The Factory Is the Supply Chain: Inside SpaceX’s One-Starship-a-Day Starfactory

Part IThe claim, stated precisely

The Space Race’s new tour inside Starbase’s production complex documents a factory designed around a single output target: one Starship per day. At over one million square feet, Starfactory is the biggest rocket manufacturing plant ever built, and the claimed ceiling — up to 365 vehicles a year from this one facility — is not a cadence any launch program in history has approached. For calibration: the video’s own contrast is NASA’s Space Launch System, which has flown twice in four years. The gap between those two numbers — roughly a factor of seven hundred — is the subject of this brief.

Production begins with raw material that looks nothing like aerospace: coils of stainless steel, about 4 mm thick and 1.8 m wide, in a custom blend SpaceX developed for the specific demands of a reusable vehicle. The alloy is engineered to hold strength from cryogenic tanking through reentry heating — and it is actually strongest cold, precisely when the rocket carries its heaviest structural load. It resists corrosion through years parked in salt air on the Gulf coast, and it remains easy to bend and weld at production speed. An automated line cuts sheets from the coil and welds them into cylinder rings; rings are stacked in threes and fives and fused by a robot arm that completes each circumferential weld in one continuous pass. Starships were once hand-welded top to bottom; as the plant moves to mass production, human welders now hold only the small, precision-critical attachments. Curved geometry — nose cone panels, tank domes — is hydroformed: liquid pressure exceeding 36,000 psi, roughly twice the pressure at the bottom of the ocean, pushing flat steel into molds.

Part IILinear adjacent flow: the unboxed rocket

The structural insight of Starfactory is what it deletes: the conveyor. The single-line architecture popularized by Ford a century ago — one belt, sequential stations, parts accreting until a finished product rolls off the end — remains the default template for most manufacturing. Tesla’s alternative, the “unboxed” method, breaks the vehicle into chunks built simultaneously at parallel stations and merged in a final assembly step. SpaceX does not use the word “unboxed” for rockets; Musk’s term is “linear adjacent flow,” and it describes the same topology: many production flows running side by side under one roof, converging late.

The mechanism that makes parallel flow work at 9-meter scale is overhead. The production floor is framed by a gridwork of steel beams that is not holding the roof up — it is a track network for cranes that lift entire rocket sections and carry them between stations. On the floor, rotating stands spin the 9 m rings so crews stand still while the work turns past them. The building itself encodes the flow: a staggered roofline, low at the end where raw coil enters, tall at the end where assembled sections exit.

A ship decomposes into three chunks. The top: nose cone plus payload bay. The middle: propellant tanks. The bottom: the engine section. (The Super Heavy booster is simpler up top — an interstage ring — and far more complex below, with 33 engines and their plumbing.) Around each big chunk, smaller parallel builds run concurrently: hydroformed header tanks that hold landing-burn propellant, installed up in the nose tip while the cone is still open and the plumbing is still reachable — one of the quiet advantages of building in chunks; wing flaps assembled from stainless skins welded over die-cast beams with three hinge points; and the Starlink dispenser — a rack SpaceX nicknamed the PEZ dispenser — that holds a couple dozen satellites in a vertical stack and feeds them out one at a time through a slot door cut in the payload bay. Below that slot, two round ports: docking hardware for on-orbit propellant transfer, the enabling primitive for every lunar and Mars mission profile on the manifest.

Thermal protection runs as its own automated flow. Robots spot-weld anchor pins across every protected surface; crews lay a black backup heat-shield sheet, then a white insulating foam layer that fills inter-tile gaps, then click hexagonal ceramic tiles onto the pins — three holes, three pins, lock. The tiles come from an in-house production line SpaceX calls the bakery — a process closer to glazed floor tile than to aerospace tradition — currently firing 1,000+ tiles per day and scaling toward 7,000.

Part IIIThe contrast: a rocket built by a supply chain

The video’s comparison case is the one OED scores as a contractor graph. SLS is not built in a factory; it is built in several, spread across the United States — California, Utah, Louisiana — with additional Artemis spacecraft hardware manufactured in Europe. Segments travel by barge, truck, and cargo aircraft to NASA’s Vehicle Assembly Building at the Cape, where the rocket is stacked from the ground up. Each node in that network is a separately contracted, separately scheduled, separately profitable enterprise: Boeing builds the core stage at Michoud, Northrop Grumman casts the solid boosters in Utah, Aerojet Rocketdyne (now L3Harris) supplies the RS-25 engines, Lockheed Martin builds the Orion capsule, and Airbus Defence and Space delivers Orion’s European Service Module. The transport layer between them is measured in weeks; the coordination layer is measured in years. Two launches, four years.

The convergence point on SpaceX’s side is the Gigabay: nearly Starfactory’s footprint with the roof raised to 116 meters — about 46 million cubic feet of enclosed volume. The explicit ancestor is the VAB, designed in the 1960s to hold four Saturn Vs simultaneously, which enabled four crewed Apollo missions inside a single year. The Gigabay holds 24 ship-and-booster work stations — enough for 12 fully stacked Starship/Super Heavy vehicles at once. Final integration happens here: the three chunks joined, propellant and pressurant plumbing connected, power wired nose to tail, final tiles set, checkout run.

One leg of the process remains geographically distributed, and it is the exception that proves the design rule: engines. Raptors are still built on the production line in Hawthorne, California, trucked to McGregor, Texas for acceptance hot-fire under flight conditions, and installed at Starbase as the final step — a partial set first (3 of 6 on a ship, 10 of 33 on a booster) for cryogenic stress testing and static fire, then the full complement back in the Gigabay before a vehicle is launch-ready. But note what kind of distribution this is: three SpaceX facilities, one balance sheet, zero external contracts. Even the residual “supply chain” is internal geography.

Part IVWhy one a day: the fleet math

The mass-production article is the ship, not the booster — and the asymmetry is operational. A booster’s flight lasts minutes; it returns to the launch site and can be ready to fly again in about five minutes, so one or two per launch tower suffice. Ships are different: they loiter on orbit for hours or days, some will fly multi-month Mars profiles, and even returning ships take days to weeks for payload refit. Sustaining a Falcon-9-like cadence — which currently means launching essentially every day — therefore requires many ships simultaneously in various stages of flight, refit, and staging. That is what “one rocket a day” actually provisions: not a stockpile, a fleet in circulation.

The demand stack that justifies the fleet is itself almost entirely SpaceX-internal: Starlink V3 deployment once flight confidence is established; the orbital data-center constellations behind SpaceX’s FCC filings (covered in OED’s February and June briefs); lunar missions that consume multiple tanker launches per single lander; and Mars windows. Today the plant already builds vehicles faster than SpaceX can fly them — finished ships and boosters queue in the Gigabay waiting for the flight program to catch up. The factory is, for now, ahead of the rocket.

Part VThe supply chain read: what OED tracks, and what it should

Here is the structural observation a launch-fan summary will miss. When a company vertically integrates to this degree, the supply chain does not disappear — it changes owners and shrinks to its irreducible external residue. OED tracks the distributed model comprehensively: every SLS prime named above is scored in the company graph, and their cumulative tracked contract value — roughly $90B across just Northrop Grumman and Lockheed Martin Space — is a measure of what the contractor-graph architecture costs. On the Starfactory side, that entire graph compresses into a single Tier 1 entity at the 100th percentile. The investable surface narrows to one ticker, and since June 12 it is an actual ticker: SPCX.

What remains outside SpaceX’s fence is upstream and unglamorous, and it is where OED’s universe has gaps worth closing. The custom alloy still enters the building as mill-rolled coil — someone smelts and rolls it, and specialty stainless producers are absent from the space-economy universe as a category. The robot arms running circumferential welds and tile-pin placement come from industrial automation OEMs; the 36,000-psi hydroforming presses, the overhead crane systems rated for rocket sections, and the ceramic feedstock feeding a 7,000-tile-a-day bakery all have manufacturers. None of these are space companies; all of them are now load-bearing nodes in the highest-capacity launch program in history. The same logic OED applied to orbital compute — track the picks and shovels before the market reprices them — applies to the machines that build the machine.

• SHOULD-TRACK / ADD TO UNIVERSE: specialty stainless coil suppliers (the airframe’s last true external material dependency); industrial welding-robotics and automation integrators; TPS ceramic raw-material suppliers; heavy overhead-crane and material-handling OEMs; high-pressure hydroforming press builders.
• METHOD-TRANSFER ADJACENCY: Tesla (TSLA) — not a space company, but the unboxed method is the direct intellectual ancestor of linear adjacent flow, and the Tesla–SpaceX–xAI industrial coupling is already OED-covered via the TERAFAB brief (Mar 2026).
• DEMAND SIDE, ALREADY TRACKED: Starlink (V3 deployment is the ramp gate); Starcloud (SEPI 27.54, Tier 2) and the orbital data-center cohort, which convert launch capacity into compute capacity; OED’s SpaceX Dependency signal already measures how much of the sector’s throughput routes through this one company.

Part VIOED assessment and market frame

Lead with the counter-thesis, as discipline requires. Nameplate is not throughput: 365 ships a year is a design ceiling claimed in a promotional-access video, not a demonstrated run rate, and the constraint that matters today runs the other direction — the flight program, not the factory, is the bottleneck, with finished vehicles parked in the Gigabay. The demand stack that rationalizes the capacity is overwhelmingly internal (Starlink V3, orbital data centers, Mars), which means the factory’s ROI is a bet on SpaceX’s own downstream execution, not on contracted third-party backlog. Tile production must scale seven-fold; Raptor output in Hawthorne must track a fleet that consumes 39 engines per stack; and reusability economics still depend on ships surviving reentry at operational rates. A skeptic can grant every fact in this brief and still price Starfactory as optionality, not capacity.

The thesis case is what the architecture does to everyone else’s cost curve if even a fraction converts. At 10% of nameplate — 36 ships a year — Starship’s available upmass would exceed the rest of the global launch industry combined; at Falcon 9 cadence the marginal cost of orbit stops being the binding constraint on the space economy, and the bottleneck migrates downstream to payloads, propellant logistics, pad throughput, and regulatory clearance. That migration is the repricing event. Every OED sector thesis that assumes launch scarcity — and several do — carries duration risk against this building. The correct instrument is not sentiment about SpaceX; it is the OED SpaceX Dependency signal, the Launch Services sector print, and the watch items below.

• WATCH: Starlink V3 deployment start — the explicit gate between building rockets and flying them at cadence.
• WATCH: Starship flight rate vs Falcon 9’s (~daily) — the crossover confirms the fleet-circulation model.
• WATCH: Gigabay inventory depth — a growing queue signals flight-program lag; a draining queue signals the ramp is real.
• WATCH: Tile bakery scaling (1,000 → 7,000/day) and Hawthorne Raptor throughput — the two named sub-bottlenecks inside the integrated stack.
• WATCH: Any disclosure of the stainless coil supplier or automation OEMs — the moment the external residue becomes visible, it becomes investable.