual
7:16product. If you accept existing
7:17solutions, you accept their cost
7:19structure. So, rebuild from physics
7:21instead. Don't ask, what do rockets
7:23cost? Ask, what should rockets cost?
7:26Elon eventually named this the idiot
7:28index, the ratio of the actual cost of a
7:30part to the cost of its raw materials.
7:32If the ratio is high, he says, you're an
7:34idiot. Consider the Falcon 1 actuator. A
7:38vendor quoted $120,000
7:41and 18 months of development. SpaceX's
7:44engineers built it for $3,900.
7:48When founding engineer Tom Mueller's
7:50team asked about a critical engine
7:52valve, the supplier kind of smirked and
7:54left after hearing SpaceX's timeline and
7:56budget. Mueller's team made the valve
7:59themselves. This pattern repeated across
8:02the vehicle.
8:03The Dragon's capsule docking mechanism
8:06was reinvented from off-the-shelf bike
8:09shocks and catalog parts instead of
8:11adopting NASA's existing design. There
8:14are probably a hundred examples like
8:16this, most not discussed in public.
8:19The philosophy extended to fundamental
8:21architecture. SpaceX uses one propellant
8:23pair,
8:24liquid oxygen and RP-1 kerosene, across
8:28all stages.
8:29The vacuum Merlin engine uses a fixed
8:32nozzle extension instead of a deployable
8:34one.
8:35This is the main point. Fewer moving
8:37parts means fewer failure modes means
8:39lower cost. Compare this to the Atlas V
8:43rocket, which uses up to three different
8:44rocket types in a single vehicle, each
8:47optimized for its flight phase. Elon's
8:50response to this was, "You've just
8:52tripled your factory costs and all your
8:54operational costs."
8:56And so he continues to give examples of
8:57how they think in all these little
8:58decisions they're making. The Merlin
8:59engine family embodies this trade-off.
9:01Russian RD-180 engines cost 20 to 25
9:04million each, while the Merlin 1D
9:06production runs around a million.
9:09How? SpaceX eliminated hydrogen's
9:11complexity by using kerosene, used
9:13regenerative cooling with the fuel
9:15itself, and optimized for
9:16manufacturability over maximum
9:18performance. The result was 95%
9:21of theoretical efficiency for 80% cost
9:25reduction. Merlin's performance is
9:27slightly lower, but good enough at
9:281/20th the cost.
9:30But identifying where to cut costs
9:32doesn't mean you can actually cut them.
9:33You still have to build the parts. Once
9:35SpaceX concluded that atoms were cheap
9:38and process was expensive, vertical
9:40integration followed almost inevitably.
9:42That's a great line. SpaceX concluded
9:43that atoms were cheap and process was
9:45expensive and so therefore vertical
9:46integration followed almost inevitably.
9:48The next subheading talks about this,
9:50becoming your own supplier.
9:52If materials are cheap and the tax is
9:54all process and overhead, you need to
9:56control the process to capture the
9:57savings. You cannot negotiate your way
10:00to a 10x cost reduction with suppliers
10:02who have profits baked in at every tier.
10:05So SpaceX became its own supplier.
10:07By building 80% of its hardware
10:09internally, engines, structures,
10:11avionics, software, and key ground
10:14systems, SpaceX collapsed the
10:16traditional aerospace stack. They
10:18outsource raw materials and commodity
10:20parts and make everything else
10:21themselves.
10:23"That's something SpaceX didn't
10:24originally set out to do," one engineer
10:26noted, "but was driven by suppliers'
10:28high prices. This wasn't an ideological
10:30commitment to doing everything in-house.
10:32It was the result of suppliers
10:34repeatedly quoting prices and timelines
10:36incompatible with SpaceX's cost
10:39targets."
10:40The benefits compound. When several
10:43tiers each add 15% margin, total cost
10:46multiplies through the layers. A NASA
10:48study found SpaceX developed Falcon 9
10:51for roughly 440 million. They estimated
10:54the same work with traditional
10:56contractors would have cost three to 10
10:59times that amount.
11:01Vertical integration also accelerates
11:03iteration, another great line. Vertical
11:05integration also accelerates iteration.
11:07When an engineer needs to change a
11:08bracket, weld, or circuit board, the
11:10manufacturing engineer is in the same
11:12building using the same CAD systems and
11:14tooling. Materials, jigs, and processes
11:17can evolve together on the scale of
11:18weeks, enabling a rapid progression from
11:21Falcon 1 to successive Falcon 9
11:24variants, each iteration improving
11:26performance and reducing cost without
11:28waiting for suppliers to retool on
11:30multi-year cycles.
11:32But integration creates a new problem.
11:34It concentrates fixed costs. If you own
11:36the factory, the machines, and the
11:38staff, you're losing money every second
11:40they aren't building something. At the
11:41traditional launch cadence of two to
11:43four vehicles per year, in-house
11:45manufacturing is a liability, not an
11:47advantage. To make the math work, you
11:49need volume.
11:51Before we get back into this, I want to
11:53tell you about the presenting sponsor of
11:54this podcast, Ramp. One of the main
11:56themes in the history of SpaceX is
11:57constantly attacking and questioning
11:59your costs. Ramp helps many of the most
12:01innovative businesses in the world do
12:02exactly that. The median company running
12:05on Ramp cuts their expenses by 5%, and
12:08just like SpaceX has demonstrated a
12:10religious dedication to controlling
12:11costs, helps increase revenue and pursue
12:13opportunities you couldn't otherwise. We
12:16see that in the Ramp data, too. The
12:18median company running on Ramp also
12:19grows their revenue by 16%. Many of the
12:22top founders and CEOs I know run their
12:24business on Ramp. I run my business on
12:26Ramp, and you should, too. Go to
12:28ramp.com today to learn how it can help
12:30your business save time and money. That
12:32is ramp.com. So, that brings us to the
12:34next subheading, build a platform.
12:37The only way to get volume is to
12:38standardize. Build a common platform
12:40that customers have to adapt to. The
12:42existing approach was bespoke vehicles
12:44per mission, custom adapters,
12:46mission-specific modifications, multiple
12:49vehicle families. This optimizes each
12:50mission at the expense of manufacturing
12:52scale. SpaceX bet the opposite, that
12:56cost savings from standardization would
12:58exceed the value of customization.
13:01Yes, customers wanted custom solutions,
13:03but they wanted low prices even more.
13:05Force them to choose, and they will
13:07adapt. The Falcon 9 became the
13:10industry's Model T, one rocket built in
13:12volume. Same nine Merlin engines on the
13:14first stage, same vacuum Merlin on the
13:17second. Same structure, same diameter,
13:19same aluminum lithium alloy, same
13:22welding methods, same avionics, same
13:25ground systems.
13:26Even the Heavy is just three Falcon 9
13:29first stages strapped together with a
13:30shared upper stage.
13:32A scaled variant from the same core, not
13:35a new vehicle.
13:37SpaceX published a Falcon User's Guide,
13:40which defined bolt circles, electrical
13:42connectors, and fairing environments.
13:44Customers designed to SpaceX's spec
13:47instead of demanding customizations.
13:49Satellite orbits adjusted to Falcon
13:52performance curves. This flipped the
13:55negotiating power. Instead of aerospace
13:57companies serving satellite
13:58specifications, satellites adapted to
14:01SpaceX capabilities.
14:03The economics of manufacturing is what
14:05makes this work. Building 40 identical
14:07Falcon 9s annually creates automotive
14:09style learning curves that are
14:10impossible in custom aerospace. As
14:13production scales, learning improves and
14:16cost declines.
14:18How this worked in practice is that
14:20every anomaly, wear pattern, or
14:22manufacturing defect fed back directly
14:24to the teams that design the parts.
14:28Finally, there's the logical conclusion
14:30of standardization, which is
14:31reusability. Reusable boosters are still
14:34the same Falcon 9 cores. You aren't just
14:37building the same model of rocket, you
14:38are literally flying the exact same
14:40hardware. Because every booster was
14:42identical, every landing attempt
14:44provided perfectly comparable data.
14:47If making 40 rockets creates a
14:49manufacturing learning curve, flying the
14:51same rocket 20 times creates an
14:54operational learning curve that's even
14:56steeper. The economics are devastating
14:59for competitors.
15:01Traditional providers launching a
15:03handful of custom vehicles per year
15:05never accumulate enough data to even
15:07start this cycle. And this is so good
15:10how Max ties this all together. You can
15:12probably see why all three tactics were
15:13necessary. First principles identified
15:16the waste. Vertical integration provided
15:18the control to eliminate it.
15:20Standardization allowed the volume to
15:22make that control profitable. Without
15:25all three, the system breaks. They work
15:28together so each flight makes the next
15:31one cheaper. What does this sound like?
15:33A flywheel. Lower costs enable lower
15:36prices, which capture market share,
15:38which increases volume, which drives
15:40costs lower still.
15:42The incumbents understood this too late.
15:46They optimized components locally.
15:48Better engines, lighter materials,
15:50incremental gains. SpaceX optimized the
15:52system for cost, accepting component
15:54level compromises for system level
15:58dominance.
15:59This flywheel relies on high volume.
16:02Competitors couldn't even imagine there
16:04was even a market for so many rockets.
16:06And they weren't willing to take the
16:07risk of spending all the money just to
16:09find out. In a world where atoms are
16:11cheap and processes expensive, the real
16:13innovation was not a single engine or
16:15material, but the decision to redesign
16:17the entire stack around the economics of
16:19cost. But a cost target doesn't build
16:22itself.
16:23First principle strategy says what to
16:25build. It doesn't say how to build it
16:28without catastrophic mistakes along the
16:30way.
16:31On to next subheading, the engineering.
16:33If the strategy is to rethink everything
16:34from first principles, how do you
16:35actually execute that without major
16:37consequential failures? The standard
16:39answer is to analyze exhaustively before
16:42building. So he's going to describe the
16:43standard answer and I love how he talks
16:44about SpaceX literally inverted this.
16:47Traditional aerospace follows this path
16:48religiously. A NASA report on the
16:50commercial crew program noted that
16:52Boeing utilizes a well-established
16:54systems engineering methodology targeted
16:56at an initial investment in engineering
16:58studies and analysis to mature the
17:00system design prior to building and
17:02testing.
17:03Must be really fun to read these kind of
17:04reports. Plan extensively, freeze
17:07requirements early, minimize test
17:09failures. This is the measure twice, cut
17:11once approach. SpaceX inverted this.
17:14Here's the problem with the traditional
17:15approach. You can't think your way to
17:17perfect solutions for problems you don't
17:19fully understand. That's another bar.
17:21You can't think your way to perfect
17:23solutions for problems you don't fully
17:25understand.
17:26Your model is always wrong in the ways
17:28you don't know yet.
17:29Complex systems have emergent behaviors
17:31that only appear when the pieces are
17:33actually bolted together.
17:35This is the paradox of first principle
17:37design.
17:38If you're questioning every inherited
17:39assumption, which you should, you're
17:41venturing into territory where analysis
17:44alone can't tell you what works.
17:46The physics might be known, but how the
17:48physics will interact with your specific
17:50materials, your specific manufacturing
17:52tolerances, your specific assembly
17:54process, that's not something you can
17:56derive from first principles. That's
17:58something you have to discover.
18:00And so what SpaceX does instead is they
18:02use reality as their validation tool.
18:04The alternative is to use reality as
18:06your primary validation tool. SpaceX
18:08focuses on rapidly iterating through a
18:11build, test, learn approach that drives
18:14modifications towards design maturity.
18:17Where Boeing invests up front in
18:18analysis, SpaceX invests up front in
18:21prototypes. The core philosophy is
18:24deceptively simple. Failures are data,
18:27not disasters.
18:29Tight feedback loops lead to a high rate
18:31of innovation and adaptation, quickly
18:33finding better solutions of what not to
18:36do. Speed is a tactical advantage.
18:40This isn't new. During World War the
18:43P-80 fighter jet went from concept to
18:46test flight in 5 months. In the early
18:491960s, the SR-71 Blackbird went from
18:52idea to roll out in 4 years. And it's
18:56still the fastest man plane ever built.
18:58That's nuts. Is that Could that possibly
19:00be true? In the early 1960s, the SR-71
19:02Blackbird went from idea to roll out in
19:054 years and it's still the fastest man
19:06plane ever built. Small teams, fast
19:10iteration, real hardware.
19:12In a 2021 Starbase interview, Elon
19:15explained the goal with each Starship
19:17prototype. Push the envelope such that
19:19it blows up. That is the end of his
19:21quote. That That is his whole quote.
19:23Push the envelope so it blows such that
19:25it blows up. This sounds reckless until
19:27you understand what he's actually
19:28saying. If the vehicle doesn't fail, you
19:30haven't learned where the limits are.
19:32Each failure is a precise data point
19:34about where reality diverges from your
19:36model.
19:37How can you reconcile this fail fast
19:39approach with the care that's needed to
19:41reliably build things where human lives
19:43are on the line?
19:44The crucial distinction is between
19:46development and operations. SpaceX runs
19:48both, but with completely different risk
19:51profiles. So, for the Dragon, it carries
19:53crew.
19:54It can never fail. So, there's large
19:56margins of safety, there's exhaustive
19:58testing, there's conservative
19:59everything. The Falcon 9, which is the
20:01operational launch vehicle, it's middle
20:03ground. It says ascent can't fail, but
20:05some landing attempts are allowed to.
20:07And then Starship, which is development,
20:09failure is instrumental.
20:11And Elon has a quote about this. He
20:13says, "Starship does not have anyone on
20:14board so we can blow things up. It is
20:16really helpful." And this is a good way
20:18to think about this. This is the same
20:19company doing two very different things
20:22with two very different groups of people
20:24and two very different risk profiles,
20:26but they're talking to each other. And
20:28so, if you go back to their very first
20:29four launches, it says, "The contrast
20:31with traditional aerospace is stark. In
20:33that world, a three failure start may
20:35have triggered years of analysis, review
20:37boards, and redesigns on paper before
20:39the next attempt." At SpaceX, each
20:41flight became the next test with fixes
20:43incorporated immediately. This pattern
20:46continued into Starship.
20:48The early integrated flights each ended
20:51in rapid unscheduled disassemblies,
20:53which is just their name for explosions.
20:55But each came after achieving partial
20:57objectives, such as clearing the pad,
20:59passing the match max Q, reaching near
21:02orbital velocity.
21:04Then finally, the famous catch of the
21:06Super booster, which is the equivalent
21:09of catching a 20-story building that's
21:11falling from the edge of space. Each
21:14subsequent flight incorporated design
21:15changes based on telemetry from the
21:17previous one. Where traditional
21:19aerospace might take years to go from
21:21flight anomaly to design change, SpaceX
21:24was doing it in between flights. Next
21:27subheading, have a high production rate.
21:29Iteration only works if you can afford
21:30many attempts. This is where SpaceX's
21:32hardware-rich approach becomes
21:33essential. This is what Elon says about
21:35this. A high production rate solves many
21:38ills.
21:39He has said this repeatedly. He
21:40continues, "Any given technology
21:42development is how many iterations do
21:45you have, and what's your time and
21:47progress between iterations? So, if you
21:49have a high production rate, you can
21:50have a lot of iterations. You can try a
21:53lot of different things. If you have a
21:55small number of engines, then you have
21:57to be much more conservative because you
21:59can't risk blowing them up."
22:01SpaceX builds many cheaper prototypes,
22:04hardware-rich fleets of test articles.
22:08They'd rather have 10 rough versions to
22:10blow up than one polished version
22:12they're afraid to break.
22:14This can lead to specific design
22:16decisions like using stainless steel for
22:17Starship, which is cheap, easy to weld,
22:20and can be welded in a tent, by the way,
22:22instead of carbon fiber, which is
22:23expensive and requires giant autoclaves.
22:26Vertical integration really helps enable
22:28this. When you own the factory, you can
22:29build fast without waiting on vendors.
22:31When you own 3D printing capability, you
22:34can produce parts on an ad hoc basis.
22:36When you can manufacture Raptor engines
22:39at high volume, losing one to a test
22:41failure doesn't set you back months.
22:44SpaceX is big into simulation as well.
22:47This moves atoms to bits where possible,
22:51letting them pre-screen designs before
22:53blowing things up. But real tests remain
22:55primary. The question being constantly
22:57asked is how quickly can it be tested in
22:59as real environment as possible. And
23:02again, Max does a great job of tying
23:03this all together down here. The pieces
23:05reinforce each other in a way that's
23:07easy to miss. First principles
23:09engineering reduces unnecessary
23:10complexity. Fewer parts means each
23:13prototype is cheaper to build. Cheaper
23:15prototypes mean you can build more of
23:17them.
23:18More prototypes means faster iteration.
23:21Faster iteration means you can push each
23:23prototype to failure without being
23:26precious about it. More failures means
23:29more data.
23:30More data means better designs.
23:33Better design means even simpler
23:35solutions, and the cycle continues.
23:39Meanwhile, fail fast iteration extracts
23:42maximum information per prototype and
23:45per flight. You're not just testing
23:47whether something works. You're finding
23:48exactly where it breaks. That precision
23:51accelerates the next iteration. The
23:53strategy, low cost, vertical integration
23:56enables the engineering approach. The
23:58engineering approach validates the
24:00strategy faster than analysis ever
24:02could. Traditional aerospace eliminates
24:05uncertainty through planning. SpaceX
24:08eliminates uncertainty through doing.
24:11But this system doesn't run itself.
24:13Running an organization that constantly
24:14questions requirements, deletes parts,
24:16and accepts visible failures requires
24:19something that can't be built through
24:21iteration. An engineering process that
24:23treats failure as data only works if the
24:25engineers themselves believe it.
24:28A system that pushes to the edge of
24:30what's possible only survives if the
24:32people doing the pushing can handle the
24:34intensity.
24:35The practices described here are the
24:37mechanism, but they're powered by
24:39something else entirely. And that goes
24:41to his next subheading, which is the
24:43people.
24:44And so Max writes, back to my original
24:46point. The practices I've described so
24:47far aren't secret. So why can't others
24:49just copy them?
24:51The standard answer is organizational
24:52inertia, bureaucracy, risk aversion, and
24:55And yet, there's truth to all of these,
24:57but it's not the whole story. The answer
24:59is that strategy doesn't exist in
25:00isolation. The same playbook in a
25:02different environment would produce
25:03different results or nothing at all. You
25:05can't copy strategy without
25:07transplanting the conditions that make
25:09them work.
25:10A fail-fast culture needs people willing
25:12to fail visibly. A first principles
25:14approach requires people willing to
25:16question experts. Skip-level
25:18truth-seeking requires people willing to
25:20deliver bad news directly to the CEO.
25:23The variable I've been circling around
25:25is people. Not in the bland HR sense of
25:27our people are our greatest asset. In
25:29the structural sense, who shows up, what
25:32they believe, and what behaviors they're
25:34willing to accept from each other.
25:36SpaceX didn't just hire good engineers.
25:38It built a system that attracts,
25:40retains, and amplifies a particular kind
25:43of engineer while filtering out everyone
25:45else.
25:47The variable starts with Elon. Without
25:49him, none of this exists.
25:51I don't say this as a hagiography, only
25:53an observation about initial conditions.
25:55Someone had to fund a rocket company
25:58when the idea seemed crazy. Someone had
26:00to decide that colonize Mars was an
26:02actual engineering target. I want to
26:05call out three factors in particular
26:06that were foundational to SpaceX's
26:08culture. The first, an ambitious vision
26:10that functions as a recruiting filter.
26:13Building cities on other planets isn't
26:15just aspirational branding. It's a
26:17sorting mechanism.
26:18The mission attracts missionaries.
26:21Engineers who would never work for just
26:23another launch company will work brutal
26:25hours for a shot at making humanity
26:27multi-planetary. When the mission is
26:29that clear, prioritization becomes
26:31automatic. That's another great line.
26:33When the mission is that clear,
26:34prioritization becomes automatic. Every
26:36decision has a simple test. Second,
26:39constant forcing functions both real and
26:41manufactured.
26:42I think this part is underrated. Some
26:45were genuinely existential, like the
26:472008 cash crisis when SpaceX had funds
26:49for exactly one more Falcon 1 attempt.
26:51Others Elon created himself, aggressive
26:53public timelines that even he knew were
26:56ambitious.
26:57Internal do-or-die milestones that felt
26:59real even when they weren't legally
27:01binding. The forcing function prevents
27:04drift. You can't endlessly study when a
27:07deadline, real or perceived, is bearing
27:08down. Even an arbitrary deadline is
27:11better than no deadline because it
27:12forces decisions. Third, direct
27:15technical engagement that bypasses
27:17organizational filters.
27:19Andrej Karpathy, who worked for Elon at
27:21Tesla, notes that Elon spends about 50%
27:23of his time talking directly to
27:25engineers, not to VPs summarizing
27:28engineering work.
27:29This sounds obvious, but it's unusual in
27:31the corporate world. The CEO has to
27:33trust the CTO, who works through layers
27:35of managers to enact a vision. Each
27:36layer is a hop where information is
27:39lost. It's like a game of telephone. By
27:41the time the technical reality reaches
27:43the CEO, it has been polished, caveated,
27:45and de-risked. It is a summary of a
27:47summary with the inconvenient details
27:49removed. SpaceX collapses the chain. The
27:52CEO, CTO, VP, engineer layers become a
27:56single conversation. By talking directly
27:58to the engineers, Elon removes the
28:00signal loss. They become the source of
28:02truth, not the filtered narratives that
28:04typically reach the CEO. This isn't just
28:07about speed and accuracy. It also allows
28:09SpaceX to make bolder technical bets.
28:11Elon stays aligned on what is actually
28:13possible.
28:14A non-technical manager can't tell the
28:16difference between a tactically painful
28:19path and a strategically necessary one.
28:22If your managers tell you a certain
28:23chassis design or an engine material is
28:25too difficult, and you don't have the
28:27technical depth to interrogate them, you
28:29have to defer. The decision to use steel
28:32for Starship over carbon fiber is a
28:33prime example. It was highly against
28:36conventional wisdom and controversial
28:38even within the company. Ultimately,
28:41Elon had to understand all the
28:42trade-offs himself and make a call. When
28:45there's a blocker, such as a rapid or
28:46production GPU supply, a regulatory
28:49delay, Elon intervenes personally.
28:51Personal phone calls to other CEOs. I've
28:53heard hilarious stories about that, by
28:54the way. Daily updates on the specific
28:56constraint until it's resolved. This is
28:58the large hammer approach, as Karpathy
29:00puts it. The hammer only works because
29:03the signal is clear. If you don't know
29:05exactly where the bottleneck is, you're
29:07just swinging in the dark.
29:09But Elon alone doesn't explain SpaceX.
29:12Plenty of ambitious, technically engaged
29:14founders have failed spectacularly in
29:15aerospace. Founders set direction, but
29:18it takes more than that to sustain a
29:19company. Start with Gwynne Shotwell,
29:21currently the president of SpaceX and
29:23part of the core founding team. She's
29:24the one, quote, holding it all together,
29:25as many believe. But it's a mistake to
29:27view her as merely the steady hand who
29:29keeps the lights on while the engineers
29:31dream. She is the strategic co-architect
29:34of the entire system. I wonder if
29:36Shotwell does any podcasts. If anybody
29:37listening to this happens to know her,
29:39I'd love to have a conversation with her
29:41on my other show, on my new show. So,
29:42back to that. She's a strategic
29:43co-architect of the entire system. In
29:45most other companies, the sales and
29:46business teams are natural enemies of
29:48engineering logic. They promise the
29:49customer whatever they want, which
29:51creates the very bespoke complexity that
29:52the first principles thinking is trying
29:54to delete.
29:55Shotwell did the opposite. She
29:56recognized that for the manufacturing
29:58flywheel to work, the market had to be
29:59forced to adapt to the rocket. She made
30:01it so the world's most conservative
30:02buyers, NASA and the Pentagon, would
30:04accept a radical new model of
30:06standardization. She ensured that every
30:08dollar saved by engineering was
30:09converted into a dominant market
30:11position. She's the reason SpaceX didn't
30:14end up as another very impressive, very
30:16bankrupt space startup.
30:19The combination also worked because of
30:20where it happened, Southern California,
30:22where aerospace culture runs deep. Not
30:24just available talent from declining
30:26programs, but the lineage back to the
30:28early aviation pioneers building flying
30:30machines in hangars. That expertise was
30:33dormant, buried under decades of
30:35bureaucracy. What Elon grafted onto it
30:38were Silicon Valley operating norms,
30:40flat hierarchy, engineer ownership,
30:42permission to walk out of unproductive
30:44meetings. From this foundation, specific
30:47cultural practices emerged. Not values
30:50on a poster, actual behavioral memes
30:53that spread through the organization.
30:55Memes that show up in day-to-day
30:57decisions about what to build, who gets
30:59promoted, and how to respond when
31:00rockets blow up. And real quick, before
31:03we get back into this, I need to tell
31:04you about Vanta. Vanta Vanta Vanta Vanta
31:07helps your company prove your security
31:08so more customers will use your product
31:10or service. You can think of Vanta as
31:11your always-on AI-powered security
31:13expert who scales with you. The more
31:15your business grows, the more complex
31:17your security needs get, and that
31:18complexity can turn into chaos, and
31:20Vanta helps tame that chaos. Vanta
31:23automates compliance, continuously
31:25monitors your controls, and gives you a
31:26single source of truth for compliance
31:28and risk. So, whether you're a
31:29fast-growing startup like Cursor or an
31:31enterprise company like Snowflake, Vanta
31:32fits easily into your existing
31:34workflows. This allows you to keep
31:36growing a company that your customers
31:37can trust. Many companies won't sign
31:39contracts unless you're certified, and
31:41this is causing you to lose out on
31:42sales. That is why the average Vanta
31:44customer reports a 526%
31:47return on investment after becoming a
31:49Vanta customer. Automate your
31:50compliance, security, and trust with
31:52Vanta. Vanta will help you win trust,
31:54close deals, and stay secure faster and
31:56with less effort. Go to
31:57vanta.com/founders
31:59and you will get $1,000 off. That is
32:01vanta.com/founders.
32:03So, Max breaks this down into five
32:05separate memes. The first one, meme
32:07number one, tip of the spear focus.
32:09Always identify and attack the biggest
32:11limiter.
32:13Don't spread effort across secondary
32:15problems. Laser in on the single
32:17constraint that, if removed, would
32:19unlock everything downstream.
32:22That is true at every level. Each SpaceX
32:25site has a single dominating objective
32:27to simplify prioritization.
32:29A NASA manager who visited SpaceX
32:31observed that when a new problem
32:33appears, it looks like a flash mob in
32:35the hallway. When a system-level
32:37bottleneck is identified, it gets
32:39disproportionate resources.
32:41When Starship development was
32:42bottlenecked on Raptor engine
32:44production, that became the company's
32:46focus. Not propellant loading, not heat
32:48shields, not launch infrastructure.
32:51Raptors.
32:52Elon gave it absolute focus, daily
32:55updates, memos to the company, resources
32:57redirected from elsewhere. Once engine
33:00production broke through, attention
33:02shifted to the next constraint.
33:04The limiter always gets the hammer.
33:07Meme number two, push through
33:09roadblocks. A roadblock isn't a reason,
33:10it's a problem statement. You either
33:12clear it or you escalate it until
33:14someone does. Admitting you're blocked
33:16isn't shameful at SpaceX, it's expected.
33:18Hiding a blocker is what gets you in
33:19trouble. As one engineer described it,
33:22solving blockers moved the needle
33:23forward on several projects. The
33:25cultural expectation is honesty about
33:27what's not working and the relentless
33:29effort to fix it. That's another good
33:31line. I got to repeat that. The cultural
33:32expectation is honesty about what's not
33:34working and relentless effort to fix it.
33:37Meme number three, scrappiness.
33:39Cost-sensitive resourcefulness over
33:41bureaucratic process. This goes hand in
33:43hand with the low-cost provider
33:45mentality.
33:46SpaceX's replacement for NASA's heritage
33:49docking system was prototyped with bike
33:50shocks and catalog parts.
33:53Crude prototypes let you test ideas
33:55before committing to expensive
33:56development. The scrappy approach
33:58extends everywhere. Reuse test hardware.
34:01Hacking tools together. Building ground
34:03support equipment from industrial
34:05components instead of aerospace grade
34:06systems. Small teams build end-to-end
34:09instead of handing off between
34:11specialized groups.
34:13Engineers are expected to design, build,
34:15and test what they own. Elon calls the
34:17alternative ivory tower engineering.
34:20Design something, throw it over the
34:22wall, and let someone else figure out
34:23how to actually make it. At SpaceX, the
34:25person who drew the bracket is the
34:27person who welds it. Meme number four,
34:30question requirements. Every constraint,
34:33customer, regulatory, internal, is
34:35treated as a hypothesis to interrogate,
34:38not a fact to accept. This is the
34:40embodiment of first principle thinking.
34:43Here's an example. Falcon 9's grid fins
34:46were originally designed to fold like
34:47traditional aerospace grid fins. The
34:50folding mechanism reduced drag during
34:52ascent, which seemed obviously
34:53necessary. SpaceX questioned whether it
34:56was worth the mass and complexity.
34:58Simulation showed fixed fins were
35:00acceptable. So, they deleted the
35:02mechanism entirely. This is what Elon
35:04said, "The best part is no part. The
35:07best process is no process."
35:10Engineers are explicitly told that
35:12requirements from {quote} "smart people"
35:14are the most dangerous because nobody
35:16thinks to question them. Every
35:17requirement must have an owner, a
35:19specific person who can defend why it
35:21exists. If the owner can't explain it or
35:23the original reason no longer applies,
35:25that requirement gets deleted. This
35:27turns into Elon's now well-known rule,
35:30"If you are not adding back at least 10%
35:32of the requirements you deleted, you
35:34aren't deleting enough."
35:37And meme number five, treat everything
35:39as learning. Failures and explosions are
35:42data for the next iteration, not
35:43disasters to be concealed.
35:46SpaceX published compilation videos
35:48titled how not to land an orbital
35:49rocket, spectacular drone ship crashes
35:53set to music. This isn't just PR, it's a
35:55genuine signal that visible failure is
35:58acceptable if you extract the lesson.
36:00Many would see the early Falcon 9
36:02booster landings as spectacular
36:04failures, rockets exploding on drone
36:06ships, tipping over, crashing into the
36:08ocean. But over time, those iterations
36:10produced a landing success rate high
36:12enough to support reusing boosters
36:14dozens of times. That reusability is
36:17what makes Falcon 9 economically
36:19dominant.
36:20You don't get there without the
36:22explosions first. The cultural frame
36:25matters. A failed test is only bad if
36:28you didn't learn enough from it. These
36:30memes reinforce each other and they
36:32reinforce the strategy. Vision attracts
36:35people who thrive in this culture. The
36:37culture then selects for more of the
36:38same. High performers stay, others
36:41self-select out. Small elite teams
36:43maintained by default. Low performers
36:45actively removed rather than
36:47accumulated. This is the real moat.
36:50SpaceX's cost advantage can
36:52theoretically be matched. Their
36:53technical innovations can be studied and
36:55replicated, but the culture requires
36:57rebuilding an organization from scratch.
37:00Are you seeing the pattern yet? The
37:02entire system reinforces itself,
37:04spinning the flywheel faster and
37:06becoming increasingly hard to copy. It's
37:08loops all the way down.
37:11And so that brings us to the last
37:12subheading, feedback loops. So what is
37:14actually hard to copy? Strategy
37:16identified the waste. 98% of every
37:18dollar going down to process instead of
37:20atoms. Engineering found the path.
37:22Iterate fast and validate using reality
37:24instead of thinking your way to perfect
37:25solutions. Culture made it move.
37:28Question requirements. Fail visibly.
37:30Attack the tip of the spear. Three
37:33systems mutually reinforcing.
37:35Each turn of the flywheel makes the next
37:37one easier. The answer to what's hard to
37:39copy isn't any single tactic. It's that
37:42the tactics are a system.
37:44Copy one without the others and it
37:46breaks down.
37:47First principles design without vertical
37:49integration gives you targets you can't
37:51reach. Vertical integration without
37:53volume makes your fixed costs a
37:55liability. A fail fast culture without
37:57people who can tolerate visible failures
37:59becomes theater.
38:01The output isn't just cheaper rockets.
38:04The real output is a generation trained
38:07to build hard things.
38:10Engineers who internalize these memes,
38:12question requirements, fail fast, tip of
38:15the spear, are now scattered across the
38:17frontier. Space startups, defense tech,
38:20manufacturing automation, energy. The
38:22cultural memes are spreading. The
38:25lessons aren't trapped inside one
38:27company. This matters for anyone trying
38:29to build something hard. Working at
38:32frontiers creates optionality invisible
38:34from the ground.
38:35Starlink wasn't in the original vision.
38:38It emerged because SpaceX was already
38:40there launching so frequently and
38:41cheaply that a 9,000 satellite
38:44constellation became feasible.
38:47Others couldn't see that opportunity
38:49because no one else was in a position to
38:51take it.
38:52When you're at the frontier,
38:54possibilities present themselves that
38:56don't exist for anyone else.
38:59Small groups with the right structure
39:00can do extraordinary things. Not
39:02brilliant individuals, structured teams
39:05with fast feedback loops, real forcing
39:07functions, and cultural tolerance for
39:10visible failure.
39:11The P-80 in 5 months, the SR-71 in 4
39:15years, Falcon 1 to Falcon 9 in 4 years.
39:18This has happened before, it can happen
39:21again.
39:22The lesson isn't be like Elon. Hero
39:24worship is a wrong takeaway. One person
39:26didn't build Starship. The lesson is
39:28that structure matters more than the
39:30hero. Get the system right and the
39:33results follow.
39:34If I had to distill to one question, it
39:36would be, how fast are your feedback
39:38loops? How fast can you get to reality?
39:41You can see this pattern in every
39:43successful frontier tech effort. The
39:46common thread is treating reality as the
39:48teacher and getting to class as often as
39:50possible.
39:51And then he describes what the book is
39:52going to be. That's the analysis, but
39:54analysis is reconstruction, pattern
39:56matching after the fact, shaped by
39:58hindsight. It's useful, but it's not the
40:00same as watching the system get built,
40:02which is what he's doing for the book.
40:04What follows is the raw material.
40:07The SpaceX company updates, over 100
40:10dispatches from 2003 to 2013. The first
40:135 years written primarily by Elon.
40:16Technical challenges explained in real
40:18time. Near-death moments. Incremental
40:21victories that didn't seem incremental
40:23at the time.
40:24The culture and memes showing up in the
40:26language before anyone named them.
40:29It starts in May 2002.
40:32Elon hires Tom Mueller, a propulsion
40:34engineer building rocket engines in his
40:36garage on nights and weekends.
40:38They incorporate a company called Space
40:41Exploration Technologies.
40:43And they begin.
40:45There's a link down below to read the
40:47entire essay. Make sure you add your
40:49email so you're notified when the book
40:50is available.

About this transcript

This page contains the full transcript of How SpaceX Works by Founders Podcast, generated from the public captions YouTube serves with the video. The transcript has 6,687 words across 1,162 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.