AI & SocietyOctober 9, 2026

Nobody Canceled the Iceman

Industries rarely disappear because a better rival wins or everyone goes bankrupt. They follow a pattern I call the Five Cs of disappearance: Capital, Compression, Container, Customer and Category. The last one is the hardest to see, and it is the one that finishes an industry.

Sahra-Josephine Hjorth in a fur-hooded coat and a leather apron carrying a block of ice with metal tongs in falling snow outside a wooden cabin

A small piece of cardboard hangs in the dining-room window of the house where John F. Kennedy was born. It has four numbers printed along its edges: 25, 50, 75 and 100. Today a visitor could easily walk past it, and anyone who stops will probably wonder what on earth it is for. In 1917, when Kennedy was born in that house, nobody on the street would have needed to ask. It was an ice card. For decades, families put one in the window and turned it so that the number of pounds of ice they needed pointed up, and the iceman read it from the street and chopped exactly that amount off the block.

That card was the order, the reorder and the renewal, all in one, and it kept an entire industry running. At its peak, the same New England trade that delivered ice to that window shipped ice as far as India. Then, within a single generation, the cards disappeared from the windows. Americans wanted more cold than ever, but they had stopped buying it from the iceman. They had bought a refrigerator, and the cold now came from a plug in their own kitchen.

Nobody canceled the iceman. The trade did not end with a rival’s victory or a single bankruptcy. It ended through five routes, which overlap and rarely arrive alone. I call them the Five Cs of disappearance:

Capital, Compression, Container, Customer and Category.

Over the past months, I have written about four of them in a series on EdTech, where I predicted that 90 percent of today’s EdTech companies would disappear, consolidate or become functionally irrelevant. My most recent essay was built around the fifth. It argued that the LLM is eating the internet in the same way that the iPhone ate the calculator, the paper map, the GPS and the digital camera.

Each C names something that leaves an industry: the money, the competitors, the form the product comes in, the biggest buyers and, finally, the decision to buy at all. The first two are moves by the market, made by investors and competitors in public. The last three are moves by customers, made in decisions nobody reports. That is why the numbers companies watch pick up the loud Cs first, and why the C that finishes an industry is the one nobody measures.

I spent twelve years selling software subscriptions, which is the modern version of delivering ice. Today I also own an AI studio that helps companies build what they used to buy, which is closer to selling refrigerators. I have been the iceman, and now I help sell the refrigerator.

No better ice company beat the ice companies

The standard explanation for the death of an industry is disruption. In Clayton Christensen’s theory, a smaller company with fewer resources finds a foothold either at the low end of a market, with a cheaper and simpler product, or in a new market of people who were not buying at all, and climbs from there until it can successfully challenge the incumbents. This is how we usually tell the big stories of business decline. Kodak lost to cheap digital cameras. Blockbuster lost to Netflix, which started as a small company mailing DVDs. Encyclopaedia Britannica, sold door to door for more than a thousand dollars a set, lost to Encarta, a cheap CD-ROM from Microsoft. Each time, the lesson people take away is the same: watch out for the competitor that looks too small to matter.

The ice trade does not fit that story. The refrigerator did not start small and cheap. It arrived at the top of the market. General Electric’s first mass-produced home refrigerator was called the Monitor Top, after the round cooling machine that sat on top of the cabinet, and it cost about $525 in 1927, more than a new Ford Model T. At the time, a hundred pounds of delivered ice cost a quarter. The refrigerator came from one of the largest companies in the United States. Most of what ended the ice companies happened without a competing ice company at all.

Capital leaves before the customers do

Capital is the first C: the money leaves before the customers do. Long before most households gave up their iceboxes, the money in the cold business had moved to machines. The United States had almost 800 ice plants by around 1900. In Chicago, 82 percent of the ice sold in 1911 was cut from lakes and rivers, and by 1922 only 10 percent was (Chicago Public Library). The harvesters still had customers, but the new money was going into the plants. That shift still happened inside the cold business, since the new plants sold ice too, but it showed where the money was heading: toward machines, and eventually toward machines that would make delivered ice unnecessary.

The same thing happened to EdTech, where venture investment fell by almost 90 percent in three years. It is now happening to traditional subscription software. AI is software too, but investors are moving their money toward the companies that build AI models and AI-native products, which took close to half of all global venture funding in 2025, and away from the subscription tools those products may replace. Capital buys a company the ability to adapt. It cannot buy a reason for the industry to exist, which is why the money leaving is a warning rather than the end.

Capital is also the loudest C. It shows up in funding reports, on investors’ websites and in the headlines that follow every fundraising slump.

Compression can make a company bigger while its industry drains away

Compression is the second C: the competitors leave. The work continues, but far fewer companies do it. The ice trade compressed fast. In 1899 Charles W. Morse combined New York’s ice companies into one giant firm, the American Ice Company, which controlled both natural and factory-made ice across much of the Northeast. In 1900 it raised prices ahead of the summer, and the scandal grew when it emerged that the city’s mayor held stock in the Ice Trust.

There is a popular version of the ice story, told in innovation talks, in which the ice harvesters failed because none of them became ice factories. Morse’s trust owned the factories too. It had the scale, the capital and both kinds of ice, and none of that changed where households were heading. Compression can make a company bigger, more efficient and more powerful while the reason for its whole industry drains away.

The Great EdTech Compression traced the same pattern through mergers and restructurings, and it is visible across software: the last quarter of 2025 alone saw 245 enterprise SaaS acquisitions worth $83.7 billion. In education, strategic acquisitions rose from 112 to 165 in the first half of 2026 compared with a year earlier, and the pace picked up again this autumn. In September alone, Pearson and McGraw Hill announced four acquisitions of AI and assessment companies between them, including TeachFX and Workera, and upGrad completed its purchase of Unacademy for about $200 million, less than a tenth of the $3.4 billion Unacademy was valued at in 2021. Compression is loud, because it produces press releases, front pages and, in Morse’s case, a political scandal, which is exactly why it draws more attention than it deserves.

The need survives the container

Container is the third C: the form leaves. The need survives, and customers still choose how to meet it, but they choose a different form. Households still wanted cold milk. They stopped wanting it delivered by wagon to a wooden box, and started owning the machine that made the cold. In 1930, 14 percent of American homes had a mechanical refrigerator. By 1940 it was 44 percent, and by 1950 about 80 percent (Cardia, NBER). The ice trade lost its customers while the market for cold kept growing.

Figure 1

The cold moved into the kitchen

14%

1930

44%

1940

80%

1950

homes with a refrigeratorhomes without one

How to read it. Each column is every American home in that year, drawn to the same scale. The pink part is the share with a mechanical refrigerator. In twenty years it went from one home in seven to four in five. The need for cold kept growing the whole time; it just stopped coming off the iceman’s wagon.

Source: Eleanor Cardia, NBER, as cited in the essay.

The iPhone story, which I told in The iPhone Killed the Calculator, is full of container shifts. People never stopped needing directions, sums, photographs or music. They stopped needing a separate object for each one. Music shows it most clearly. In 2025 the US recorded music industry earned a record $11.5 billion in wholesale revenue, and only $312 million of it came from CDs. The need for music has never been bigger, but the disc, the CD player and the record store it came in have almost disappeared. Where the need went next, into a product made by a company that had never been in the business, is the fifth C.

Learning is making the same move today, out of separate platforms and into ChatGPT, Claude and YouTube, as I described in EdTech Is Still Mailing DVDs. A container shift is easy to misread from the inside because demand looks healthy. The ice companies could point to growing cities, hotter summers and more food to keep fresh. Every one of those trends was real, and every one of them was flowing into a different box.

Container is quieter than it looks. It shows up in usage data, but only for a company that tracks where the need goes, and most companies track how often people open their own product.

The best customers start making their own ice

Customer is the fourth C: the biggest buyers leave to make it themselves. In the ice trade it came first. Breweries and meatpackers were the trade’s biggest buyers, and in Chicago the two industries together used about 350,000 tons of ice a year (Chicago Public Library). They were also the first to leave. Carl von Linde built his refrigeration machine for the Spaten brewery in Munich in 1873, and by the 1880s most breweries made their own ice mechanically. In 1896 the meatpacker Armour installed a 350-ton Frick machine that ran for forty years. Instead of shopping for a cheaper ice company, the best customers built their own cold.

The same thing is spreading across software, as I described in When Your Customer Becomes Your Competitor. In McKinsey’s 2026 State of AI survey, 32 percent of respondents said their organizations had decided against buying at least one software product or feature because they could build the functionality in-house with agentic coding tools. None of those decisions shows up in a vendor’s pipeline as a lost deal, because the deal never existed.

Customer reaches the most valuable accounts first, because they are the ones for whom building pays. A brewery buying thousands of tons of ice could justify a machine, while a family buying fifty pounds at a time could not. It shows up only as an absence, a pipeline that never forms, while the smallest customers still look loyal.

The category can be taken by a company that was never in it

Category is the fifth C: the decision leaves. The job is absorbed into something people buy for other reasons, until nobody makes the choice at all. One question separates it from Container: is anyone still choosing? The families who switched from the icebox to the refrigerator were choosing. They weighed the iceman against General Electric, an electrical company that had never harvested ice, run an ice plant or owned a single wagon, and they picked the appliance. The households formed after them never made that choice. Cold was simply something a kitchen had, and the iceman never came up.

The iPhone shows the same move more cleanly. The calculator, the camera and the GPS box each became an icon on a phone that people bought for other reasons. Most people today never decide against buying a separate camera, because the question never comes up. AI chatbots like ChatGPT and Claude are starting to do the same to software. People open a chat window for a dozen different reasons, and one product category after another becomes something they simply ask for there.

Category is the silent C. Container leaves a decision behind, a household choosing a new box. Category leaves nothing in the reports a company already reads, because nobody writes down why they never considered buying something.

Each C shows up on a different dashboard

Put side by side, the Cs run from loudest to quietest, and each leaves its evidence in a different place. The ice card was the trade’s dashboard, and it had four edges: 25, 50, 75 and 100. There was no edge for zero.

Figure 2

An edge for every order, none for zero

Today: 25 pounds

Families turned the card so the pounds they needed pointed up. The iceman read it from the street and chopped exactly that amount off the block.

The zero is not on the card. A household that bought a refrigerator simply took the card down.

How to read it. A drawing of the card in the window of the house where John F. Kennedy was born. The card turns through all four orders and comes to rest on 25. Every edge records an order of a different size. The dashed zero below it is the order the card could never show.

Source: National Park Service, Fresh Pond Ice Company card.

The Cs reached the ice trade in the worst possible order. A quiet one came first, as breweries built their own cold in the 1870s and 1880s. The loud ones, the money moving into ice plants and the Ice Trust, came in the middle and took the headlines. The quietest, a generation that never needed an iceman, came last and finished the job.

Figure 3

Each C leaves its evidence in a different place. Category leaves none.

loudestsilent

Tap a C to see how it showed up in the ice trade.

How to read it. The Cs run from loudest at the top to silent at the bottom, and each box fades as its C gets quieter. Capital and Compression make headlines. Container and Customer show up only if you go looking. Category, in pink, sits in an empty dashed box, because it leaves nothing for a company to measure.

Sources: Chicago Public Library; Charles W. Morse (Wikipedia); Cardia, NBER; Beer & Brewing, as cited in the essay.

Software companies are built to watch the loud ones. In February 2026, Bain noted that software companies were still keeping around 90 percent or more of the revenue from their existing customers, while software stock prices were down about 25 percent from their highs. The retention numbers looked healthy. The market was pricing in the customers who would never arrive.

Some of this really is disruption

The strongest objection is that this is Christensen with new labels, and part of it is. A container shift often looks like classic disruption: a new way of meeting the same need starts small, improves and takes the market. Streaming did that to the DVD. The Five Cs keep that theory and add the routes it was never built to describe. Money leaving, an industry consolidating, customers building for themselves and a category absorbed by an outsider involve no small rival climbing from the low end or from a new market.

A second version of the objection says that a category absorbed by an outsider is simply disruption by another name. In everyday speech it is, because the word now covers any change that ends an industry. But a word that covers everything stops telling you where to look. Christensen’s theory gave incumbents a precise warning: watch the product that starts small, cheap or aimed at people you do not serve. Category absorption arrives from the opposite direction, as an expensive product from one of the largest companies in the world, bought for reasons that have nothing to do with you. When the iPhone launched in 2007, Christensen himself predicted that its “probability of success is going to be limited,” because by his theory it was an improvement on existing phones rather than a disruption. He was comparing it with other phones. Years later he said it had disrupted the laptop instead, a new-market disruption in his own terms. But even that correction missed most of the damage. As I argued in my last essay, the industries the iPhone really emptied were the calculator, the camera and the GPS box on the dashboard, and none of them had ever considered a phone a competitor. TomTom lost a fifth of its value in one day in 2009, when Google put free navigation on the phone. The most respected theory of disruption in the world looked straight at the device that ended those categories, and it did not see them.

There is also a sharper difference. Christensen’s incumbents failed by listening too closely to their best customers. In the ice trade, the best customers were the first to stop calling.

Nobody cancels the iceman

In 1940, more than one in four Chicago households still bought ice (Chicago Public Library). Ten years later, about one Chicago home in twenty did. A family did not need to call the ice company to cancel. Taking the card down was the cancellation. They bought a refrigerator, and the next time the wagon came down the street, the card was not in the window. There was no complaint to record, no lost deal to analyze and no competitor to blame, only a window without a number in it.

Software is different in one respect. Its customers do cancel, and every cancellation is logged, analyzed and presented to the board as churn. But a cancellation only records that a customer left. It does not record where the customer went. The form offers price, missing features or a competitor, and has no box for the customer who simply asked an AI instead. Most of the loss never becomes a cancellation at all. The rest is the customers who never sign up: the company that builds its own tool instead, and the new business that never even considers buying one. That is how retention can stay above 90 percent while the market stops arriving.

Every software company has its own version of the card. It is the renewal that arrives on time, the login on Monday morning, the search that brings a buyer to the pricing page. I watched those signals for twelve years, and they are honest. They also only ever count the windows that still have a card in them. They tell you when a household stops ordering, never that it bought a refrigerator, and they say nothing at all about the children who will never learn what the card was for.