In the October 14 space roundup, there was an item that looked routine: Starship’s upper stage, Ship 40, had returned to Starbase, Texas. A spacecraft that had flown into space came back not as wreckage, but as an intact object that could be stood upright in a hangar and studied.
But what grabs you here is not the range record. It’s one line in the report: some of the heat-shield tiles on the underwater side “were not washed away during the entire time in salt water.” It sounds like an amusing little detail from a swimming report. In fact, it is an event of another order.
Ship 40 is Starship’s upper stage, the second stage of the largest rocket system in history. It launched on July 24, 2026, on its thirteenth test flight: it reached space on a suborbital trajectory, deployed satellites, lit an engine in vacuum, and entered the atmosphere at near-orbital velocity. Under the standard scenario, Starship stages sink afterward. This one did not.
And that is where the real story begins: for the first time, engineers had a physical, complete, flight-proven sample of the heat shield—the very component SpaceX calls the hardest unsolved problem in the entire program. Thirteen flights before it had yielded telemetry, video, and individual samples. Now there is a whole vehicle, with 18,000 tiles, each of which has been through the fire.
The timeline alone deserves pages of its own. After a soft splashdown in the Indian Ocean, the ship behaved against expectations: instead of rolling over and breaking apart like previous stages, it stayed afloat and kept transmitting data. SpaceX had not planned for this. Flight 13 was intended as a standard suborbital flight, and the value of recovered hardware had never been part of the mission.
What followed was improvisation on the open ocean. Sailors kept the ship afloat by hanging inflatable Yokohama fenders around it—the same giant rubber bumpers used to berth vessels. A few days later, the tugs Normand Ranger and Skimmer Tide arrived, while the salvage vessel Go Australis had remained nearby since landing. The ship was towed engines-first for several hundred miles toward Christmas Island—an Australian territory in the Indian Ocean.
On August 7, Elon Musk publicly wrote off the operation: “Unfortunately, recovery of the ship is looking unlikely at this point.” Eleven days later, on August 18, the ship was brought to the island after all. By then, it had spent around 24 days in the ocean.
It is worth being precise with the words here, because they blur in retellings. “Intact” does not mean “undamaged”: official photos show a scratched-up ship with heat-shield material flaked away and partly missing. “Recovered” does not mean “at the dock”: as of August 18, it was simply floating offshore near the island. Engineers climbed aboard and sampled the shield right there, in calm water; then, on August 26, it was loaded onto the semi-submersible vessel Forte for transport across half the planet. The ship arrived at the Port of Brownsville and then at Starbase on October 7—75 days after splashdown.
It is also worth spelling out why this operation inspired doubts in the first place. Starship was designed as a spacecraft, not a floating vessel, and as a towed object it is awful in every respect. It has no crew to take a line. No steering in its cooled-down, powerless state. No conventional hull to hold steady in the waves—instead, it has two enormous, lightweight pressurized steel tank-vessels, shaped for flight rather than resistance to waves. Its aerodynamic flaps, built to steer a falling spacecraft, spent weeks taking the blows of ocean swell. And over all of that—the same thousands of ceramic tiles, each essential during atmospheric entry and each vulnerable to impact and abrasion afterward.
The ship is 171 feet long, around 52 meters—the height of a seventeen-story building. Even lying on its side, it presents enormous surface area to waves and currents. One serious bout of rolling or one snapped towline would have been enough to turn an intact ship into scattered debris. So Musk’s pessimism on August 7 was not posturing. It was a sober assessment.
What exactly was salvaged becomes clear once you look at how the shield is built. Starship’s thermal protection system is roughly 18,000 ceramic tiles covering the windward side—the side the spacecraft presents as it enters the atmosphere. The tiles are hexagonal and roughly the size of a dinner plate. Each is designed to take the flow of superheated gas and keep that heat from reaching the structure.
Starship’s crucial difference from all earlier hardware is the material of its primary structure. The shuttles and the Dragon capsule used aluminum, which melts and loses strength long before a tile has time to fail. Starship uses a special stainless-steel alloy that holds up to 800 °C. The practical meaning is simple: where losing a tile meant catastrophe for the shuttle, Starship can “survive” some of the heat. The system is forgiving, as engineers put it—tolerant of mistakes in places where mistakes are not normally tolerated at all.
A tile is not needed for the ship to make it home. A tile is needed for the ship to make it home twice.
The ancestor of this technology is the Space Shuttle’s thermal protection system, and its design story is worth recalling because it explains why this whole ocean circus was needed in the first place.
At its core was RSI, a reusable ceramic insulation made of silica fibers, developed by Lockheed back in the late 1950s. The material was 89% void space, weighed about 9 pounds per cubic foot, and withstood repeated heating cycles to 2500 °F. It did not burn away like ablative coatings; it released moisture, creating a flow of cool gas outward. The trouble was its fragility: covering the entire vehicle with it was impossible, so it was cut into six-by-six-inch tiles and bonded with gaps of about 0.01 inches—so the metal beneath could deform under heat.
By mid-1979, it emerged that in some areas the tiles “did not have sufficient strength to survive the tensile loads of one flight.” A race began. A special densification process—impregnating the back side of the tile with a mixture of Ludox colloidal silica and alcohol, drying it for a day, then baking it at 150 °F—raised the system’s strength from 6 psi to the required 13. Tiles already installed on the orbiter had to be load-tested at 125% of maximum flight load, and 13% of those inspected failed. Acoustic sensors listened for internal fibers breaking inside the ceramic.
For twenty months, technicians worked three shifts, six days a week. The result: 30,759 tiles installed, checked, in places removed, reinstalled, and checked again. The average pace: 1.3 tiles per person per week. In June 1979, Rockwell estimated that 10,500 tiles needed replacement; by January 1980, nine thousand of them had been replaced, while the queue had grown to 13,100—because new tiles kept failing inspections one after another. Fewer than a thousand remained by Thanksgiving.
This is not an anecdote from the distant past. It is an exact inventory of what SpaceX swore it would not repeat. Musk stated the goal plainly: land the ship, load it with propellant, launch it again—without a “labor-intensive inspection of 40,000 tiles.” He used 40,000 as a metaphor, but the metaphor grew out of the shuttle’s real arithmetic.
Back to Ship 40. Even before it returned, images of the shield after splashdown spread through the industry and sparked a dispute worth presenting honestly, because both sides have numbers to cite.
The good part: compared with Flights 10 and 11, the heat shield looked noticeably more even, with only a handful of tiles lost. Starship V3, which first flew on the twelfth mission, received a new tile geometry and improved retaining clips developed from data gathered in previous launches—and, judging by the images, it worked. One commentator called the shield “the real star” of the twelfth flight.
The bad part is clearer in the details. White streaks ran across the shield, and all of them began at the boundaries between tiles. They are traces of hot gas leaking through the joints and flowing beneath the tile: underneath is an ablative barrier inherited from Dragon’s thermal protection system, and when it burns away, it leaves a white residue. SpaceX official Bill Gerstenmaier explained it in plain language after the tenth flight: “We realized that we need to seal the tiles.” At the same time, he candidly reported on a failed experiment—three metal tiles installed in place of ceramic ones in the hope of simpler manufacturing and greater durability. The metal oxidized in the oxygen-rich environment—“extremely beautifully”—and proved unusable. The rusty-orange stain along the side of the ship, noticed by the public, was the trace of precisely that failure.
The response was “crunch wrap”—material wrapped around each tile, which a robot presses in with the tile and trims flush with the surface. The idea is to seal the joints without gap fillers, which added complexity on the shuttle and sometimes came out in flight.
After the thirteenth flight, three former and current NASA professionals—Dan Rasky, Charles Camarda, and Charles Miller—watched footage of the shield and published their analysis. Their conclusion sounds harsher than the image: only a handful of tiles were lost, but “many tens, if not hundreds of tiles are cracked, chipped, and severely damaged.” In other words, the shield’s apparent health is deceptive—white streaks, cracks, and chips become legible only under close examination.
This is where the fracture lies.
On August 4, during SpaceX’s first-ever quarterly earnings call, Musk said he considered the heat-shield problem solved. A few days later, Ars Technica published the opposite verdict: “Starship’s current thermal protection system is a dead end for any missions requiring full and rapid reusability.”
The author of that quote is Dan Rasky, a man with 38 years at NASA Ames, a co-author of PICA—the thermal-protection material used on Crew Dragon—and one of those who brought TUFI technology to flight on the shuttle in 1994. The heart of his objection is not that the tiles are bad. It is that they are good for exactly what they were designed for: a single atmospheric entry. Mechanically attached “shuttle-style tiles” are fragile by their very nature, while vibroacoustics and the dissipation of enormous thermal power are not problems solved by careful installation.
Then come the consequences, and they are unpleasant. With this heat shield, according to the trio, Starship will become fully reusable and reduce launch costs below Falcon 9. It will be enough for deploying the Starlink V3 constellation and other large low-Earth-orbit systems. But it will not reach the flight rate and turnaround speed needed for lunar bases, Martian settlements, orbital data centers, and space-based solar power. The cost per kilogram to orbit will remain above $100–250. Starship will not become airline-like with this shield.
And the main, most cutting point in their analysis: every technology flying on Starship today was developed 30–40 years ago—in the shuttle, National Aerospace Plane, and X-33 programs. Fundamental U.S. research into thermal protection for Mach 25 entry was wound down in the 1980s, cut further in the 1990s, and mostly shut down after the cancellation of the 2GRLV and NGLT programs in the early 2000s. No one today is seriously funding this work at low technology-readiness levels. A recommendation to create a new NASA thermal-protection initiative went nowhere.
The dispute did not end there, and that matters. The trio’s argument was publicly challenged: there is an analysis podcast that takes their position apart point by point—and asks Rasky to say what he really thinks. So this is not an established fact, but a live engineering debate in which both sides have names, credentials, and data. SpaceX, for its part, shows not the slightest intention of changing course this late in the game.
Back to where I began. Why drag a vehicle across the ocean if it will not be reused anyway?
The answer is in what they are doing with it now. On October 13, SpaceX stood Ship 40 upright in Starbase’s Megabay alongside three other ships, and stated outright that it would provide “first-of-its-kind data” on a flown heat shield, Raptor engines, and avionics. The company did not specify how long the teardown would take or when results might emerge.
The logic is simple, and it explains everything else. Telemetry shows when the structure flexed, when temperature rose, when pressure fell. Cameras show the moment a tile disappeared. Neither shows what happened next—the physical chain of events that caused damage to this particular joint, this particular fastener, this particular seam. A white scar permanently frozen on a specific tile is an answer to a question no instrument can be asked.
And there is one more detail that makes this specimen genuinely valuable: some of the damage was sustained during atmospheric entry, some during splashdown and rollover, and some during three and a half weeks of waves, salt, and towing. Separating those causes is not easy, and that is exactly why the ocean does not diminish the find—it turns it into a task for a forensic investigator. Everything visible in photos before the storm can be subtracted from what is visible afterward. What survived a long soak can be studied alongside what did not.
The shuttle parallel reverses itself here. If the shuttle was a story about fragile tiles forcing people to count debris, Ship 40 is the story of how debris can finally be counted precisely.
The temptation to declare Ship 40’s return a victory for reusability is great, and it should be resisted.
This ship proves nothing about turnaround. It did not land on a pad, was not caught by the tower’s mechanical arms, did not undergo a limited inspection, and did not fly again. It drifted for 24 days and traveled 10,000 miles aboard a semi-submersible vessel. Salt is hostile to engines, valves, electronics—to everything that will have to be trusted on the next flight. There are no public indications that Ship 40 will be reused, and given the nature of the operation, there could not have been.
The program’s list of unresolved issues still includes items the recovered ship does not cross off: thirteen flights without propellant transfer between ships in orbit, without catching an upper stage at the tower, without refurbishing and relaunching even one stage. Meanwhile, the surrounding context is pressing in: the FAA is preparing an environmental review for 456 launches a year versus the current 25, while third-generation Raptors are repeatedly failing in parallel—on the twelfth flight, engines let down both the booster and the ship; on the thirteenth, the failure of four engines scrubbed a July 16 launch attempt; and during the booster’s descent on July 24, engines failed again.
But this is what Ship 40’s return truly changes. It turns the argument over thermal protection from a matter of faith into a matter of measurement. Musk says “solved,” Rasky and his colleagues say “dead end,” Jalopnik snarks about gaps in the panels—and all of them are right to the extent that people arguing from photographs and telemetry can be right. The ship in the Megabay is the first thing in this argument that can be taken apart and touched by hand.
There is still no answer to whether the shield can withstand hundreds of flights a year. And that is perhaps the most honest conclusion to this entire story: the question has finally been framed correctly, and the answer will have to be extracted in the laboratory, tile by tile.
Just as the shuttle’s thirty thousand tiles were once counted at the rate of one per person per week. The difference is that this time, it seems, someone is finally recording the results.