In nineteen fifty-four, engineers on two continents started designing the fastest passenger aircraft anyone had ever attempted. Twenty-two years later, twenty of them entered service. The forecast had been three hundred and fifty.
And every airline that wasn't state-owned had already walked away by then.
Daniel's got a whole thing this week about exactly that feeling. He wants five specific examples, machines and systems and products, not broad categories, where a technology reached the absolute summit of its paradigm. The finest, most expensive, most painstakingly engineered version ever built. And then got overtaken almost immediately by something else.
And he wants the absurdity of the timing front and centre.
Right. He's asking for a timeline on each one. When development began, how much money or effort went in where that's even knowable, when it entered service or peaked, and when the replacement made it irrelevant. What made it impressive, what engineering problems got painstakingly solved, what displaced it, and why the transition happened so fast. He says to prioritise the cases where the contrast is most dramatic. Years of work, enormous resources, and then only a few years at the frontier. Ideally shorter.
He's not asking for failures.
No. That's the thing. He's asking for the opposite of failures. He wants the machines that were the best ever at what they did, and it turned out that didn't matter. His phrase is that the engineers reached the summit of an entire paradigm just in time to discover everyone else had started climbing a different mountain.
That's a good frame.
So let's climb five mountains that turned out to be the wrong ones.
Before we start, I want to nail down what this pattern actually is, because it's not the usual story about technology dying.
Go on.
The normal story is that something was worse and got replaced by something better. That's not what we're talking about. In every case Daniel's pointing at, the machine that died was the finest example of its kind ever built. Nobody built a better ocean liner than the SS United States. Nobody built a faster airliner than Concorde. The Cray-1 was the most powerful computer on earth and it was beautiful.
So what killed them?
The paradigm moved. And there's a distinction I want to keep straight the whole way through, because it matters for how you read these stories. Technological obsolescence and commercial obsolescence are different things.
Explain.
Concorde flew passengers for twenty-seven years. Technologically it was never obsolete. It did exactly what it was designed to do, right up to the last flight in October two thousand three. But it was commercially dead almost from the day it entered service. The SS United States sailed for seventeen years and was killed by the jet within a decade of its maiden voyage. Meanwhile the Tu-144 and Buran had almost no service life at all. Those are short. So the pattern has two flavours and we should be honest about which is which.
Which is the more interesting flavour?
The one where the machine works perfectly and the world stops wanting it. That's the one that's strange.
So five cases. Give me the list.
Concorde and the Tu-144 for supersonic transport. Buran and Energia for the Soviet shuttle. Iridium for satellite phones. The Cray-1 for supercomputing. And the SS United States for ocean liners. With the Boeing 2707, Saturn V and the N1 rocket as supporting characters.
That's a lot of dead rockets.
It's a lot of dead everything. And there's a mechanism running underneath all of them that I want to flag now, because it explains most of the carnage. Long development cycles lock in assumptions about markets and competing technologies that are guaranteed to be stale by the time the product ships.
Say that more concretely.
Dan Colussy, the man who eventually rescued Iridium out of bankruptcy, put it about as well as anyone has. He said the Iridium business plan was locked in place twelve years before the system became operational. Twelve years. You write a business plan in a world where nobody has a mobile phone, and you execute it in a world where everybody does.
That's the whole episode in one sentence.
It's most of it. But let's start with the aeroplanes, because they're the most expensive version of the mistake.
Concorde.
Concorde. Studies began in nineteen fifty-four. The Anglo-French treaty was signed on the twenty-ninth of November, nineteen sixty-two, with an original cost estimate of seventy million pounds. That's about one point seven seven billion in today's money.
And the final bill?
One and a half to two point one billion pounds by the time it entered service in nineteen seventy-six. Twelve to sixteen point seven billion in today's money. Roughly six times the original projection.
Six times. That's not an overrun, that's a different project wearing the first one's coat.
And here's the number that actually kills it. They forecast a market of three hundred and fifty aircraft. They built twenty.
Twenty.
Twenty airframes. And every airline except Air France and British Airways cancelled. Pan Am, TWA, United, Qantas, Lufthansa, Japan Airlines, all of them pulled out between nineteen seventy-two and nineteen seventy-five. Over a hundred options evaporated.
What did the engineers actually solve, though? Because this is the part I want on the record. These weren't incompetents.
No, they were the best in the world. The ogival delta wing came out of Küchemann and Weber's work on slender-delta vortex lift. Analogue fly-by-wire, first on any airliner. Digital air-intake control units, which was the first full-authority digital control of an essential system on a passenger aircraft. They used the fuel as a heat sink to cool the airframe. The droop nose so the pilots could see the runway on approach.
And the airframe grew.
Three hundred millimetres in flight. Twelve inches, from kinetic heating. They had to cap the skin at a hundred and twenty-seven degrees Celsius, which is what set the speed limit at Mach two point oh two. Every one of those is a hard problem that had never been solved before.
And the thing that killed it was a fuel bill.
Fifteen point eight passenger-miles per gallon. The Boeing 747 got forty-six point four. The DC-10 got fifty-three point six.
So it was three times thirstier than a jumbo jet.
Three times thirstier than a jumbo jet, carrying a quarter of the passengers. Then the oil crisis hit in nineteen seventy-three, and wide-body jets were already scaling. Concorde was commercially irrelevant almost the moment it arrived. It flew until October two thousand three, but it was a museum piece with a schedule for most of that time.
And the Soviets built one too.
The Tu-144 is the more extreme case, and I think it's the purest version of Daniel's prompt in the whole episode. First flight on the thirty-first of December, nineteen sixty-eight. Two months before Concorde.
Deliberately.
Entirely deliberately. They flew it on the last possible day to hit a five-year-old government deadline. Brian Calvert, who was Concorde's flight manager, said the rush exacted a heavy penalty later, and he was being generous.
How bad was it?
Passenger service ran from the first of November, nineteen seventy-seven, to the first of June, nineteen seventy-eight. Fifty-five passenger flights. It was pulled after a crash-landing. In a hundred and two flights it logged over two hundred and twenty-six failures.
Two hundred and twenty-six failures in a hundred and two flights.
That's more than two per flight, yes. The designer, Alexander Poukhov, said in nineteen ninety-eight that the project was ten to fifteen years beyond the USSR's capabilities at that time. His own words. And Howard Moon, who wrote the book on the Soviet SST, called it an astounding achievement and a magnificent failure in the same breath, which I think is exactly right.
And then NASA bought one.
Spent three hundred and fifty million dollars converting it into a flying laboratory in the nineties. The Tu-144LL. So the machine that couldn't carry passengers ended up as a research aircraft for the people who'd beaten it.
There's something almost poetic about that. The loser gets a second career as a test rig.
It's the most dignified ending any of these things got.
Buran next.
Buran is where the numbers stop being abstract. The programme formally began in nineteen seventy-six as a direct response to the American shuttle. The cost came to seventeen point eight billion roubles in nineteen ninety figures. That's about seventy-four billion dollars in today's money. The largest and most expensive project in Soviet space history.
And how many times did it fly?
Once. The fifteenth of November, nineteen eighty-eight. Uncrewed. Fully automated flight and landing.
Which the American shuttle couldn't do.
Which the American shuttle could not do. That's the part people forget. The one time Buran flew, it did something the Shuttle programme never managed. It came back and landed itself, autonomously, in a crosswind, without a pilot aboard.
So it was more advanced in that one respect.
In that one respect, yes. And then it never flew again. It was destroyed in a hangar roof collapse on the twelfth of May, two thousand two. Killed eight workers. The programme itself was cancelled by Yeltsin on the thirtieth of June, nineteen ninety-three.
What was the actual problem?
Cost per launch, and no mission. Launching twenty tons on Energia-Buran cost an estimated two hundred and seventy million roubles. The same payload on a Proton rocket cost five and a half million.
That's a factor of...
Forty-nine.
Forty-nine times the price for the same job.
And the cosmonaut Oleg Kotov said it plainly. We had no civilian tasks for Buran and the military ones were no longer needed. That's the whole thing. They built the most sophisticated spacecraft in their history and there was nothing left for it to do.
There's a thread running through these three that I want to pull on.
Go ahead.
All three of them were partly political symbols. Concorde was Anglo-French prestige. The Tu-144 existed because Khrushchev wanted one. Buran existed because the Americans had a shuttle. These weren't built because a market demanded them. They were built because somebody needed to be seen building them.
And that's an accelerant and a trap at the same time. It gets you funding you'd never get commercially, and it locks you into a specification you can't walk away from when the world changes.
The Tu-144 is the cleanest example. They flew it on New Year's Eve because a deadline from five years earlier said they had to.
And it cost them. Every shortcut they took to hit that date came due later. You can't rush a supersonic airliner. The physics doesn't care about your five-year plan.
Those were all aerospace projects with national treasuries behind them.
Every one.
But the same pattern shows up in commercial technology. Where the money was private, the timeline was just as long, and the market moved just as fast.
Iridium. And I think Iridium might be the single best answer to what Daniel's asking, because it's the clearest case of something being obsolete before it launched.
Tell it properly. It starts with a man on holiday.
It starts with Bary Bertiger, a Motorola engineer, in nineteen eighty-five. His wife was trying to reach clients from the Bahamas and couldn't. So he went home and designed a constellation of satellites so that nobody would ever be out of coverage again.
That's a very engineer response to a holiday inconvenience.
It is. And the idea was good. Seventy-seven satellites in low earth orbit, handing calls off between each other as they moved. Seventy-seven is the atomic number of iridium, which is where the name comes from. They ended up flying sixty-six active.
And the build rate?
Motorola was producing a satellite every four point three days at peak. That's a production line, not a workshop. They developed it on a fixed-price contract from July nineteen ninety-three, and service went live on the first of November, nineteen ninety-eight.
And the total?
Over five billion dollars.
For a phone network.
And here's where the twelve years matter. The business plan was written when a mobile phone was a brick that rich people had in their cars. By the time the satellites were up, cellular coverage had gone from a luxury to something approaching a default. The thing Iridium was built to solve had been solved by something else, more cheaply, while they were still building.
What did the handset look like?
A pound in weight. Three thousand dollars. Calls at three to eight dollars a minute.
Three to eight dollars a minute.
And the coverage was extraordinary. You could call from the middle of the Pacific. There just weren't very many people in the middle of the Pacific who needed to.
So it filed.
Chapter eleven on the thirteenth of August, nineteen ninety-nine. One of the twenty largest bankruptcies in American history at the time. It had drawn fewer customers in ten months than the plan expected in the first year.
Ten months from launch to bankruptcy.
Ten months. And Colussy's line is the one that should be on the wall of every product manager's office. The business plan was locked in place twelve years before the system became operational.
Iridium was done in by cellular networks.
Yes.
The Cray-1 was done in by something even smaller. And the contrast is even more dramatic, because the Cray-1 wasn't beaten by a cheaper version of itself. It was beaten by a completely different idea of what a computer was.
Seymour Cray spent four years designing it after he left Control Data. He'd been working on the CDC 8600 there and it had failed, so he walked out and started again.
What did he build?
Announced in nineteen seventy-five, first installed at Los Alamos in nineteen seventy-six. Seven point nine million dollars. About forty-two million in today's money.
For one computer.
Two hundred thousand gates. A C-shaped chassis, and the shape isn't aesthetic, it's electrical. He arranged the boards in a curve to shorten the wire lengths, because at that speed the wire is the delay. Twelve and a half nanosecond cycle time. Eighty megahertz. Cooled with Freon.
Liquid cooling in nineteen seventy-six.
Because air couldn't carry the heat away fast enough. And the thing that made it matter was the vector processor. First successful one. A hundred and sixty million floating point operations per second, several times faster than anything else on earth.
And it looked like a piece of furniture.
It looked like a curved bench with a loveseat in the middle. People used to sit on it for photographs. Seymour Cray reportedly got the idea for the shape from watching a football game.
Of course he did.
There's a story attached to the predecessor that I love. The CDC 6600, which Cray also designed, was the fastest computer in the world from nineteen sixty-four to nineteen sixty-nine. And Thomas Watson Junior at IBM wrote a memo on the twenty-eighth of August, nineteen sixty-three, saying, I fail to understand why we have lost our industry leadership position by letting someone else offer the world's most powerful computer.
And Cray's reply?
It seems like Mr. Watson has answered his own question.
That's the best line in the episode.
It's the best line in the history of computing, and I will not be taking questions.
So what killed the Cray?
The microprocessor. And the timing is almost cruel. The supercomputer industry crashed in the early nineties, because commodity chips got good enough and you could wire thousands of them together into a massively parallel cluster. You didn't need a bespoke vector machine anymore. You needed a lot of ordinary ones.
How far did the gap close?
By twenty thirteen, a typical smartphone processor was hitting about one gigaflop. That's roughly six times the Cray-1.
A phone in your pocket, six times the fastest computer on earth, thirty-seven years later.
And it cost a few hundred dollars instead of seven point nine million. The Cray-1 was the fastest machine of its kind for years. But its kind stopped being the kind that mattered.
The SS United States was the fastest ship of its kind, period. A record that still stands. And it was obsolete almost before it left the dock.
William Francis Gibbs designed it. Ordered in nineteen forty-nine, launched on the twenty-third of June, nineteen fifty-one, maiden voyage on the third of July, nineteen fifty-two. Cost seventy-one point eight million dollars. About six hundred and ninety-four million in today's money.
And the government paid for part of it.
They paid for what were called national defence features. The ship was designed so it could be requisitioned as a troop carrier at a moment's notice. That's why there's no wood anywhere aboard. It's a fireproof ship. Military-grade damage control throughout.
And the speed.
Two hundred and forty thousand shaft horsepower from four Westinghouse geared steam turbines. Thirty-eight point three two knots on trials. That is still the fastest Atlantic crossing ever made by a passenger ship. The Blue Riband was never contested. Nobody even tried.
She took it and retired the trophy.
Effectively, yes. They also fitted secret five-bladed inboard propellers, which is the kind of detail that only shows up when a navy is quietly involved in a passenger liner.
And then the jets came.
The first transatlantic jet passenger service was the fourth of October, nineteen fifty-eight. BOAC Comet. Six years after the SS United States entered service.
Six years.
Passenger numbers held up for a while, because flying was still expensive and uncomfortable and the ship was neither. But through the mid-sixties it collapsed. She was withdrawn in a surprise announcement on the fourteenth of November, nineteen sixty-nine. Seventeen years of service.
The finest ocean liner ever built, and it got seventeen years.
And the ending is strange. She's been sitting at a pier in Philadelphia for decades, and she's now being prepared to be sunk as the largest artificial reef in the world, off Destin, Florida.
The fastest ship ever built becomes a fish habitat.
It's not the ending Gibbs had in mind.
Give me the supporting cases before we pull this together, because Daniel asked for five and I want the shape of the thing clear.
Boeing 2707. The American supersonic transport. Kennedy launched the programme on the fifth of June, nineteen sixty-three. Boeing won the design competition on the first of January, nineteen sixty-seven. It was cancelled in nineteen seventy-one when the House voted two hundred and fifteen to two hundred and four to cut the funding. Both prototypes were still unfinished.
And the order book?
A hundred and fifteen unfilled orders from twenty-five airlines. All of them gone. The collapse cost Boeing over sixty thousand jobs. People called it the airplane that almost ate Seattle.
Sixty thousand jobs for an aircraft that never flew.
Never flew. Not one prototype. And then Saturn V. Thirteen launches between nineteen sixty-seven and nineteen seventy-three. Six point four billion dollars for the programme, about thirty-four and a half billion today. A hundred and eighty-five million per launch, which is roughly a billion in today's money.
The most powerful rocket ever flown.
And after Apollo it never flew again. The United States had no comparable heavy-lift capability for decades. They had to build a new one from scratch.
They threw away the tooling.
And the N1, the Soviet lunar booster. Four launches between nineteen sixty-nine and nineteen seventy-two, all four failed. Cost per launch, six hundred and four million in nineteen eighty-five dollars, which is about one point eight billion today. Cancelled in nineteen seventy-six.
Four launches, four failures, one point eight billion dollars a go.
The Soviet moon programme died with it.
So let's pull the pattern out, because there are three distinct mechanisms here and I don't think they get separated often enough.
Go.
First mechanism. The locked-in business plan. Iridium is the cleanest, twelve years from plan to operation. But Concorde did the same thing with its three hundred and fifty aircraft forecast. The Tu-144 did it with its twenty-aircraft plan. You commit to a market assumption at the start of a development cycle so long that the assumption cannot possibly survive it.
And the longer the cycle, the worse it gets. Aerospace is the worst offender because the cycles are measured in decades, not quarters.
Second mechanism. The cost-per-unit absurdity. Buran cost forty-nine times what a Proton did for the same payload. Concorde burned three times the fuel of a 747 for a quarter of the passengers. Being the most advanced machine can be economically fatal, because advanced usually means expensive, and expensive usually means you lose to the cheap thing that's good enough.
The displacing technology is almost never more sophisticated. That's the part people get wrong. The 747 is a much simpler aircraft than Concorde. A cellular network is much simpler than a satellite constellation. A microprocessor is much simpler than a vector supercomputer. Simpler, cheaper, and scalable is what wins.
Third mechanism. The summit just as the mountain moves. The Cray-1 and the SS United States both peaked at the exact moment a fundamentally different technology was scaling. Microprocessors. Jets. The engineers solved the hardest problems of the old paradigm precisely when the paradigm stopped mattering.
And that's the thing that makes it tragic rather than stupid. They weren't wrong about anything technical. Every problem they solved, they solved correctly. They just solved the wrong set of problems.
Because forecasting isn't an engineering problem.
No. And it doesn't yield to engineering methods. You can't test a market assumption the way you test a wing. You can't iterate on it in a wind tunnel.
There's one more angle on this that we haven't touched. The human scale of it.
The billions of dollars are almost too big to feel.
They are. Once you're past a certain number it stops meaning anything.
It's the same problem with all of them. You hear seventy-four billion dollars for Buran and your brain just files it as a big number. It doesn't land.
The individual engineer's eighteen months doesn't have that problem.
No. That lands.
I spent a while thinking about that this week. The people who wrote the specifications, the test procedures, the failure analysis. All that documentation for a machine that never got built.
It went somewhere.
It went into a drawer, mostly.
I think the money is the least interesting part of this.
So do I. The money's gone either way. What's interesting is the knowledge.
Whether it transfers.
Because if the next paradigm doesn't need it, it just sits there. Correct, careful, useless.
I don't know how you'd even measure that.
You can't. That's why nobody talks about it.
The documentation is the part that has no champion. Nobody writes a retrospective about the test procedures for an aircraft that never flew.
Right. The books get written about the money and the politics. The specifications go in a box.
And then one day somebody's clearing out a garage.
There's a delivery coming that needs a signature and the window closes in about ten minutes.
Sorry?
It's fine. Go on.
That was Hilbert.
I know.
He's gone.
He'll be back. He's always back.
He still has the drawings. And the museum didn't want them.
Because the machine was never built.
So the documentation of the finest version of a technology that never flew is sitting in a box in a garage, and the only reason we know it exists is that he mentioned it on his way out the door.
There's something in that. The archive of the thing that didn't happen.
Every one of these cases has one. Somewhere there's a filing cabinet with the complete engineering record of a machine that lost.
And no historian will ever open it, because the interesting story is the one that flew.
The interesting story is the one that flew. The box is the story of the one that didn't.
Which brings me to the thing I keep circling.
Go on.
Is this pattern accelerating or slowing down?
That's the right question and I don't know the answer.
Because the development cycles have compressed enormously. Iridium took twelve years from plan to operation. Software takes twelve weeks.
Which could cut either way. Faster iteration means you find out sooner that you're building the wrong thing. It also means you can commit to a wrong assumption and execute it completely before anyone notices.
So are we building more Iridiums or fewer?
My instinct is more, but the failure is quieter. A five billion dollar satellite constellation going bankrupt is a newspaper story. A company burning two years and forty million dollars on the wrong architecture is a footnote.
And the footnote version doesn't get a Smithsonian retrospective.
Which means we only see the pattern when it's expensive enough to be visible. The cheap versions happen constantly and nobody counts them.
The uncomfortable implication here is that the engineers weren't wrong.
They weren't. Not once. Concorde's wing was a masterpiece. Buran's automated landing was a genuine first. The Cray-1 was the fastest machine on earth. The SS United States still holds the record.
The failure was in the forecasting, not the engineering.
And forecasting is not an engineering problem. That's the sentence I'd put on the wall next to Colussy's.
So the next time you see a machine that represents the absolute summit of its paradigm, the question isn't how good it is. It's what mountain everybody else is climbing.
And whether they're already most of the way up it.
If you enjoyed this episode, please leave a review. It helps other listeners find the show.
It does. And it's the cheapest thing you can do for us.
Thanks as always to our producer, Hilbert Flumingtop, who is currently signing for a delivery.
He'll be back.
He's always back. This has been My Weird Prompts.
The human-AI collaboration podcast. We'll be back soon.