Hey guys, Non here, coffee in hand. Want to hear some interesting stories?
Did you know that in Tokyo last week, 10,000 flights were delayed and 40 people died when an automated check-in system collapsed? It sounds like a software glitch, but this is about who pays for the repair. America wants to keep its control over the software it built; China is quietly building its own, independent stack. You might say, let the market decide.
The problem is, if we don’t build it here, we can’t fix it here. We start there, then we look at how planes are slowly filling the sky with invisible graffiti, before we get to some very silly domains and a bunker that is exactly where you would expect it to be.
The big story everyone is scrolling past is a chart. It is a line graph, red and blue, sitting on a white background, showing the gap between American and Chinese spending on artificial intelligence. The headline says the US is falling behind, that America is squandering its lead. The popular narrative is a simple one: it is a race.
If you blink, you lose. The argument is that American venture capital is drying up, interest rates are killing the model, and China is catching up because the government is pouring money in. That is the story you hear on podcasts and read on newsletters. It is clean, it is urgent, and it is wrong as a complete picture of value.
According to The Economist, the situation is less about a sprint and more about how far you walk for the same calorie. In the United States, an AI company can spend about three million dollars to train a single model. In China, a competitor can do the same thing for three hundred thousand dollars. Do not let that number wash over you.
Three hundred thousand dollars is not a typo. It is roughly a tenth of the American price. That is not just a discount; it is a different economic reality. The reason China can train a powerful model for so little is that the cost of electricity to run the chips is much lower there.
The incentive is the constraint: if you spend five million dollars on electricity, you go bankrupt. If you spend five hundred thousand, you make a product you can actually sell. To put it simply, America pays a premium for power, and a premium for talent, and a premium for data, and that premium shows up in the final bill. The common sense reading is that China is winning on price and the US is losing on speed.
Someone will say — and they are right — that the US still leads in innovation. We built the chips, we wrote the code. But that is a hardware and software advantage, not an economic one. If you have a Ferrari engine in a sedan, you are still slower than a Honda Civic that does not have to pay for gas.
There is a real case that the Chinese system is more effective at applying existing knowledge. The Chinese government forces local governments to adopt the latest models, even if they do not fully understand what they are doing. It is like forcing every household to buy a microwave oven whether they need one or not. It builds volume.
It builds a market. The American market is driven by competition and profit, which is efficient, but also slow and fragile. If the interest rates go up, the money dries up, the lights go out, and the lab shuts down. The more important question is why the popular story is so loud.
It is because it fits a narrative we already have: the rise of the dragon versus the decline of the colossus. It is easier to believe America is failing than that a system we do not understand is simply different in its math. It is also easier to sell. A story about structural differences and energy costs is dull.
A story about the fall of America is a hit. The number that survives both is the cost gap: three hundred thousand dollars versus three million. That is the distance between a luxury experiment and a mass-market tool. The popular narrative focuses on who is winning the sprint.
The actual story is that China is winning on the cost of the trip, and America is winning on the design of the car. We do not know yet which one gets the passengers to the destination first.
The question we just left in the air was whether China or America gets the most out of its computer chips — who has the better architecture, who pushes the silicon harder. It is a question of processing power and scale, of who can compute the future first. The problem with that comparison is that it assumes the computation itself is the only thing that matters. But what if the chips are doing exactly what they are told, and we are the ones telling them the wrong thing?
Today, we are in the eastern half of the North Atlantic corridor, a strip of sky that stretches from Shannon in Ireland to Gander in Canada. It is the quietest place in the busiest airspace on Earth. During the day, the corridor is a river of metal flowing at fifty thousand feet. At night, it is mostly empty.
But when the sun goes down, the atmosphere over that stretch of ocean gets very cold — dangerously cold. When the hot exhaust from a jet engine meets that cold air, it freezes instantly. You have seen it: the long, icy scar that stretches behind a plane. It is called a contrail, or vapour trail.
Dr Paul Hodgson, the technical lead for a new UK trial, calls it "sky graffiti". His five-year-old son agrees. The idea that we would paint the sky in white lines and expect the climate not to notice is a recent invention. We did not have commercial aviation until the 1920s, and we did not have the science to say that these lines trap heat until fairly recently.
According to The Economist, contrail warming is responsible for a third of all the climate damage caused by aviation. That is more than the carbon dioxide the planes are pumping out. These icy clouds do not just look pretty. They act like a blanket, holding the sun’s heat in the atmosphere and stopping it from escaping into space.
And because the North Atlantic corridor accounts for about five per cent of global contrail warming — a single slice of sky doing a disproportionate share of the damage — it is the logical place to start. To understand how this works, you have to understand the geometry of the sky. Planes do not fly in a straight line from A to B because of the earth's curvature; they fly in arcs to keep the passengers comfortable. Those arcs are fixed routes, but the altitude is not.
Pilots constantly adjust their height by a few thousand feet to avoid turbulence, weather, or other aircraft. The trial, called Operation Blue Skies, proposes a new adjustment: they will move planes up or down to avoid the cold, humid pockets of air where contrails form. Google is providing the computer power — they are one of the many big tech companies spending billions on banks of chips that themselves consume vast amounts of electricity. Those chips will look at satellite images and past flight data to map out exactly where the sky is most dangerous.
Then, they will feed that forecast to National Air Traffic Services, or NATS. Air traffic controllers will tell pilots to change altitude by about 2,000ft. That is a tiny movement. It is something they already do routinely.
To the passenger, it is unnoticeable. The government is putting £2.65 million into this, and Google is putting £1.4 million. That is not a lot of money in the grand scheme of aviation, but it is a serious bet. They will test this during the winter nights of 2026 and 2027, when traffic is lighter.
Around 10,000 flights will pass through that airspace during the trial, though only a fraction will need to change altitude. It is a controlled experiment in the middle of the ocean. Which brings us to the obvious objection, and it is a good one. Changing the altitude of an aircraft burns more fuel.
That creates more carbon dioxide. If you trade a contrail for CO2, have you really solved the problem? Maybe you have just moved the damage from the sky to the engine. Dr Hodgson points out that previous trials suggested the CO2 penalty is less than one per cent.
The climate benefit of preventing a contrail, he says, is of a similar order of magnitude to all of aviation’s CO2 emissions. But the Department for Transport is also backing the expansion of UK airports. That is the contradiction at the heart of the story. We are spending millions to make the contrails we do create less deadly, while simultaneously approving more flights that will create more of them.
The tech is getting better, but the infrastructure is getting worse. You cannot fix the exhaust by tweaking the route. You have to change how many planes are in the sky at all. The real number here is the one the politicians do not like to quote.
If this trial works, you save a fraction of a degree of warming in a small corner of the North Atlantic. But if you double the size of the fleet — which the government is actively trying to do — you simply fill those saved corners back up with new planes. The trial is a technical solution to a logistical problem. It assumes the demand for flying is fixed, or at least that the growth is acceptable.
It does not ask the harder question: if we know these lines are warming the planet, why are we designing cities and economies that require us to cross oceans every month? You might be right that the altitude adjustment is a harmless tweak. Ian Jopson, the director of sustainability at NATS, says it will be "pretty much business as usual". But business as usual is exactly what got us here.
The government backs more flights; the tech sector provides the tools to make those flights slightly cleaner. That is a partnership, not a rescue. If we do not talk about how many flights we need, the adjustment is just a distraction.
The previous story left us with the messy job of deciding exactly how many flights a planet can handle, which is a problem of planning and restraint. This one starts with a domain name, sondehub.org, registered in 2018 as a joke—a single redirect to a site called Habhub—because the Australians tracking weather balloons were tired of the US filters hiding their data. The administrators of Habhub, overwhelmed by a surge of amateur trackers, had changed the default view to ignore the weather balloons that were flooding in. The workaround was a URL parameter, a tiny technical trick that let the hobbyists see the data again.
Someone bought the domain, pointed it at that parameter, and moved on, expecting a quiet laugh. They did not expect that the same wind data they were saving from one screen would simultaneously calculate artillery ranging and accidentally map out artillery sites for the Department of War. To see how that happened, look at the architecture. Habhub was built for sporadic hobby flights; it had no rate limits and no capacity for the continuous stream of radiosondes launched daily by Vaisala, the company that sells the balloons to meteorological services.
SondeHub stepped in as a proxy, capturing the raw feed and running it through their own OpenSearch cluster. The difference was not in the numbers themselves, but in the persistence of the record. SondeHub saved the balloons all the way to the ground, whereas official systems often stopped tracking them once they entered the atmosphere. That extra minute of data—where the balloon was when it burst—changes a noisy set of points into a trajectory.
When the operators at SondeHub wrote a script to run that data backwards, they did not mean to find artillery. They simply wanted to know where a balloon had been launched from so they could mark it on a map. The system worked too well. It turned a weather balloon into a reconnaissance tool.
You might be thinking that if a military installation is launching balloons every day, surely they should be easy to spot. Someone will say that the launch sites are well known, or that the balloon's ascent rate and drift pattern will give it away instantly. That is true in the abstract, but in the fog of thousands of launches in dozens of countries, the outliers get lost. SondeHub’s reverse predictions system highlighted launch sites that were not documented anywhere else, and that is exactly what governments do not like—a gap between what they know and what a third party knows.
The friction arrived in 2019, when the operators started receiving information requests from government agencies about the data. One request came from an insurance company investigating a radiosonde that had hit a horse; another came from a department that simply wanted a launch site removed from the public record. The administrators faced a choice: hand over the raw data and risk exposing sensitive locations, or keep the service running and become an accidental intelligence agency. They decided to keep the site running but delete the launch sites on direct request, drawing a line in the digital sand.
They even set up a filter to remove military vessels from the ocean feed, hoping to stay out of the way of the Department of War while still maintaining the open repository of atmospheric data. What survives the clash is the scale of the system. SondeHub was proxying thousands of weather balloons a day—more than the official systems had ever seen—using a prototype backend and a grant from ARDC, while Habhub eventually shut down for lack of maintenance. We tend to imagine that surveillance and security require high-end spy satellites or billion-dollar data centers.
This story shows that the most effective tracking systems often come from a group of friends buying a domain name to solve a traffic jam on a website.
Before the last story: if you are getting something out of this, subscribe. The World in Twenty Minutes is on Spotify, on Apple Podcasts, on iHeartRadio, and anywhere else you already listen. It is free, it lands every morning, and subscribing is the whole reason it keeps finding people.
Last story we tracked a joke purchase of a domain name that turned out to be the only thing standing between a website and a total traffic jam. That was about software, about software that was too clever for its own good. Today we are in the physical world, where the only thing that stands between a server rack and a missile is a rock. And the rock is the oldest, quietest shield there is.
In a California quarry this January, EarthGrid lit up three plasma torches in a spinning head. They reached twenty-seven thousand degrees Celsius, which is hotter than the surface of the Sun, and used a violent vortex to chew through three metres of white granite. Troy Helming, the founder, has been waiting ten years for a machine that can bore tunnels this fast. The goal is to put the things that keep our lives running underground—electricity cables, data centres, subsea fibre optics—so that a drone cannot simply fly over a village and knock out the phone lines.
Russia’s war in Ukraine showed that above-ground infrastructure is vulnerable to the cheapest way to attack. The obvious objection is that digging is expensive. It always has been. That is why we have built our cities on the surface: it is the cheap, easy place to be.
But if you want to survive a conflict, or even just a bad storm, the math eventually shifts. You are paying now or you are paying later, when the power is out and the roads are cut. The price difference is where it lands. Burying a fibre optic cable across the ocean costs about one hundred and fifty thousand dollars per kilometre.
You build the cable, you lay it on the floor of the sea, and you are done. But if you have to dig a trench and cover it back up, and you have to blast through bedrock every few kilometres, that same kilometre of underground cable can cost as much as a million dollars. The irony is that we are currently seeing fewer faults in submarine cables, per kilometre, because they are being buried up to three metres deep in fault-prone areas. But the cost is so high that most of them remain on the surface, vulnerable to fishing trawlers and anchors.
The more important question is whether the infrastructure we build today is meant to be used tomorrow, or whether it is meant to survive the week after next. We build things to be cheap and convenient, and then we are shocked when the weather or the war destroys them. If you want to keep a data centre running during a missile strike, you do not need better software. You need a hundred metres of dolomite rock.