Raptor

Why the engine gates

A 39-Engine Vehicle Is an Engine-Production Problem

One Starship stack burns 33 booster engines and six ship engines. A fleet is thousands. The engine has to be built like a car part, not a spacecraft part.

Raptor burns liquid methane and liquid oxygen in a full-flow staged combustion cycle — the most efficient and most complex arrangement of a rocket engine, attempted in flight hardware only twice before SpaceX. Methane is the choice for two reasons that both point at Mars: it burns cleanly without the coking that fouls kerosene engines between reuses, and it can be made on the Martian surface from carbon dioxide and water by the Sabatier reaction. The same chemistry lets Earth propellant be carbon-neutral.

But the binding constraint has never been performance. In November 2021 Musk warned staff of a Raptor “production crisis” that he said could bankrupt the company and stall the next-generation Starlink deployment. The engine that flies is the engine that can be mass-produced, and each Raptor generation is as much a manufacturing redesign as a thrust upgrade — fewer parts, less external plumbing, more that a machine can build unattended. That is why the McGregor site is on this page: the engine and the rate it is built at are one subject. Observed

The full-flow cycle

Both Propellants Go Through Both Preburners

Full-flow staged combustion is why Raptor reaches its chamber pressure and reuse life. It is also why so few engines have ever used it.

In a full-flow staged combustion engine, all of the fuel and all of the oxidizer pass through preburners and enter the main chamber as hot gas rather than liquid, driving two separate turbopumps — one oxygen-rich, one methane-rich. The gases mix by diffusion and ignite on contact, so the main chamber needs no igniter. Running both propellants through the turbines keeps each turbine cooler for a given power, which is what allows the very high chamber pressure and the long reuse life SpaceX wants. Before Raptor, only the Soviet RD-270 of the 1960s and Aerojet Rocketdyne’s Integrated Powerhead Demonstrator of the 2000s had reached testing on this cycle, and neither flew.

Three generations

Each Version Removes Hardware From the Outside of the Engine

Thrust rose about 50% across three generations. The bigger change is what disappeared: pipes, sensors, cables and, on Raptor 3, the heat shield.

The decisive change is Raptor 3. By pulling the regenerative cooling channels and secondary flow paths inside the engine, SpaceX eliminated the external heat shielding that earlier Raptors needed — which is what let the Block 3 booster and ship shed most of their engine-bay shielding, over a tonne per booster engine. Raptor 3 also relights more freely: all 33 Block 3 booster engines can restart for boostback, where Raptor 2 boosters could relight only the inner 13. The parts count fell far enough that Musk credited additive manufacturing, consolidating many machined and plumbed components into single 3D-printed pieces.

Sea level and vacuum

Two Nozzles, One Engine Core

Every Starship stack flies both variants. They share a core and differ at the nozzle.

The sea-level Raptor has a compact bell sized for the atmosphere. The Raptor Vacuum (RVac) adds a large regeneratively cooled nozzle extension of brazed steel tubes, raising the expansion ratio to about 90 and vacuum specific impulse to around 380 s against roughly 350 s for the sea-level engine. The booster flies 33 sea-level engines. A Block 1–3 ship flies three sea-level plus three vacuum; a Block 4 ship is planned at three sea-level plus six vacuum. The vacuum engine is the harder one to produce — as of late 2025 only a single Raptor 3 Vacuum had been observed at McGregor, against booster engines in the dozens, which makes RVac the near-term production bottleneck for the upper stage.

McGregor

McGregor, Texas: Where Every Raptor Is Built and Fired

The world’s most active rocket-engine test site. Production moved here from Hawthorne; nothing flies until it fires here.

McGregor is SpaceX’s propulsion development and test facility in central Texas, and it is where Raptor is both built and proven. Engine production migrated from the Hawthorne, California headquarters — where the first Raptors and the crisis-era engines were made — to a dedicated Raptor factory at McGregor built for high volume; Hawthorne continues to build Raptor Vacuum and develop new designs. Every engine is acceptance-fired here before it is shipped to Starbase for installation, and the site also tests Starship components such as the redesigned header tanks used on Block 2 ships and Block 3 boosters.

McGregor is the original SpaceX test site — it has fired Merlin for Falcon since the 2000s — so Raptor testing shares the grounds with ongoing Falcon 9 engine work. That shared load is why a given week’s firing count mixes both engine families.

Production rate

The Rate Is the Program

Musk’s stated fleet math needs engines by the hundreds per year now and the thousands per year eventually.

The public targets have been consistent for years: a near-term rate of one engine every day or two, a stated aim of two to four engines a day, and a long-run goal of roughly 800 to 1,000 engines a year — the number Musk tied to building a Mars fleet over a decade. NASA reported SpaceX hitting about seven Raptors a week in the Raptor 2 era, roughly one a day. Raptor 3 serials passing 68 by late 2025 show the rate holding into the new generation. The arithmetic is unforgiving: one expendable stack is 39 engines, and even fully reused, a high flight rate means a continuous engine pipeline. The Louisiana buildout and its stated 30-flights-a-day ceiling only make sense against an engine line running far above today’s.

Reliability

Relight, Not Thrust, Is Losing Boosters

Raptor 3’s open problem in flight is restarting reliably for boostback and landing, not producing power on ascent.

Both 2026 Block 3 flights lost the booster to engines that would not relight or failed after relighting. On Flight 12, only 20 of 33 engines lit for boostback and an explosion in one engine took out neighbors. On Flight 13, only 10 of the 13 planned center engines lit for the landing burn, and the booster hit the water hard. Flight 13 was also scrubbed once for ice building up in several Raptor 3 turbopumps. The ships on both flights reached their targets, so the engine makes power reliably on ascent — the unsolved case is the cold, fast, multi-engine relight near the ground. McGregor testing has visibly concentrated on relight conditions and ignition timing for exactly this reason.

Contested

Deliberately Not Claimed

The load-bearing uncertainties, tagged with the confidence they actually earn.

Record

Provenance and Data Sources