Ep 8: Elon Musk: An Operator's Algorithm, Tesla, SpaceX

Episode Description 

Everyone's knows about Elon Musk as a visionary, but few know him as an operator. This episode is not a profile about Mars,  "first principles" as a personality type, or the ten-companies-one-genius arc. This is an operator's autopsy: how Musk actually ran two of the most consequential manufacturing systems of the last twenty years, what he built, what he broke expensively and publicly, and the five-step doctrine that emerged from the wreckage. The Model 3 production collapse: first-pass yield as low as 14%, robots pulled out, a tent assembly line improvised in a Fremont parking lot and the five-step improvement algorithm it produced. The Gigafactory network as a speed and trade-exposure strategy: a 168-working-day Shanghai build that closed a 55% tariff cost disadvantage before competitors could respond. SpaceX's iterative manufacturing doctrine applied to Falcon 9 and Starship, from Raptor 3's part-count reduction and cost trajectory, to nine-day booster turnarounds, to the four-hundredth drone-ship landing. The logistics layer nobody profiles: the Tesla Semi solving an internal freight bill on a 260-mile route that Musk originally floated running through a hyperloop tunnel, the maritime recovery fleet designed to operate like an airport, and a satellite factory producing seventy units a week paced deliberately against the launch cadence built to absorb them. And the part most profiles skip: what all of this actually cost. In capital. In regulatory friction. In very public mistakes that survivorship bias has turned into charming anecdotes. One codified doctrine. Six transferable principles. One episode about the most over-mythologized executive of the past twenty years, without the mythology.

SHOW NOTES

Key Concepts

Gigacasting — High-pressure aluminum die casting replacing ~70 discrete stamped and welded underbody parts with a single casting. ~40% cost reduction on the rear underbody section. ~600 robots eliminated on the Model 3 body line. Validated rapidly using 3D-printed sand binder-jet prototype tooling before committing to metal dies.

Build-Fly-Fix-Repeat — SpaceX's Starship development methodology. Physical iteration funded at a scale that treats destroyed test articles as a line item. Requires the capitalization to absorb repeated full-asset losses as planned program cost, not crisis.

Asset Utilization as Competitive Moat — Falcon 9's actual advantage over expendable-rocket competitors was not propulsion technology. It was treating the booster as scheduled equipment rather than a disposable artifact — reuse economics applied to orbital hardware.

Targeted Vertical Integration — Bringing in-house specifically the most exposed, least redundant node in the supply chain. Not the most visible, not the easiest to acquire. The Tesla lithium refinery in Corpus Christi is the model: it eliminates a ~20,000-mile intercontinental shipping loop by targeting the refining step that was the actual single point of failure in the chain.

The First-Mover Tax — The R&D burden of proving a new manufacturing category is absorbed by the pioneer; fast followers buy the mature technology at a fraction of the proving cost. A real and underdiscussed cost of manufacturing innovation at the frontier.

THE ALGORITHM — FIVE STEPS 

The improvement sequence codified by Musk from the Model 3 ramp collapse. The order is the entire point.

1. Question every requirement — traceable to a named individual, not a department. The smarter the source, the more scrutiny it deserves. Includes your own prior decisions.
2. Delete — remove the step, part, or process entirely. If you don't re-add at least 10% of what you cut, you didn't cut aggressively enough. Restoration is not failure; it's calibration.
3. Simplify and optimize — only after deletion. Making a process that shouldn't exist more efficient is waste at a higher velocity.
4. Accelerate cycle time — speed up the already-simplified version. Not before.
5. Automate — last, always last. Automation is a reward for a process that has already earned simplicity. Tesla violated this step, at scale, at two separate facilities, before it became doctrine.

SOURCES AND FURTHER READING

All figures cited in the episode are drawn from primary disclosures, authoritative trade press, or cross-corroborated reporting. Key sources below.

• Walter Isaacson, Elon Musk (2023) — source of the five-step algorithm framing, production-hell retrospective, and hands-on management corollary
• SpaceX IPO filing (2026) — Starlink satellite production figures; reported via GeekWire
• Bloomberg / Xinhua News Agency (October 2019) — Shanghai Gigafactory 168-working-day build timeline
• CarNewsChina (December 2025) — Shanghai 4 millionth vehicle milestone and 30-second production cadence
• Space.com (March 2025) — Starlink Bastrop kit factory production figures, sourced from SpaceX video
• Spaceflight Now (August 2025) — 400th drone-ship landing milestone
• New Space Economy (July 2024) — SpaceX offshore recovery fleet overview
• SpaceX.com mission page — Starship Flight Test 12, May 22, 2026
• Electrek / Teslarati (2017–2024) — Tesla Semi internal route, freight cost framing, convoy cost figures
• Procurement Magazine / Supply Chain Digital / Mining Digital (February 2026) — Corpus Christi lithium refinery; figures consistent across all three sources
• Stellarix industry analysis (March 2026) — Giga Casting 2.0 pause; single-source, recommend a confirming pass before citing on-air
• BDC Network (2018) — Shanghai tariff/freight cost disadvantage figure
• NASASpaceFlight.com / TechTimes (2025–2026) — Falcon 9 fleet turnaround statistics

Operational Velocity is for education and general information only and is not investment, financial, legal, or tax advice, and nothing in it is a recommendation to buy or sell any security. The views expressed are the host's own, the company and figures discussed are drawn from public sources believed reliable but not guaranteed, and you should do your own research and consult a qualified professional before making any decision.

Transcript

SPEAKER_01 0:10

Great returns aren't luck. They're built on the floor. Operators and investors always pushing for more from the deal sheets signed to the systems in place. Speed without precision is just running a race. Operational velocity where execution sets the pace.

SPEAKER_00 0:25

Hey, welcome to the Operational Velocity Podcast. Here's something that you probably heard a thousand times. Elon Musk is a visionary. And here's something that you probably haven't heard as often. That Elon spent eight months in 2017 and 2018 sleeping on a couch on the factory floor in Fremont, California, personally rewriting the wiring harness routing on the Model 3 line because the original design required eight miles of cable per car, and he thought that number was insane. And here's something else, and I think this is one of my favorites because nobody puts this in the highlight reel. In 2017, he looked at what it costs to move battery packs 260 miles between the gigafactory in Nevada and the assembly plant in Fremont, called the number gigantic, and seriously floated building an underground hyperloop tunnel through his tunneling company to solve it. Then he didn't build a tunnel, he built a truck. Tesla became its own first customer for the Tesla semi, specifically to solve an internal freight problem nobody outside the company ever heard about, because internal freight problems don't make keynote slides. And that's the show today, not the tunnel and not the truck. This show is called Operational Velocity. And one of the four lenses of this series is the operations first leader. And these are the folks who make money by improving how things run. The operations leaders, the supply chain rebuilders, the PE operating partners, all for the unglamorous mechanics of value creation. And today we're going to talk about Elon Musk, because I think he's over-mythologized over the last 20 years. And we're going to challenge that mythology a little bit, or maybe add some things that you're not so much aware about as Elon Musk, the operator. So no Mars, no first principles as a personality trait, no 10 companies, one guy, generational genius, narrative arc. If you want that show, there are 400 of them already, most paid for one way or another by someone's public relationships budget. What we want is narrower and I'd agree, more useful. How did this person actually run factories? What did he get wrong expensively in public? What's the repeatable operating logic underneath the chaos, the actual doctrine codified that a supply chain leader or operations leader could steal tomorrow? So there's six stops on this map. Tesla's manufacturing system and the codified rule it produced, the gigafactory network as a speed and trade exposure strategy, SpaceX engine and rocket manufacturing, that is structurally the most interesting industrial story in this episode, the physical logistics underneath both companies, the trucks, the ships, the barges, the satellite factories, which nobody profiles because it's boring, and it's not coincidentally where most of the real operating discipline lives. The actual cost of all of it in dollars and regulatory friction, because operators who tell you risk taking is free are lying to you. And finally, what's transferable to the rest of us who don't have a launch pad. So let's get into it. Let's actually start with a failure, because the failure is more instructive than the win. And this will kind of set everything else in this episode. So in July 2017, Tesla launched the Model 3. This was the car the entire company's future is staked on. And he launched this at a stage in Fremont, and Musk hands the first 30 cars to employees. The plan, publicly stated, is to hit 5,000 vehicles a week by the end of that year, 20,000 a month, and ultimately a million vehicle annual run rate within a few years. Musk told the assembled employees directly that the company was heading into what he called production hell. And he wasn't being falsely modest. The Model 3 was being built with roughly 10,000 unique parts, sourced through a global supplier network, and he told that same crowd that almost anything going wrong anywhere in that network, if Tesla hadn't buffered the supply chain, would interrupt the whole line. And that's honestly a good assessment of fragility from a CEO standing in front of his own factory floor, which is rarer than it should be. And by the back half of 2017, Tesla is producing a few hundred Model 3s a week, not 5,000, a few hundred. And here's what was actually happening mechanically. Tesla had tried to rebuild the most automated automotive line in the world. Robots doing nearly everything, including task robots, that are genuinely bad at like handling soft battery modules or threading wiring through tight chassis cavities. First passed yield on the parts line was reportedly as low as 14%. And that means for every part coming off that station, roughly six of seven needed rework before it was usable. And the battery module supply was lagging final assembly. The body design used a mix of steel and aluminum that was created, joining and welding complications nobody had fully solved at volume. And remember that wiring high risk from the cold open, eight miles of cable by Musk's own account, running through every Model 3 body, because the design had never been pressure tested against the question: can a robot or a human actually route this efficiently at 5,000 units a week? Nobody had asked. And that's the failure mode in one sentence. Nobody asked. And it's worth saying, too, that Fremont wasn't even the first time this exact failure mode showed up. The Nevada Gigafactory, built in parallel to support Model 3 battery production, went through its own version of the same crisis on the battery module line, with heavily automated equipment struggling to handle the same kind of soft, irregular components that were jamming the Fremont body line. Two factories on two different products hitting the identical wall at the identical moment because they had been designed by the same underlying assumption. That more automation applied earlier was strictly better, that's not a coincidence. That's a single bad assumption replicated across an entire entire company's plan before anyone had the data to know it was bad. And here's the part most retrospective skip. Tesla's fix was not more automation better executed, it was the opposite. Musk pulled the robots out. He brought in more human labor, he built a literal tent, a fabric structure in the Fremont parking lot as an improvised general assembly line, because the existing building didn't have enough room for the manual processes they now needed. And people, it turned out, were dramatically better than the robots at the specific job of adapting to parts that didn't quite fit, fixtures that weren't quite calibrated, and the thousands of small judgment calls that high mix manufacturing actually requires in its early unstable phase. There's a documented line from Musk on this, paraphrased here rather than quoted verbatim: humans are simply better than robots at handling the unexpected, and a factory in its first 18 months of life is nothing but the unexpected, repeated daily. Now, why does this matter beyond being a good story? Because must turn the failure into a codified rule, and the rule is the most exportable piece of operating doctrine in this entire episode. He calls it somewhat self-mockingly the algorithm. Five steps always in this order, because the order is the entire point, and he's reportedly described himself as having become a broken record, repeating it to its own executives on the theory that repeating something to an annoying degree is the only way it actually changes behavior at scale. So, step one, question every requirement, not generically, specifically. Every requirement should be traceable to a named individual, not a department. Legal requires it or safety requires it is not an answer. Who specifically decided that and why? The logic here is uncomfortable but correct. Requirements that come from credentialed, intelligent people get questioned the least, precisely because nobody wants to be the person who challenges the expert. That's exactly backward. The smarter and more authoritative the source, the more scrutiny the requirement deserves, because an unquestioned smart person's assumption ages into permanent organizational scar tissue. And add the corollary that matters most for operators sitting in this audience, and this applies to your own prior decisions too. The requirement doesn't get a pass just because you're the one who wrote it 18 months ago. Step two, delete. Not simple out, simplify, but delete. Remove the part, the process, the approval step entirely. And there's a built-in calibration mechanism here that I think is genuinely clever. If you don't end up rereading at least 10% of what you cut, you didn't cut aggressively enough. That's a target failure rate for the deletion process itself. Most operators are too conservative about cutting because they're optimized to avoid the embarrassment of having to put something back. Musk's framing inverts that. Step three is simplify and optimize. Only now, after the requirement has survived, scrutiny and the unnecessary parts are gone. You're allowed to make what remains more efficient. Step four, accelerate cycle time. Speed up the process, but only the process that's already been pared down and simplified, because speeding up a process that shouldn't exist just means you're generating waste faster. And step five, automate. It's the last. This is the one Tesla violated catastrophically at both the Nevada Gigafactory and Fremont during the Model 3 RAM. They tried to automate processes before questioning, deleting, or simplifying them, which meant they built expensive, rigid robotic systems to perform tasks that on inspection didn't need to exist in that form at all. And Musk has been fairly candid in retrospective interviews that this was his own most expensive operating mistake, not a vendor failure or a one-off engineering miss, and that he made the same mistake twice, once in Nevada and again in Fremont, before it stuck. There's a set of corollaries that travel alongside the five steps that I think are underrated relative to the headline sequence. One, every technical manager needs hands-on experience in the thing they manage. Must's own stated rule was that software managers should spend a meaningful share of their time actually writing code, and that managers overseeing solar installation should personally have installed roofs. Logic being that a manager who's never touched the work is in his own framing, like a cavalry commander who can't ride a horse. So, whatever you think of the cavalry metaphor, the underlying point holds up under scrutiny. Distance from the work is exactly how a bad requirement survives unchallenged for years. So if you're running an operating company and you only take one thing from this episode, take this. Automation is a reward for a process that's already earned simplicity. It's not a substitute for asking whether the process should exist. Most fail digital transformation projects in portfolio companies, and I'd bet most of you listening have seen at least one, or step five thinking applied to a process that never survived step one. And there's a second related principle, Tesla operationalized out of this period that's worth dwelling on. Design for manufacturability has to be a top-line requirement, not engineering's afterthought. After production hell, Elon explicitly told design teams that ease of manufacturing moves to the top of the requirements list, above, in some documented cases, aesthetic or performance preferences that engineering had previously prioritized. A car that's elegant on a CAD screen and impossible to assemble at 5,000 units a week is not an elegant car. It's a liability with good marketing. And it's worth being honest that even a codified battle-tested doctrine doesn't make every subsequent product launch clean. The Cybertrucks ramp at Giga Texas, years after production hell and years after the algorithm had supposedly been internalized company-wide, still took longer than Tesla's own initial promises suggested it would. And the exotic materials carry their own manufacturability tax that no amount of process deletion fully erases. Stainless steel doesn't stamp, weld, or finish like conventional steel or aluminum. And that's a material science constraint, not a process discipline one. The lesson there isn't that the algorithm failed, it's that the algorithm fixes process problems, not material science problems. And a mature operator should know which kind of problem they're actually staring at before reaching for the five steps as a cure. And one more thread before we move on, because it's quiet proof point that the doctrine actually stuck rather than just being a good story. Tesla tells about itself. If you walk into Giga, Texas today, you'll find the Model Y line, the Cyber Truck Line, the Optimus Robot Assembly, operating inside variations of the same physical facility, sharing tooling philosophy and in places sharing floor space and equipment classes across what are on paper three completely different products. A sedan, adjacent crossover, a steel-bodied pickup, and a bipedal robot. That's only possible if the underlying manufacturing system was built on a modular deletable from the start, rather than bespoke to one vehicle. The algorithm isn't just a crisis recovery tool. If applied early, it's what lets one factory floor host three unlimited products without rebuilding the line from scratch each time. Most coverage of the gigafactories treats them as a scale story. Bigger factory, more batteries, line go up. That's true, but that's the least interesting layer. The interesting layer is that the Gigafactory Network is fundamentally a speed and trade exposure strategy wearing a manufacturing costume. And the clearest proof of that is a single factory in China that most people only remember as a headline number without remembering why the speed itself was the actual point. Here's a setup nobody mentions when they tell the Shanghai story. When Tesla decided to build in China, internal analysis reportedly put the company at a 55 to 60% cost disadvantage on every car sold there, purely from import tariffs and ocean transport relative to a competitor building locally. And that's not a rounding area if you fix with a better procurement team. That's an existential cost gap that makes market structurally unwinnable unless you change where the car is built. So Tesla didn't optimize around the tariff. It built a factory inside the tariff wall, and it built it at a pace explicitly designed to close the gap before the competitors could react. The numbers on that build are worth saying slowly, because they are genuinely abnormal even by Chinese industrial construction standards, which are themselves abnormal by global standards. Site grading began in January 2019. Officially, it took 168 working days, about six months, to go from securing permits to having electricity running through the building, a pace state media in China described as fast, even by domestic benchmarks, achieved by running construction crews around the clock. In some phases, moving first structural pillar to roof, pegging in roughly a single month. The first vehicles came off the line in roughly nine months from the groundbreaking. Deliveries to customers happened within the same calendar year, and compare that to the rule of thumb the Tesla skeptics were using before the project started. 17 months was considered the prior speed record for a comparable industrial facility, and Shanghai just didn't beat that, it roughly cut it in half. So why does speed matter more than size? Because the entire value of the regional manufacturing is time discounted. A factory that closes a 55% cost disadvantage in 18 months has captured a market position competitors can't easily dislodge before they've even broken ground on their own answer. A factory that takes four years to build the same facility has handed that window to ever moves faster. Shanghai is now by itself one of Tesla's most productive single assets in the world. Recently reporting puts it at past four million total vehicles built, with a new car rolling off that line roughly every 30 seconds at full rate. And the plant has at points supplied not just China but European deliveries before Berlin's own capacity came fully online. None of that scale matters if speed to first delivery hadn't happened first. Scale is what you get to keep after you've already won the speed race. The pattern repeats itself across the rest of the network with the same underlying logic, even if no other site matched Shanghai's specific record. Nevada, the original gigafactory for battery cells. And PACs, Berlin, serving the European market under the identical trade exposure logic, produce where you sell, inside the trade zone, inside the regulatory jurisdiction, inside the currency exposure you're already managing. By 2025, reporting had Berlin as the fastest Tesla's growing plant outside of China, even while it absorbed real friction. The U.S. facilities hadn't had to navigate at the same intensity. The German labor unions pushing for stronger collective bargaining than Tesla was used to, and sustained local opposition over the plant's water usage in a region with real water stress concerns. Both of those are genuinely operating costs of localizing in a jurisdiction with a different regulatory and labor culture than Fremont or Austin, and they're worth naming rather than skipping past. Then Austin, the Giga Factory Texas, the leading edge site for Model Y and CyberTruck, and per the last segment, now a genuinely multi-product floor. Now let's go one layer deeper into the casting story, because it's the same speed and deletion logic applied to physical parts instead of geography, and it deserves a level of detail most business coverage skips. The conventional rear underbody of a mass market car is assembled from roughly 70 individual stamped and welded pieces. Tesla's gigacasting process using an enormous high-pressure die casting machine called the GigaPress, originally sourced from the Italian manufacturer IDRA, replaces those 70 parts with a single aluminum casting. One piece, one pour. Tesla reported the resulting cost reduction on that section at roughly 40%. And on the Model 3 body line specifically, the consolidation eliminates something on the order of 600 robots that had previously existed solely to weld and join those 70 discrete pieces together. There's a genuinely clever piece of process engineering buried inside that, easy to miss. Testa validated its dye designs using 3D printed sandbinder jet prototypes rather than conventional metal tooling, which let engineers test and revise the casting geometry at a fraction of the cost and time a traditional tooling iteration would have required. That's the algorithm again, quietly applied to the prototyping process itself rather than the production line. Accelerate cycle time, but only after you simplified what you're iterating on. Elon Musk's own framing of the casting decision from internal commentary that's since become public is blunt and worth repeating because it captures something true about manufacturing physics. The worst job on that line, by his own account, was sealing the gaps between joined pieces, applying adhesive sealant across dozens of seam, any one of which could fail. A single casting has no seams, it has no sealing job, it has no welding station, it has no 70-part inventory and logistics chain feeding that station. You haven't optimized the process, you've deleted it. That's step two of the algorithm, applied here to sheet metal instead of approval workflows. But here's where it gets interesting. The gigacasting strategy hit a real wall, and it's worth dwelling on because it's a genuine operational lesson about the limits of consolidation. Tesla stated ambition was Gigacasting 2.0, collapsing the entire underbody, not just the front and rear sections, but the whole structural floor into one or two pieces, reducing the total part count from roughly 400 to a handful. And that required new presses with dramatically higher clamping force on the order of 16,000 tons or more, at a capital cost running into tens of millions per machine before tooling and facility modification. Reporting through 2025 and into 2026 indicates Tesla actually paused that full single-piece underbody ambition, stepping back to a three-piece design instead. The reason matters operationally. A single massive structural casting, if damaged, in even a minor collision, can mean replacing essentially the entire underbody of the car, repairability and insurance cost problem that a modular three-piece design avoids. Toyota, Hyundai, and others pursuing their own versions of this technology, they call it hypercasting or megacasting, have flagged the identical concern. And there's a competitive footnote here worth adding, because it changes how you should think about who actually captured the value of this innovation. Tesla bore the entire first mover advantage of proving the gigacasting at scale, the failed early die casting designs, the production line redesign, and the supplier qualification process. The FastFollowers got to skip most of that. BYD, for instance, has reportedly installed its own large tonnage die casting equipment on the order of 9,000 tons of clamping force at a fraction of the RD cost Tesla absorbed, pioneering the category, because BYD could simply buy a mature version of the same press technology from the same supplier ecosystem Tesla helped to create. That's a real cost of being the operator who proves a new manufacturing category works. You frequently end up subsidizing your own competitors' learning curve. And it doesn't make the original bet wrong. It does mean you should go into that kind of bet with open eyes about who ultimately captures the long-run margin. That's a consolidation lesson in full. It has an optimum point, not an infinite asymptote. And the company that proves the technology doesn't automatically keep the advantage it created. The last piece of this segment is lithium, which is vertical integration applied with genuine discipline rather than corporate e-commerce. Tesla operates a lithium refinery in Corpus Christi, Texas, roughly a billion-dollar facility that processes raw spodamine ore directly into battery-gauge lithium hydroxide on site. The conventional supply chain for that material involved shipping raw ore, roughly 20,000 miles, extracted in one hemisphere, refined in Asia, and then shipped back to a Western factory for battery assembly. Tesla's refinery collapses that round trip into a single domestic processing step. Tesla didn't try to mine lithium. It targeted the refining step specifically, because that was the actual single point of failure in the chain. The narrowest, least redundant link, not the most visible or easiest one to acquire. Start with a contrast that frames everything else in this segment. NASA's Space Launch System versus SpaceX Starship. SLS followed the traditional aerospace development model, extensive paper analysis, simulation, and review cycles attempting to anticipate every failure mode before a single full-stack test across more than a decade of development before its first flight. Starship inverted that model entirely. Built a prototype, fly it, watch it precisely, what breaks, fix the specific thing, build the next one, fly again. SpaceX's own internal language for this is buy, fly, fix, repeat, and the public-facing term, rapid, unscheduled disassembly, or rud. It's deliberately gallowing humor functioning as institutional permission. If a test article exploding is in an organizational crisis, your engineers will report bad news faster. And bad news reported faster is the entire value of the iteration loop. It's also worth noting that the physical layout choice underneath this, SpaceX's engineering offices sit adjacent to the manufacturing floor, and Must's own desk has reportedly been positioned directly on the production floor at the points in the program, specifically so engineers see the failures firsthand and can act on them immediately rather than through a meeting three weeks later. Work inside the program is organized in roughly two-week sprint cycles, borrowed almost directly from software development practice applied to people who weld stainless steel for a living. The doctrine didn't start with Starship's orderable flights. It goes back to the very first hop tests. A stubby test vehicle named Starhopper completed low-altitude hop flights back in 2019, providing the basic Raptor engine and landing leg concept before anyone attempted a full-scale vehicle. And the early full height prototypes through 2020 failed during testing with some regularity, but that's a footnote, and that's the design intent. By May 2021, prototype SN15 achieved the program's first successful high-altitude flight and landing, closing out years of accumulated lessons from the prototypes that came before it and didn't survive. That's roughly two years from the stubby test article hopping a few hundred feet to a full-scale prototype flying to altitude and landing intact. A timeline traditional aerospace development simply doesn't produce because traditional aerospace development doesn't budget for destroying a dozen prototypes along the way as a planned cost of the program. The flight history of a full orbital class vehicle makes the doctrine visible in even sharper relief. Integrated Flight Test 1 in April 2023, multiple Raptor engines failed during the ascent, and the vehicle was destroyed by its own flight termination system around the four-minute mark, and the launch pad itself suffered serious damage. Integrated Flight Test 2 that November, the vehicle achieved a first successful hot stage separation while still losing the booster during its return maneuver. And through 2024 and into 2025, successive flights validated engine relights in space, payload deployment, and partial reuse milestones, each flight closing out a specific named unknown from the prior one, with the ninth flight test in May 2025 demonstrating the first reuse of a super heavy booster, even as the upper stage broke up during its return. Progress and partial failure occupying the same exact flight, which is precisely what an honest iteration loop looks like in practice rather than in a highlight reel. And as of this recording, the most current data point is genuine and instructive about what operational risk taking looks like in practice, not in retrospect. In late May 2026, SpaceX flew test flight 12, the first flight of the upgraded Starship and Super Heavy V3 vehicles and the new Raptor 3 engine, launching from a newly built second pad at Starbase. All 33 Raptor III engines on the booster ignited successfully on the ascent, itself a non-trivial feat of synchronized engineering, but a single engine shut down mid-ascent, and the booster's return maneuver only partially completed before an early shutdown, which triggered an FAA anomaly review. That's not a footnote included to be balanced for the sake of balance. That's the actual current state of an operating doctrine that treats partial failure as the expected output of pushing a new vehicle generation and treats the FAA review that follows as a known priced-in cost of operating at that edge, not a crisis communications emergency. And here's the part that ties this back to operations rather than a spectacle. Space has reportedly invested more than $7.5 billion in Starbase and the broader Starship Program. This methodology is not cheap improvisation. It's a capital-intensive experimentation deliberately funded at a scale that allows a company to absorb the cost of a destroyed test article as a line item rather than a setback. The willingness to blow things up is not a free-spirit risk tolerance. It's a finance bet that the cost of physical iteration is lower over the relevant time horizon, and that the cost of years of paper analysis trying to simulate away every unknown in advance. And now the engine, because this is where the manufacturing story gets genuinely impressive, separate from the flight drama. The Raptor engine, the methane-fueled engine powering, both Starship's upper stage and super heavy booster, has gone through three major design generations. And the trajectory is a clean case study in the discipline cost reduction through design simplification rather than just supplier negotiation. The simplification habit actually predates the named generations. As early as 2016, during subscale development, roughly 40% of the engines parked by mass were already being produced through metal 3D printing rather than conventional machining, because additive manufacturing let engineers build injector heads and turbo pump flows pads with complex internal geometry that would have been difficult or impossible to machine conventionally. Geometry that reduces weight and improves performance precisely because it doesn't have to compromise for a traditional manufacturing process. SpaceX own reported figures on the generational jump, Raptor III relative to Raptor 1, runs at roughly double the thrust at roughly a quarter of the production cost, while shedding more than 2,000 pounds of weight per engine. The mechanism behind that isn't a single breakthrough. It's the accumulation of part count reduction, continuing the additive manufacturing habit from years earlier. Raptor 3's design integrates components that were previously separate weld-seamed assemblies, the turbo pump pausing, for instance, into a 3D print structure, eliminat both the weld operation and the leak path that welds seams represent. The engine has reportedly shed nearly 30% of its total parts count relative to its predecessor through its consolidation. And while the McGregor Texas test and the production facility, which runs its own X-ray inspections, pressure tests, and performance evaluations on every engine before its ships, scaled toward output approaching a Raptor engine roughly every day at the prior generation, with a target capacity of around 300 engines a year for Raptor 3 to support an eventual 100 launch per year cadence for the full Starship program. If that pattern sounds familiar, it should. It's gigacasting's logic applied to a rocket engine instead of a car chassis. Fewer parts, fewer joints, fewer failure points, lower cost, faster build. The algorithm doesn't care whether the substrate is aluminum body panels or a combustion chamber rated for orbital flight. The last piece is arguably the most commercially consequential, because this is the manufacturing story that's actually finished and generating revenue today rather than still mid-iteration. Falcon 9 reuse economics. SpaceX's fastest documented booster turnaround. The same first stage rocket flying twice was nine days in March 2025, flying a NASA science mission and then a national security payload less than a week and a half apart. That's the headline number. The more operationally meaningful number is the average, roughly 40 to 50 days turnaround across the active fleet, a figure that's been remarkably stable for several years, meaning SpaceX needs a standing fleet of roughly 20 flight-ready boosters to sustain its current launch cadence of a flight every two to three days. One booster has now flown nearly 30 times, well past the 25 flight mark, at which SpaceX considered a Block 5 booster, fully depreciated on its books. Meaning the marginal economics of every flight past that point are structurally almost pure margin against refurbishment cost alone. That's the actual payoff of treating a rocket the way an airline treats an aircraft instead of the way the rest of the launch industry has historically treated a rocket as a one-time use artifact you build, fly once, and discard into the ocean. The entire industry's launch cost assumptions for a half a century were built on expendability. SpaceX manufacturing and operations strategy didn't out-engineer the competition on rocket science, it out-engineered them on asset utilization. The single most boring operational metric in this entire episode applied to the least boring asset class imaginable. This is a segment I almost left out, and it's the one I'd argue is the actual heart of this episode, because it's where the operating doctrine gets the least credit and does the most work. Start back to where we open, the Tesla semi. In 2017, Tesla's own VP of truck programs said it plainly in a conference presentation in the Netherlands. Tesla would be its own first customer for the electric semi, specifically to move battery packs and drivetrains, the roughly 260 miles between Gigafactory Nevada, where the batteries are made, and the Fremont Assembly plant, where the cars are built. Elon Musk had already called the existing trucking cost on the route gigantic and had floated semi-seriously of building an underground freight tunnel through the Boring Company to solve it. What actually shipped years later was the boring answer, literally and figuratively, built into the company's own production schedule. And by 2024, Tesla was posting footage of a daily semi-fleet running the exact same route, carrying battery packs from Nevada to Fremont as standard operating procedure, with the route itself doubling as a real-world proving ground. The trucks crossed the roughly 7,000-foot elevation of Donner Pass on the way, which means every single delivery run also functions as a continuous stress test of the truck's regenerative braking system on sustained mountain grades long before Tesla ever asked a paying customer to bet on that same hardware. Tesla simultaneously stood up a dedicated semi-production line in Nevada designed for five trucks a week as the initial rate before scaling further in Texas. Why tell this story at its length? Because it's perfect. Undramatic case study in the algorithm applied to a problem that had nothing to do with cars or rockets. It was a freight cost line item. Question the requirement. Does this freight have to move by conventional diesel trucking, paying whatever third-party carrier charges indefinitely? Delete. Don't build elaborate new infrastructure if you don't have to. And a tunnel, however appealing as a thought experiment, is exactly that kind of elaborate new infrastructure. Simply, the actual fix was to build a truck you were already planning to sell and use it on your own highest volume internal route first. Which let Tesla validate the vehicle's real-world economics. Tesla's own published figures at unveiling a claimed cost at roughly 85 cents per mile in a three-truck convoy configuration, using its own freight as a test bed before ever asking a customer to bet on it. That's a company eating its own dog food in the least metaphorical sense possible. And it's the kind of decision that never makes a highlight real because there's no explosion and no record-setting headline. It's just a truck doing a route every day that use costs more. Now scale that same logic up to orbital recovery, because SpaceX runs what is functionally a maritime logistics operation that almost nobody covering the company describes in those terms. Recovering a Falcon 9 first stage at sea requires an autonomous spaceport drone ship, a modified ocean barge, roughly 300 feet long, holding station with an extraordinary precision using its own space station, keeping engines with no crew aboard during the actual. Landing. SpaceX operates a fleet of these, three at Last Count, two on the East Coast, and one on the West, with names lifted from science fiction novels that can tell you something about the engineering culture. Of course, I still love you, and just read the instructions. Both nods to sentient starship names from Ian M. Banks culture novels, plus a shortfall of Gravitas. Alongside the drone ships, dedicated recovery ships fish the payload fairing halves, each one valued at roughly $3 million, intact out of the ocean, rather than let them sink as scrap. As of August 2025, SpaceX has completed its 400th successful drone ship landing of an orbital class rocket. SpaceX engineers have explicitly described the goal of this fleet internally as building towards an airport-like recovery operation, meaning booster lands, gets towed back to the port, gets refurbished, and flies again on a cadence as routine and unremarkable as a commercial jet turning around at a gate. That phrase is the whole doctrine in five words. The rocket isn't a missile anymore. It's scheduled equipment. And the next chapter of that story is already underway. As Starship moves towards full reusability, SpaceX is working through the much harder version of the same problem. A Starship scale drone ship needs to be dramatically larger and structurally reinforced relative to anything used for the Falcon 9, and SpaceX has reportedly begun to secure international permissions near the Bahamas and in talks elsewhere simply to operate recovery vessels in waters far enough offshore to support the vehicle's flight profile. That's logistics as foreign policy, which is not a sentence I expected to write for this episode, and it's a real constraint that doesn't show up in any of the flashy engine specifications. And there's also a third leg to this segment that ties Tesla and SpaceX together through a piece of infrastructure most people have literally never thought of, and that's the barge. SpaceX moves completed starship and super heavy hardware over land and by sea from the starbase production site in Texas to the Kennedy Space Center in Florida, using a modified barge specifically because the vehicles are too large to move effectively any other way. And SpaceX has been staging this infrastructure at the Florida site ahead of the FAA's full launch certification there, explicitly betting that physical readiness should outrun the regulatory timeline rather than wait for it. Reporting from earlier this year described a production pipeline with multiple Starship vehicles numbered in the 40s, already under construction, queued for missions that, as of this recording, haven't even been publicly announced yet. That's a logistics posture, not just a manufacturing one. Build the inventory, stage the transport, and let the launch schedule catch up to the hardware rather than the other way around. And the last stop for this is the one I think most listeners will find genuinely surprising: Starlink, which is usually covered as a satellite internet story. And it's actually one of the cleanest consumer electronics manufacturing stories in either the company's portfolio, and it's also the segment that proves manufacturing capacity is worthless without matching logistics capacity to deploy it. The Redmond Washington facility that builds the satellites themselves was at first publicly disclosed point, producing roughly 70 satellites a week, north of 3,600 a year, at that run rate, according to SpaceX figures, disclosing in its own IPO filing. That production rate only matters because Falcon 9's launch cadence and reuse economics from the last segment exists to absorb it. Satellites have historically launched in batches of roughly 60 at a time on a single Falcon 9, with newer, larger satellite generations launching in batches of 40 to 60 aboard Starship once that vehicle is flying operationally. And built 70 satellites a week with nowhere to launch them, and you built an expensive warehouse. Build them at a rate alongside a launch casement of one Falcon 9 flight every two to three days, and you built the fastest satellite constellation deployment in history. And that manufacturing number and the logistics number only make sense read together. Separately, the ground equipment, the actual dish and router kit that shows up at the customer's door, is built at a facility outside Austin that per SpaceX owned public statements has gone from standing start to producing 15,000 kits a day, more than 70,000 kits a week in under two years, which is explicitly working to in-source more of its component manufacturing so the company can take raw plastic pellets and raw aluminum in one door and ship a finished box kit out of the other entirely within its own walls. The phased array antenna inside that kit is its own vertical integration story. By designing custom chips in-house and automating the antenna's assembly, SpaceX reportedly took a piece of hardware that cost tens of thousands of dollars in its original military grade form down to a retail cost of under $500. Wow. That collapse in unit cost is the entire reason a customer can put a dish on the roof for a few hundred dollars instead of tens of thousands. And it happened because the company treated an antenna the way it treats a rocket engine. Question every part, delete what is earning its place, and then scale the line. Put all these four stories next to each other. The semi solving an internal freight bill, the drone ship fleet built to feel like an airport, the barge moving rockets over land because no other transport makes under scale, and then the Redmond and Bastro factories quietly mass-producing satellites and antennas like Consumer Electronics pace precisely against the launch cadence that has to absorb them. And what you get is the actual through line of this episode. The visionary story is always about the destination. The operator story is always about what's actually moving, on what schedule, and what cost per unit, and who's paying for the truck. I want to spend some real time here because most profiles of operators who take big swings skip the line item for what the swing actually cost. And that omission is what turns a case study into a fan letter. The Model 3 production hell period nearly took Tesla to the edge of insolvency. And that's not hyperbole. It's a documented framing from inside the company at the time, aggressive automation that didn't survive contact with reality, cost months of output, consumed enormous rework labor against a first past yield in the double digits at its worst, and required an emergency improvised fix in a tent in a parking lot. And that is not, by any normal definition, a planned operating outcome. If you ran a portfolio company and your operating partner described, we built an unplanned tent assembly line because the real factory couldn't make the rework numbers work. That's a board-level crisis conversation, not a charming antidote. It only reads as charming in retrospect because the company survived it and the stock went up afterward. Survivorship bias is doing a lot of narrative work in most of Elon Musk's coverage, and this episode is trying to not let it. You see, the Starship program's cost of iteration is the same dynamic at a greater scale. Multiple full vehicle loss, a damaged launch pad early in the program, an FAA anomaly investigation as recently as a flight test in May of this year, 2026, and against a program that reportedly absorbed more than $7.5 billion at Starbase alone. The buy-fly fix-repeat doctrine works. And the Falcon 9 economics prove it eventually works extremely well. But it works because SpaceX has been capitalized at scale that very few operating companies and frankly very few PE black platforms ever will be. This is not a methodology you import into a mid-market manufacturing rollup without first asking whether you can actually fund six figures worth of destroying prototypes as a normal. Expected repeated cost of business, because most can't. And that's not a criticism of the doctrine. It's a constraint on who gets to use it. There's also a first mover cost worth restating from the segment two, because it's the cost that's easiest to forget when you're impressed by a manufacturing innovation in the moment. Tesla absorbed the entire RD burden of approving the gigacasting at an industrial scale. And that includes every failed early dye design and every supplier qualification cycle. And competitors are now buying mature versions of the same technology at a fraction of that cost. And that was due to Tesla's learning curve. And pioneering a manufacturing category is not free even when it works. It's a subsidy you pay the rest of the industry, priced in years before anyone else figured out the same thing was possible. There's also the regulatory dimension, and it's worth naming directly rather than euphemistically. Operating at this edge means living inside FAA review cycles, environmental permitting flights around the starbase, international maritime permissions for recovery vessels, and on Tesla's side, labor relations friction in Germany, and sustained local opposition over water usage at Giga Berlin, and that the company had not previously needed to navigate in a more union-light, less water-constrained U.S. facility. None of that is disqualifying. All of it is a real cost, paid in time, in regulatory relationship capital, and occasionally in a production delay. An operator evaluating this doctrine for their own organization should price all of it before deciding to build fast, break things, and fix in public model is the one to copy. That's an honest synthesis, and I think this is the fairest thing to say about both companies. This is not a story about a person who is reckless and got lucky. It's a story about a person who correctly identified that in two very different industries, automotive manufacturing and orbital launch, the prevailing operating doctrine of the incumbents had become risk-averse to the point of structural stagnation. And who built organizations explicitly capitalized and culturally wired to absorb a high failure rate in exchange for a faster iteration loop. And that bet has paid off enormously at SpaceX, where Falcon 9 reuse economics are now simply the industry's reality that every competitor is forced to react to. It has paid off unevenly at Tesla, where the casting strategy and the Shanghai speed record are genuine and durable manufacturing advances, and the full automation instinct at Fremont and Nevada was an expensive public mistake that the company had to publicly walk back twice before it learned the lesson. Good operating doctrine survives contact with its own failures, gets sharper. That's the actual standard here. Not did this person never fail, but did the system get better specifically because of the failure? In a way you can point to and name. On that standard, both companies pass, not flawlessly, but specifically. So let's start to bring this home. If you're an operations leader and you're trying to extract something usable from this episode, here's the doctrine compressed into six points. Number one, sequence your improvement effort correctly. Question the requirement, delete what's not necessary, simplify what's left, accelerate it, and then and only then automate. Because most operating improvement programs that I see, they go straight to the automation or to a software layer because it's the most fundable, the most demoable step. It's also per this entire episode's evidence the step most likely to waste capital if you haven't done the first three. Audit your last technology investment against this sequence. I'd bet a meaningful share of you skip straight to step five. And pressure test it against the hands-on corollary too. If the manager who approved that automation spent has never personally done the job being automated, that's worth flagging before the next funding round, not after. Step number two. Speed itself is emote, but it's just an outcome. Shanghai didn't just build a factory, it closed a structural cost disadvantage before competitors could organize a response, and the 168-day build time was the entire part of the project, not a side benefit. Ask honestly, whether your own roll-up or platform treats time to first delivery as a strategic variable with its own dedicated budget, or as a number that falls out passively once everything else is decided. Number three, vertical integration is a targeting exercise, not a philosophy. Tesla didn't try to own everything, it targeted the single, most exposed, least redundant node in its supply chain, lithium refining, and bought that specific link in-house. Before you vertically integrate anything into your own company, ask the Tesla question. Which single link, if it broke tomorrow, would take down the whole chain? And is that thing you're actually bringing in-house, or are you integrating the part that's easiest to acquire instead of the part that's actually dangerous? Step four, consolidation has an optimum, not an infinite asymptote. And the company that proves a manufacturing innovation doesn't automatically keep the advantage it created. Gigacasting went from genius to overreach the moment Tesla pushed it past the point where deletion started concentrating risk somewhere new. In this case, Aiden Collision Repair Economics. And Tesla is now watching fast followers buy mature versions of the same press technology at a fraction of the RD course it absorbed pioneering the category. Every let's eliminate the handoffs initiative in your own operations has an equivalent cliff edge. And every category you pioneer has a fast follower waiting on the other side of it. Find the first before your customer does, and go into the second with your eyes open about who eventually captures that margin. Step number five: the unglamorous logistics layer is where the doctrine actually proves itself. Nobody profiles the Tesla Semir route or the drone ship fleet because there's no explosion and no keynote slide. But that's exactly where question every requirement, then delete gets applied with the least ego and the cleanest result. A freight bill solved with a truck, a recovery operation that quietly became airport-like, one landing at a time, a satellite factory whose output is worthless unless you've matched it deliberately to a launch cadence built to absorb it. If your own operational improvements only ever show up in the flashy parts of the business, you're probably leaving the cheapest wins on the table in the boring parts, and probably failing to check whether your production capacity and your logistics capacity are actually paced to each other at all. And last but not least, number six, and this is the one I'd like to actually put on a wall. The best operators in this story didn't get the algorithm right the first time. Elon Musk has been candid and on the record that he personally over-automated Fremont and Nevada twice before he'd earned the right to automate anything by his own five-step logic. The doctrine isn't valuable because an author never violates. It's valuable because it's specific enough that you can catch yourself violating it in real time and correct, which is precisely what happened when Tesla tore robots back out of the Model 3 line and put humans back in. The system isn't be right. The system is have a specific sequence enough that being wrong is diagnosable. So that's the episode. Not a visionary, an operator who burned a lot of capital finding out, in public where the sequence breaks, and who wrote the sequence down carefully, enough that the rest of us don't have to burn ours finding out the same way. If you got something useful from this episode, then please subscribe and send it along to an operator that may need to hear it. Until next time, this is Gotham Basso. Take care.