AI & EdtechSeptember 23, 2026

I Meant Ninety Percent

Close-up portrait of Sahra-Josephine Hjorth

What I expect to happen to EdTech companies by July 2029.

Industries rarely disappear because every company fails at once. They disappear because companies become divisions, products become features, categories migrate, and customers find a shorter route to the same outcome. The capability survives. The need for thousands of independent suppliers does not. That is what I expect to happen to EdTech.

On July 25, I predicted that 90 to 95 percent of today's EdTech companies would close, consolidate or become functionally irrelevant within two to three years.[1] The number sounds extreme because most people hear it as a bankruptcy forecast. They picture thousands of liquidation notices arriving on the same morning, look around at the EdTech companies still sending invoices, and conclude that I have confused a hard market with an ending. That is not how an industry disappears. It is not even how the industry everyone agrees the internet killed actually disappeared.

The travel agent did not die when the internet arrived

The familiar story is that the internet killed the travel agent. Travelers learned to book their own flights, the websites were cheaper, and the storefronts with posters of distant beaches in the window slowly emptied. That story is right about the outcome and wrong about the order in which it happened.

The first blow came from the airlines, not the browser. In February 1995, Delta capped the commissions it paid travel agents, and the rest of the industry followed with caps and cuts until, by March 2002, most airlines had stopped paying base commissions altogether.[2][3] Those commissions had been the agents' primary source of revenue.[5] The airlines had concluded, as one industry consultant described it at the time, that agents had become a cost to control rather than a necessity.[4] Delta did not announce the end of the travel agent. It announced a cost reduction, which is how the end of an intermediary usually sounds when it begins.

The internet arrived in an industry that had already lost its financial protection. Airlines sold directly on their own websites, online agencies aggregated what used to require a phone call, and travelers discovered they could reach the ticket without passing through anyone. Accredited agency locations in the United States fell from about 33,600 in 1995 to fewer than 13,000 twenty years later.[3][2] The number of travel agents fell from roughly 132,000 in 1990 to 74,000 in 2014.[5] Travel itself kept growing throughout.[6] People never stopped wanting to go somewhere. They stopped needing a separate company to get them there.

The more interesting part of the story is how the agencies disappeared, because most of them did not simply close. Many were sold or folded into larger networks. By 2001, firms with under $2 million in annual air sales made up 84 percent of agencies but sold only 14 percent of airline business, while less than one percent of firms sold 57 percent of it.[3] Some kept the storefront open, added service fees and lived on customers who had not yet found another route.[4] The agencies that survived changed what they sold. They moved away from issuing tickets and toward cruises, tours and complicated itineraries, where packaging several products together still required judgment a booking engine could not supply.[2] Order-taking agencies, as Travel Weekly later put it, had to die or reinvent themselves as advisors.[7]

That is the shape of what I expect for EdTech. It will look like closure, sale, consolidation, migration and a long tail of storefronts that stay open after the market has found a shorter route.

EdTech will fall faster than the travel agents did

The comparison has limits, and they cut against EdTech. Accredited agency locations fell by about 60 percent over twenty years, and the number of travel agents by about 44 percent over twenty-four. One analyst summarized the whole arc as a collapse that took ten years.[6] If EdTech simply followed the travel agents, ninety percent in three years would be wrong. I do not think it will follow them, because EdTech is exposed in ways the travel agents never were.

The internet shortened the route to the product. It did not replace the product. The airline still flew the plane, and no website could manufacture a flight. General-purpose AI does more than route the learner around the EdTech company. It produces much of what the company sold: the explanation, the practice question, the tutoring conversation and, increasingly, the course itself. Travel agents also had storefronts, personal relationships and corporate accounts that took years to unwind, while many EdTech products are software destinations a learner can abandon between one assignment and the next. And learners are not waiting a decade to change routes. OpenAI reported in 2025 that more than a third of college-aged young adults in the United States used ChatGPT, and that about a quarter of their messages were related to learning or schoolwork.[27]

The travel numbers also understate the collapse in the sense that matters for this forecast, because they count storefronts rather than independence. Between 1992 and 2002, accredited locations fell 23 percent while the number of agency firms fell 34 percent, as independent businesses were sold or consolidated into branch networks.[3] Even the industry's own trade press eventually stopped treating accredited locations as the measure of its health.[2] My forecast counts independent companies. By that measure, the travel agents disappeared faster and more completely than their storefront count suggests.

Every exit is closing at the same time

Travel agents could have survived the commission cuts alone. Many did for a while, by charging customers fees for what the airlines had once paid for. They might have survived the internet alone too, had they entered it with healthy margins and time to reinvent themselves. What they could not survive was both at once, because each blow closed the exit that would have softened the other. The commission cuts removed the money needed to adapt to the internet, and the internet removed the customers who might have paid the new fees.

EdTech is in the same position, with more pressures and less time. The money left first. Global EdTech venture investment fell from $20.8 billion in 2021 to $2.4 billion in 2024 and recovered only to $2.6 billion in 2025, while specialist investors widened their mandates until education became one theme among work, skills and human capital.[8][9] Then the demand moved. Learners did not stop wanting explanations and practice; they started getting them inside general-purpose AI instead of entering a separate learning product.[10] Then the purchase began to disappear, as institutions that once had no choice but to buy software found they could build the narrow piece they actually needed.[11] And the industry began to compress through mergers, absorption into larger platforms and direct bypass.[12]

An EdTech company could survive any one of these. A company that loses its investors can live on revenue. A company whose learners drift away can sell to institutions. A company whose institutional customers start building can raise money to out-build them. A company in a shrinking category can be acquired. What it cannot do is use those exits when all of them are narrowing together. That is why the number is so high even though many of these companies are well run. The routes that used to rescue weak and strong companies alike are closing at the same time.

I arrived at this from inside the system. For more than a decade I built CanopyLAB around the belief that adaptive learning would transform learning platforms, and I spent years advising customers not to build software they could buy. The technology I had been waiting for finally arrived, and the consequence was not the one I expected. It made many learning platforms unnecessary faster than it made them better. Today I also own an AI studio that helps companies build what they used to buy. I have sold the product, and I now help customers replace it.

EdTech was never the size of education

HolonIQ values the global education market at $7.6 trillion, with governments providing between 60 and 70 percent of the spending.[13] Education is a market, a public institution, a human activity and, depending on where you live, a right, and none of that is disappearing. EdTech is much smaller and much less stable. UNESCO found that estimates of the 2022 EdTech market ranged from $123 billion to $300 billion, a spread that says less about the market than about how differently people define it.[14] TIME and Statista reviewed more than 7,000 companies for their 2025 ranking, HolonIQ screened more than 10,000 for its Global EdTech 1000, and neither claims to be a census.[15][16]

The forecast concerns companies operating when I published whose primary commercial identity depended on learners or institutions entering, licensing or subscribing to a separate educational technology product. That includes learning platforms, courseware, tutoring, homework help, authoring, assessment, language and skills applications, corporate learning, classroom software and education-specific administration. It does not include schools or universities, OpenAI merely because people learn through ChatGPT, or an HR, consulting or productivity company in which learning has become one feature among many.

The commercial boundary is simpler than the list. If the company's core economics depend on learning continuing to require a separate software destination, it belongs inside the forecast. A travel agency whose business was issuing tickets belonged inside the equivalent forecast in 1995. The hotel did not.

Most of the ninety will never file for bankruptcy

Closure will be a substantial part of the forecast, but only part of it. This series has traced the other routes. In Your Edtech Investor Wants an Open Relationship, investors and then companies migrated out of the category, repositioning learning businesses as workforce infrastructure, financial platforms or general AI.[9] In Edtech Is Still Mailing DVDs, the demand stayed while the container changed.[10] In When Your Customer Becomes Your Competitor, the purchase itself began to disappear.[11] In The Great EdTech Compression, companies merged, specialists were absorbed into HR and consulting, and learners bypassed the vendor entirely.[12]

A company disappears in this forecast if it closes, loses independent control through merger or acquisition, moves its primary business outside EdTech, or stays legally alive after the need for its product as a separate purchase has gone. Each company counts once. An acquisition counts whether it is a rescue or a spectacular exit, because the forecast is about how much independent industry remains, not about who deserved to win or whether the investors got paid. The travel agency that was sold into a larger network did not go bankrupt either. It simply stopped being an independent business.

Functional irrelevance is the category most easily stretched, so it needs a narrow meaning. Falling valuation, layoffs or slower growth do not qualify on their own. The evidence has to show up as product withdrawal, sustained collapse in usage or renewals, the disappearance of new commercial activity, or survival mainly through legacy contracts after customers have moved to another route. Where the evidence is ambiguous, the company survives.

The baseline was already brutal

The familiar claim that 90 percent of startups fail is too loose to support a three-year forecast, because failure can mean closure, a disappointing exit or simply not producing venture returns. The official baseline is more useful and still severe. Of US private-sector establishments founded in 2013, 65.3 percent were gone ten years later, including 70.9 percent in the information sector and 61.1 percent in educational services.[17] More relevant is what happens after a business has already survived four years. Information-sector survival fell from 49.5 percent at age four to 36.8 percent at age seven, which means roughly a quarter of the four-year survivors disappeared over the following three years. In educational services the figure was about 23 percent.[17]

Those are ordinary rates in ordinary conditions. EdTech is entering the same three-year window after a funding bubble, a collapse in venture investment and a technological shock that changes both how products are made and how learners reach the outcome. Carta was already recording shutdowns accelerating year over year by 102 percent at seed, 61 percent at Series A and 133 percent at Series B, and warned that its confirmed closures undercounted the real total.[18] Closure has the strongest statistical anchor and will account for a large share of the ninety. The rest will come through the quieter routes the travel agents took.

More software, fewer companies

The obvious objection is that new EdTech products are appearing faster than ever, and they are. RevenueCat's 2026 analysis of more than 115,000 subscription apps found that monthly launches rose from roughly 2,000 three years ago to almost 15,000 today.[19] AI removed a production constraint. It did not create an equivalent increase in durable demand.

Only 4.6 percent of newly launched apps reached $10,000 in monthly revenue within two years. Apps launched before 2020 still generated 69 percent of subscription revenue, against 3 percent for apps launched in 2025 or later. AI apps converted downloads into trials better than other apps and then kept their subscribers worse, with median monthly retention of 6.1 percent against 9.5 percent.[19] That is not company-survival data. It describes the economy into which new EdTech products are being born.

Travel went through the same thing. The number of places where you could book a flight multiplied while the number of independent agencies collapsed, because building a booking site was never the scarce part. When almost anyone can create the product, creation stops being the gate, and distribution, trust, retention and outcomes take its place. The number of products can explode while the number of durable independent businesses contracts. Proliferation is not the opposite of compression. It may be one of its causes.

The prediction is not eccentric

In a 2026 survey of 621 enterprise leaders, 30 percent said their organizations were already reducing or reconsidering at least one software category because of agentic AI, and 86 percent expected to consolidate at least two platforms into a unified infrastructure layer within eighteen months.[20] Brett Queener of Bonfire Ventures predicts 75 percent fewer employee-facing application-software companies.[21] Norrsken VC predicts that 80 percent of today's AI startups will disappear.[22] Gartner estimates that agentic AI could expose $234 billion, about a fifth of enterprise application SaaS spending, to disruption by 2030.[23]

These numbers measure different things, and none of them proves mine. They show that forecasts of dramatic consolidation now extend well beyond EdTech. EdTech sits at the exposed end of that range because AI changes both sides of its market at once: how learning products are made and how learners reach the outcome.

The model provider walked onto campus

Seattle Colleges merged three Canvas environments into one, giving students a single entry point and giving instructors an assessment engine inside infrastructure the institution already runs.[24] That does not prove an assessment vendor was cancelled. It shows the precondition for aggregation: functionality that once justified a separate tool can increasingly live inside a system the institution already pays for.

The same shortening is happening above the traditional EdTech stack. OpenAI's campus-wide relationship with INSEAD gives ChatGPT Edu directly to students, faculty and staff, with a roadmap that includes AI-powered case studies, personalized learning, agentic workflows and measurable learning outcomes.[25] In the first essay of this series I argued that the everything app for learning would be the AI platform that already knows what the learner is working on, rather than a learning platform. This is that argument arriving at an institution. It is also the airline website of this transition. Nobody has to cancel a contract with a named incumbent for the route to change. The contract cancellation is not yet the evidence. The change in route is.

The bubble broke first

The strongest alternative explanation is that EdTech had a pandemic bubble and bubbles burst. That is true. Emergency demand, cheap capital, aggressive acquisitions and extraordinary valuations exposed weak companies when classrooms reopened and money became expensive, and some companies failed because of debt, governance or execution that had nothing to do with AI. If the same number of independent companies were now selling the same products at lower valuations, this would be a correction. If fewer companies, categories and commercial relationships are needed to deliver the capability, it is compression.

The travel agents lost their commissions before they lost their customers, and the order of those two blows is what made the second one fatal. EdTech lost its capital before general-purpose AI became a serious substitute. The bubble removed the financial protection. AI changed what could grow back.

Public education will be late

This process will not move at the same speed everywhere. A consumer can cancel a homework-help subscription today and ask ChatGPT tomorrow. A company can remove a marginal learning tool at its next renewal or build a narrow internal alternative. Public K–12 cannot move that quickly.

Schools and districts buy through budgets, tenders, approved-vendor processes and contracts. They must account for privacy, accessibility, child safety, integrations, training and political responsibility. Digital Promise has documented districts centralizing requests for proposals, vendor selection, pricing, contract negotiation and data-privacy agreements before a tool reaches the classroom.[26]

This creates residue. UNESCO reported that an average of 67 percent of US education-software licenses were unused and 98 percent were not used intensively. It also cited research finding that 85 percent of roughly 7,000 pedagogical tools, representing $13 billion of spending, were either a poor fit or implemented incorrectly.[14] In public education, lack of use does not automatically cause removal. The learner can stop depending on the product before the procurement system stops paying for it.

The same delay runs through the companies themselves. Public markets repriced the category first, because a listed company's value is tested every trading day: Chegg went from roughly $14.7 billion at its 2021 peak to about $100 million.[10] Private companies are not tested that way. They carry the valuation of their last funding round, renew multi-year contracts and report revenue that can continue long after the reason for the product has started to disappear.[1] The listed companies were the leading indicator. The private majority, which is most of what this forecast describes, will show the same change later.

Neither is an exemption from compression. Each is a delay between behavioral irrelevance and financial recognition, the same delay that kept the storefront agency open for customers who had not yet learned to book for themselves. Public K–12 will probably be the last place where the forecast becomes visible. The lag will look like survival.

What the booking engine could not sell

I expect the change to become visible between July 2028 and July 2029. It will not arrive as a wave of liquidation notices. It will arrive the way it arrived for travel agents: through closure, sale, consolidation, migration, absorption into platforms that used to be customers or suppliers, and companies that keep invoicing after the market has stopped needing them independently. Learning may grow throughout. Technology spending may grow. A handful of winners may become enormous, the way a few travel companies did.

But look at which travel agents are still here. It is not the agency that issued tickets fastest. It is the one that stopped selling the ticket. The survivors did not win the old game. They found the part of the need that the shorter route could not absorb.

I meant ninety percent. The remaining question is what earns a place in the ten.