NASA just bought four more Moon landings for $590 million. Three of its last four landers tipped over.

NASA announced this week that it is buying four more trips to the Moon. The headline number is nearly $590 million, split among three companies: Astrobotic gets $297.9 million for two deliveries, Intuitive Machines $148.3 million for one, and Firefly Aerospace $144.2 million for one. All four are meant to land in late 2028, each carrying the same trio of science instruments, as part of what NASA now calls its Moon Base Program. It is a tidy, confident announcement. Here is the number it does not lead with: of the four commercial landers that have actually attempted a lunar touchdown so far, three ended up on their sides or never got there at all.

That is not a gotcha. It is the whole business model, stated honestly. NASA is not paying for four guaranteed landings. It is paying for four attempts, at a price per attempt low enough that it can afford to lose some. Whether that is smart or reckless depends entirely on a spreadsheet, so let's do the spreadsheet.

What a landing costs now

Start with the price, because the price is the entire argument for doing it this way. These four deliveries land at roughly $145 million to $150 million each. A Moon lander built the old way — designed, owned and operated by NASA itself — runs to the billions; the agency's own robotic missions of the past routinely cost more than a billion dollars apiece before anyone left the launch pad. Buying a delivery at a fixed price from a private company, instead of building the spacecraft in-house, is a way of turning a billion-dollar science mission into something closer to a courier invoice. That is the pitch of Commercial Lunar Payload Services, the program these contracts fall under, and on cost alone the pitch is real.

The structure underneath it matters as much as the number. CLPS is what the contracting people call an indefinite-delivery, indefinite-quantity vehicle: a standing arrangement with a cumulative ceiling of $2.6 billion through 2028, against which NASA writes individual task orders as it wants deliveries. More than $1.4 billion of that had been ordered by May 2026. These are fixed-price deals, which is the part that transfers the risk. If a lander costs more to build than the contract pays, the company eats the difference. If it tips over on arrival, the company — not the taxpayer — has largely already absorbed the cost of getting it there. NASA is buying outcomes, and only paying full freight for the ones that work.

NASA is not paying for four guaranteed landings. It is paying for four attempts, at a price low enough that it can afford to lose some.

The landing rate is the variable that decides everything

Now the uncomfortable column of the spreadsheet. Run the record to date. Astrobotic's Peregrine, in January 2024, sprang a propellant leak and never reached the Moon. Intuitive Machines' first lander, Odysseus, touched down a month later — the first private spacecraft to do so — but came to rest on its side with a broken leg, cutting its mission short. The company's second try, Athena, tipped over inside a shadowed crater near the south pole in early 2025 and lasted about an hour. Only Firefly's Blue Ghost, in March 2025, set down upright and worked as intended, delivering ten NASA payloads through a full lunar day. Four attempts, one clean landing. If you are pricing the program like an accountant, your expected number of fully successful landings out of these next four is not four. On the record so far, it is closer to one.

That is the math NASA is betting against, and to be fair to the agency, it has a reason to think the odds improve. Every one of these four deliveries uses an upgraded version of a lander design that has already flown, not a clean-sheet vehicle. Firefly is flying the descendant of the one lander that worked. Intuitive Machines is flying the descendant of two that reached the surface, learned hard lessons about center of gravity and landing legs, and presumably fixed them. Even Astrobotic, whose Peregrine never made it, is flying refined hardware. The theory is that a landing is a thing you get better at by doing it, and that the second and third attempts on a given design should tip over less than the first. The reused-design detail is the single most important line in the announcement, because the whole business case rides on that curve bending the right way.

The tell is in the payload

Look at what each of these four landers is carrying, because it gives away how NASA is thinking. Every one of them flies the same three instruments: a stereo camera system called SCALPSS that watches how the rocket exhaust throws up dust during landing, a laser retroreflector for precise navigation, and a radiation spectrometer to measure what the environment does to hardware and, eventually, people. That is not how you fly a one-off flagship science mission, where every kilogram is fought over and customized. That is how you run a bus route. You standardize the payload, fly it repeatedly to different spots, and treat each landing as one data point in a series rather than a singular event. Standardization is the quiet admission that this is infrastructure, not exploration — and infrastructure is exactly the kind of thing where cheap-and-repeated beats expensive-and-perfect, as long as the vehicles hold up.

It is also worth naming who is on the receiving end of these checks, because the fixed-price model only works if the companies survive their own failures. Astrobotic collected the largest share here, two deliveries, despite Peregrine never reaching the Moon and despite a separate, larger lander program that has been repeatedly delayed. Firefly, fresh off the one unambiguous success, recently went public and can use the revenue. Intuitive Machines has landed twice and tipped twice. NASA is spreading four deliveries across three companies partly to keep more than one of them alive, because a delivery service with only one surviving provider is not a service. It is a single point of failure with a lander attached.

Does the math close?

So back to the question I always end on: does it actually pencil out? On cost, yes, and it is not close. Even if only one or two of these four landers works, NASA will have bought that lunar science for a fraction of what a single agency-built lander would have cost, and it will have paid nothing like full price for the failures. A 25 percent success rate at a tenth of the cost still beats a 90 percent success rate at ten times the cost, if all you care about is science-per-dollar. The commercial model does not need every landing to work. It needs enough of them to work that the average delivered payload comes in cheaper than the alternative, and even at today's grim landing rate, it probably does.

The risk is not on the cost line. It is on the durability of the companies and the phrase doing the heavy lifting in NASA's own announcement, which is "Moon Base." Lori Glaze, NASA's associate administrator for human spaceflight, said the awards "demonstrate our commitment to accelerating our effort to build a long-term presence on the lunar surface." A base is a cadence, not a delivery — it assumes these flights keep coming, on schedule, for years. That only holds if fixed-price contracts don't quietly bankrupt the thin-margin startups flying them, and if the landing rate really does climb the way the reused-design theory promises. Buy four attempts and you are buying an experiment in whether commercial lunar delivery can become routine. The moment to grade it is not the 2028 launch. It is the ledger a year later, when we can count how many of the four are sitting upright and sending data — and whether the companies that built them are still in business to fly the next ones.