Stratechery · running since 2013; the sitemap lists 2,946 public post URLs, between 160 and 320 a year from 2013 to 2026 (plus two pieces backdated to 2009 and 2010) · still publishing at a high cadence, most recent post 22 September 2026 · one Article a week is free, three Daily Updates a week are for subscribers (per the author's About page); all 13 posts read for this page come from the free tier · a clear pivot in late 2022: the subject matter swings to AI, and the author's own concept index now lists AI first · this page is built on 13 posts spanning 22 September 2013 to 21 September 2026, including 4 from the earliest period (2013–2015) and at least 4 on each side of the AI pivot · selection follows the author's own Concepts index — this is him saying what to read
In October 2014, a tech commentator set out to explain why the New York Times was in decline. The evidence he reached for was the stock chart of a Taiwanese optics component maker.
The company was Largan Precision, then worth $9.6 billion, doing exactly one thing: camera modules for phones. Foxconn, which assembles the entire iPhone, was worth only five times as much. A small shop that makes one lens, against the giant that builds the whole device — a gap of five.
His point was Acer founder Stan Shih's "smiling curve": plot value against the stages of a value chain and you get a curve that turns up at both ends and collapses in the middle. Critical components on the left, systems integration and brand on the right; the middle — the part that screws the components together — gets ground down to single-digit margins by competition. Acer itself sat at the bottom of its own founder's curve.
Then he mapped the curve onto publishing. Google and Facebook on the right, writers on the left, and newspapers in the middle.
There simply isn't that much money in the middle.Publishers and the Smiling Curve, 28 Oct 2014
Something else had happened that same week. German publishers had banded together to demand that Google pay for the article snippets in its search results. Google's response was not to pay but to remove the snippets. Traffic fell so hard that within days the publishers came back asking Google to restore them, no compensation required. Everyone's takeaway was that Google had proved its value to publishers. His angle was pointed the other way:
the publisher's demonstrated that they provide no value to their writers.ibid.
The writer is Ben Thompson, an American who spent a decade writing from Taipei before relocating back to the United States in 2025. He started Stratechery — Strategy plus Technology — in 2013, went full-time on it in 2014, and has published close to three thousand posts since.
In July 2015 he looked back at his own last three pieces — on Airbnb, on Netflix, and on why web pages had become so unbearably slow — and realised they were one piece.
In retrospect, there is a clear thread. In fact, I believe this thread runs through nearly every post on Stratechery, not just the last three. I am calling that thread Aggregation Theory.Aggregation Theory, 21 Jul 2015
The thread fits in a paragraph, and it is probably the most-quoted paragraph in a decade of tech commentary.
Before the Internet, every consumer value chain ran suppliers → distributors → users, and the money stopped at the distributor, because distribution was scarce. A newspaper owned the delivery routes of a city, so it integrated backwards into making the content. Broadcast networks, book publishers, taxi companies, hotel chains — all the same shape: control distribution, then reach back and lock up supply. Why backwards? Because there are always vastly more users than suppliers, and in a world where transactions cost something, owning the smaller side is far more leverage.
The Internet turned that upside down. Distribution of digital goods became free, neutralising the distributor's old advantage; transaction costs went to zero, which meant a company could, for the first time, own hundreds of millions of customers directly and at scale.
This has fundamentally changed the plane of competition: no longer do distributors compete based upon exclusive supplier relationships, with consumers/users an afterthought. Instead, suppliers can be commoditized leaving consumers/users as a first order priority.ibid.
Which leaves user experience as the only thing still scarce: the best experience wins the most users; the most users force suppliers to show up on the aggregator's terms; more suppliers make the experience better still. A virtuous cycle that eats the whole market.
Three years later he added a test for it, borrowed from Bill Gates. In 2007 Facebook had just held its first developer conference and was loudly calling itself a platform; Gates, at a funding meeting, was unimpressed (the line comes to us via Chamath Palihapitiya, who ran Facebook Platform at the time):
A platform is when the economic value of everybody that uses it, exceeds the value of the company that creates it. Then it's a platform.Bill Gates, quoted in The Bill Gates Line, 23 May 2018
From which Thompson wrings the sharpest distinction in his whole body of work: a platform facilitates the relationship between third-party suppliers and end users; an aggregator inserts itself into that relationship and controls it. Windows and AWS are platforms — Microsoft used to brag about capturing only a minority of the value in its own ecosystem. Google and Facebook do not merely fall below the line; they do not appear on the chart at all.
The sharpest application of the knife was Yelp. Yelp was pushing Google to stop filling its answer box with Google's own reviews and to use Yelp's instead. Thompson's reply: the thing you are asking for will only make Google stronger. Yelp's real problem is not whose data fills the answer box but the answer box itself — the more useful Google is, the less viable Yelp's business becomes, even in a world where every word in the box came from Yelp. He let Yelp's own CEO deliver the verdict:
If I were starting out today, I would have no shot of building Yelp.Jeremy Stoppelman, quoted in The Bill Gates Line
But remember only Aggregation Theory and you have read half this mind. The blade he actually returns to is older and more general, and it is five words long: your business model is your destiny.
In 2013 a recent business-school graduate published a piece titled "What Clayton Christensen Got Wrong". Christensen was the father of disruption theory and Silicon Valley's favourite professor, and between 2006 and 2012 he had predicted three times that Apple would lose — the iPod would be commoditised, the iPhone would not succeed, modularity would eventually do its work. The flaw Thompson found was inside one of Christensen's own examples: "You also see it in aircrafts and software, and medical devices, and over and over."
That is the problem: Consumers don't buy aircraft, software, or medical devices. Businesses do.What Clayton Christensen Got Wrong, 22 Sep 2013
Low-end disruption requires three assumptions: buyers are rational; every attribute that matters can be documented and measured; modular suppliers can become good enough on all of them. All three hold in enterprise procurement — a CIO has to justify a purchase, which usually means a feature list against a price. None of them holds in consumer markets. And Christensen's sample was almost entirely enterprise buyers.
The killing blow had been planted three years earlier, in a paper he wrote as a student:
it is impossible for a user experience to be too good… has anyone turned to an "inferior" product because the better one was too enjoyable?quoted in What Clayton Christensen Got Wrong
(A footnote adds: "Much to my delight, Christensen said I did a 'good job'.")
That instinct — ask who is paying and what their ledger looks like before you ask anything else — kept growing. In 2018 he used Bezos's "divinely discontent" customers to overturn his own earlier version: he had pictured experience as an asymptote, but in fact expectations themselves keep rising, so you can never overshoot, which makes it a perpetual business rather than a finite one. In 2016 he had noticed Google's "I'm Feeling Lucky" button and pointed out that the button guaranteed that search would make Google no money at all, because a user who doesn't choose among results never gets the chance to click an ad. In 2016 that was a pretty observation. After ChatGPT it became an existential one: AI is the ultimate expression of "I'm Feeling Lucky." Where do the ads go?
In the same piece he said something almost nobody was willing to say: Kodak got it right. Kodak invented the digital camera and correctly judged that it would not threaten film for twenty-five years; it collected profits and paid out billions in dividends in the meantime, and when it went bankrupt in 2012 it was because consumers had something better.
The idea that this is a happy ending is, to be sure, a contrarian view: most view Kodak as a failure, because we expect companies to live forever.AI and the Big Five, 9 Jan 2023
By September 2026 the same blade had swung at an entirely new target: the moral rhetoric of AI safety. Same move — not "is this claim correct" but "what does the ledger of the person making it look like". He takes Anthropic's call to pace the frontier and decomposes it into four overhangs — capability, product, pricing, capital. Slowing down happens to relieve every one of them.
Anthropic is fine with Anthropic being in the lead; anyone else requires government intervention.Frontier Overhangs, 21 Sep 2026
What is worth taking from this mind is not the conclusions — you can disagree with those, and people have, forcefully. It is the order of the questions: before asking whether a company is right, or whether a statement is decent, ask where the money comes from. Answer that and who will do what, who won't, and who could but definitely won't, mostly sorts itself out.
Incidentally, he has turned the blade on himself. Subscriptions are the vast majority of his income, deliberately, so that his incentives align with his readers'. And he admits that for over a decade he has analysed and advocated for advertising as a business model — a model he has never once used.
Glossary (the handful of prerequisites this page leans on, one line each)
On 28 October 2014, an article about the publishing industry opened with the name of a company most English-language readers had never heard of: Largan Precision (TPE:3008).
The iPhone 6 had just launched, and the single thing the launch event spent the most time on was the camera. Only a handful of companies in the world can build a phone camera like that, and Largan is one of them; it had supplied Apple for several models. The article carried Largan's all-time stock chart and two numbers: Largan was worth $9.6 billion, and Foxconn, which actually builds the entire iPhone, had a market cap only five times greater.
That is a strange ratio. Foxconn employs hundreds of thousands of people and assembles the whole device. Largan makes one lens.
His explanation was Stan Shih's smiling curve: plot value-added across the chain and it turns up at both ends and sags in the middle. Critical components on the left (Intel, Microsoft, Largan), systems integration and customer relationships on the right (IBM, Accenture), and in the middle the people who bolt the parts together. Acer, whose own founder drew the curve, is the epitome of the bottom — it and the other Windows OEMs competed each other down to single-digit margins.
There simply isn't any money in the undifferentiated middle.Publishers and the Smiling Curve, 28 Oct 2014
Then he laid the same curve over publishing: aggregators on the right, writers on the left, and the publishers in the middle. He supplied the market caps of the day: Google $369.7bn, Facebook $209.0bn — and the most august publisher of all, the New York Times, at $2.03bn.
He also had a natural experiment, cruelly clean, from that very week. German publishers had sued to force Google to pay for article snippets. Google didn't pay; it removed the snippets. The traffic drop was severe enough that within days the publishers came crawling back asking for the snippets to be restored, compensation waived.
Everyone read this as Google proving its value to publishers. He read it the other way round:
The general takeaway is that Google proved it was adding value to the publishers, but I have a different angle: the publisher's demonstrated that they provide no value to their writers.ibid.
An article about newspapers, evidenced by a Taiwanese lens maker, concluding that newspapers and Acer are the same kind of thing. That is how this mind works: find the shape of the skeleton, then go looking for the same skeleton in any industry at all.
Ben Thompson, American. Undergrad at the University of Wisconsin, an MBA from Kellogg in strategy and marketing, plus an MEM from the McCormick engineering school in design and innovation with a human-centred-design emphasis. Three employers before this: Apple (as an intern), Microsoft, and Automattic — strategy, developer relations and marketing for Apple University, Windows and WordPress.com respectively.
He started Stratechery — Strategy plus Technology — in 2013, and it became his full-time job in 2014. He lived in Taiwan for many years and wrote most of this body of work from Taipei: one person, one site, one subscription, relocating back to the United States only in 2025.
One detail says everything about what the job became. His About page lists his speaking rates in the open — compensation "ranges from $400k to $1m" depending on customer, event and topic — immediately followed by "I do not do any consulting at this time." The same page notes subscribers in over 85 countries.
What follows the rates is more worth remembering. He holds no individual stock in any company he writes about; he pays his own way to company events; since 2016 he has taken no speaking or consulting arrangements with companies he covers (the three that preceded that date are each named). And then this:
Subscriptions are the vast majority of my income; this is purposeful on my part as I want my incentives to be aligned with my readers'.About / Ethics Statement
Hold that sentence until section six. This man's life's work is using "where does your money come from" to explain "what you will do" — he applies it to Apple, to Google, to newspapers and to AI labs, and he has written it into his own masthead.
The site is still going hard: 2,946 public post URLs on the sitemap, between 160 and 320 a year from 2013 through 2026, the most recent as this page was written dated 22 September 2026. One thing must be said plainly, though: of those three thousand posts, only the weekly Article is free; the three Daily Updates a week belong to subscribers. All 13 posts read for this page come from the free tier — which is also, conveniently, where the load-bearing concepts live, by his own reckoning in the Concepts index.
Spread thirteen years of writing out and what he keeps returning to is not "which company wins" but something lower down, drier, and far more explanatory:
On any value chain, why do the profits settle at one stage rather than another, and what makes them move?
He did not run into this question in a library. He ran into it in a classroom. He has written that diving deep into Christensen in a corporate innovation class was "a breath of fresh air from management theory that explained all of corporate America but for its most successful company."
That company was Apple. And Apple was precisely Christensen's bugaboo: in 2006 he predicted the imminent demise of the iPod; in 2007 that the iPhone, being sustaining rather than disruptive relative to Nokia, would not succeed; in 2012 that "the transition from proprietary architecture to open modular architecture just happens over and over again… so I worry that modularity will do its work on Apple."
Faced with that, a fresh graduate normally has two options: trust the theory, or trust his eyes. Thompson took a third, and the choice more or less defines the next thirteen years: he did not go looking for counterexamples. He went looking for where the theory's sample came from.
That is the starting point of the first idea below. The other half of the root question — what makes profits move — grew in 2015 into the thing he is best known for. Put together they are the skeleton of the entire site: locate which stage the money stops at, identify the structure pinning it there, then ask whether some technology can remove that structure in a single stroke.
In July 2015 he did something writers rarely do: he reread his own last three pieces and discovered they were one piece.
Those three were about Airbnb (the sharing economy commoditised trust, which made a business of aggregating resources and owning the customer relationship viable), Netflix (which placed the same thing inside Christensen's Conservation of Attractive Profits: profits accrue to whoever integrates in a value chain, and they move when someone modularises the incumbent and integrates elsewhere), and a piece about why programmatic advertising had made web pages so slow — whose conclusion noted, almost in passing, that the way ad networks were commoditising publishers fit the same frame.
In retrospect, there is a clear thread. In fact, I believe this thread runs through nearly every post on Stratechery, not just the last three. I am calling that thread Aggregation Theory.Aggregation Theory, 21 Jul 2015
Start with the old world. Any consumer market splits into three: suppliers, distributors, users. To earn outsized profits you either take a horizontal monopoly in one part or integrate two of them into a vertical advantage — and before the Internet the latter almost always depended on controlling distribution.
Printed newspapers were the primary means of getting content to readers in a geography, so newspapers integrated backwards into content creation and earned their profits on advertising. The identical shape repeats everywhere: book publishers (distribution plus control of authors), video (broadcast availability plus content purchasing), taxis (dispatch plus medallions and cars), hotels (brand trust plus vacant rooms).
Note that the distributors in every one of those industries integrated backwards. His reason is blunt: there have always been far more users than suppliers, so in a world where transactions are costly, owning the supplier relationship is much more leverage.
The Internet did two things that turned the dynamic on its head:
First, the Internet has made distribution (of digital goods) free, neutralizing the advantage that pre-Internet distributors leveraged to integrate with suppliers. Secondly, the Internet has made transaction costs zero, making it viable for a distributor to integrate forward with end users/consumers at scale.ibid.
The plane of competition changes completely: distributors no longer compete on exclusive supplier relationships, suppliers become the thing that can be commoditised, and users become the first-order priority. Which leaves exactly one decisive factor — user experience. Best experience wins the most users; the most users draw the most suppliers, on the aggregator's terms; more suppliers improve the experience. A winner-take-all loop.
The original article runs seven pairs of examples in a row, each two lines long, formatted with almost gymnastic regularity: Google modularised individual pages and articles out of publications; Facebook modularised advertisements by letting advertisers target people directly rather than by proxy; Amazon modularised book distribution; Netflix modularised broadcast availability into any time, any order; Uber modularised fleet management into independent drivers; Airbnb modularised vacant property by building a reputation system for trust.
He singles out the last two. Rooms and cars were never digitised, yet they were disrupted. The reason is that every industry conceals a critical differentiator, and the moment that differentiator can be digitised, competition moves to user experience — which is exactly where new entrants are strongest and incumbents dullest. His prediction at the time: nearly every industry will belatedly discover it has such a function.
Two years later he came back and shipped a version upgrade. The 2017 piece "Defining Aggregators" opens with an unusually candid note:
(Note: this is not a typical Stratechery article; there is no over-arching narrative or reference to current news. Rather, the primary goal is to provide a future point of reference)Defining Aggregators, 26 Sep 2017
He gives aggregators three necessary characteristics; miss one and you are something else. A direct relationship with users (payment-based, account-based, or merely habitual). Zero marginal cost to serve them (no cost of goods, no distribution cost, no transaction cost). Demand-driven multi-sided networks with decreasing acquisition costs — the prettiest of the three: an ordinary company's customer acquisition costs rise, because the earliest customers are the best product fit and each subsequent cohort yields less surplus until it goes negative; an aggregator's fall, because marginal users are drawn in by the accumulating supply itself.
He then grades aggregators by their relationship to supply. Netflix is Level 1: it has to buy its content, which makes it slower to build and more precarious. Uber is Level 2: it owns no cars, but putting drivers on the platform costs time and money (background checks, vehicle verification), which caps supply growth and therefore demand growth. Google is the archetypal Level 3: suppliers — websites — are not only accessible by default but actively make themselves easier to find, to the point of sustaining an entire industry called SEO. Social networks are Level 3 too: initial supply comes from the users themselves, and once enough attention pools there, professionals contribute content for free.
Above all three sits the Super-Aggregator, with zero marginal costs on all three sides — users, suppliers and advertisers. There are exactly two in the world: Facebook and Google.
In 2018 he added the last piece, and the most usable tool in the set. He was working out what separates a platform from an aggregator when he remembered the Gates line (see the Essentials). What Gates supplies is a testable quantity: does the economic value created by everyone in the ecosystem exceed the value of the company that created it? By that test Windows is a genuine platform — Microsoft used to brag about capturing only a minority of the total value of its ecosystem — and AWS and Azure are its direct successors. Apple's App Store has all the trappings of a platform but sits clearly below the line: Apple captures the great majority of the ecosystem's value, which is why so few strong, durable businesses grow on it. Not an accident; an outgrowth.
And Google and Facebook? They do not even register on the graph.
platforms are powerful because they facilitate a relationship between 3rd-party suppliers and end users; aggregators, on the other hand, intermediate and control it.The Bill Gates Line, 23 May 2018
The sharpest application was Yelp, then filing a fresh EU antitrust complaint arguing that Google's answer box unfairly used only Google's own ratings and reviews and ought to use Yelp's. Thompson's response: what you are asking for will only make Google stronger.
Three layers to the argument. First, the answer box originally did scrape sites like Yelp, and Google stopped in 2013 under pressure driven partly by Yelp's own complaints. Second, in the years since — by virtue of sitting on top of search results — Google's own reviews have grown substantially; user-generated content is the ultimate low-cost supply, and losing that supply to Google is a bigger problem for Yelp than any advertising revenue at stake. Third, and decisively: Yelp's enemy is not whose data fills the answer box, it is the answer box. He puts two versions of a results page side by side and asks which is better for Yelp's business. Obviously the one without an answer box, where the user has to click through — even in a hypothetical world where the box is 100% Yelp data.
The more useful Google is, the less viable Yelp's business becomes. That is exactly what you should expect in an aggregator-dominated value chain: aggregators disintermediate suppliers completely and reduce them to commodities. Hence his strategic advice runs the other way — the best strategies avoid Google and Facebook entirely. Look at Amazon, which had just stopped buying Google Shopping ads, which it can afford to do because half of shoppers start their product searches on Amazon.
This idea predates Aggregation Theory by two years and is more general. It was born in an argument that looked hopeless.
September 2013: Apple had just shipped the iPhone 5C — a new shell at an old price point, sitting exactly where the discontinued prior flagship would have gone, $100 below the new one. Analysts were aghast. How could Apple not produce a genuinely low-priced iPhone? Wasn't this begging to be disrupted?
Behind all that aghastness stood one man, and Thompson wrote a piece titled simply "What Clayton Christensen Got Wrong". His opening posture is worth noticing: he first separates Christensen's two theories and declares that he fully subscribes to the first (new market disruption, from the disk-drive paper and The Innovator's Dilemma). Only the second — low-end disruption — is up for demolition.
And then he doesn't reach for a counterexample. He asks a question one level down: where did this theory's sample come from?
The flaw is in Christensen's own 2006 sentence: "You also see it in aircrafts and software, and medical devices, and over and over."
That is the problem: Consumers don't buy aircraft, software, or medical devices. Businesses do.What Clayton Christensen Got Wrong, 22 Sep 2013
Low-end disruption presumes three things at once: buyers are rational; every attribute that matters can be documented and measured; modular providers can become good enough on all of them.
He takes them in turn. First, business buyers really are extremely rational — a CIO must justify a software purchase, and the justification usually reduces to a feature list weighed against a price; whatever scores best wins. Consumers are not like that, though he prefers a different framing: consumers are perfectly rational, and it is our definition of rationality that needs to expand far beyond what is easily quantified. Second, the attribute consumers value most (assuming the product is in the vicinity of a need) is ease of use, and ease of use can only be felt, not bullet-pointed — the iPhone had a better touchscreen at lower resolution, bigger camera pixels but fewer of them, better performance-per-watt with fewer cores and lower clocks. Third, modular products, by prioritising standardisation and interconnection, necessarily fall short on attributes tied to the experience of use.
The decisive blow had been buried three years earlier in a paper he wrote as a student:
it is impossible for a user experience to be too good. Competitors can only hope to match or surpass the original product when it comes to the user experience; the original product will never overshoot (has anyone turned to an "inferior" product because the better one was too enjoyable?).Apple and the Innovator's Dilemma (2010), quoted in the above
"Overshoot" is borrowed straight out of disruption theory, and he uses it to locate a blind spot: the entire theory rests on the premise that a product can get too good, and there is one dimension where that never happens. Apple had been standing in that blind spot for fifteen years.
A footnote adds, charmingly: "Much to my delight, Christensen said I did a 'good job'."
Five years later he demolished and rebuilt his own idea. In May 2018, arguing about whether Apple or Amazon would reach a trillion dollars first, he quoted Bezos's annual letter:
One thing I love about customers is that they are divinely discontent. Their expectations are never static — they go up. It's human nature. We didn't ascend from our hunter-gatherer days by being satisfied… yesterday's 'wow' quickly becomes today's 'ordinary'.Jeff Bezos's 2017 shareholder letter, quoted in Divine Discontent, 2 May 2018
And conceded that he had thought too narrowly:
When I wrote that piece five years ago, I was thinking of the opportunity provided by a focus on the user experience as if it were an asymptote: one could get ever closer to the ultimate user experience, but never achieve it.Divine Discontent, 2 May 2018
Reality is better than an asymptote: the line of expectations itself keeps moving up. Today's amazing is tomorrow's table stakes. Which inverts the conclusion into something near a paradox — if a company defines itself as delivering the best possible experience for consumers, it will never achieve its goal, and that is precisely what makes it a perpetual business. He calls Amazon, on that basis, a sort of perpetual motion machine.
Meanwhile the same follow-the-ledger instinct was fermenting on another track. At Google I/O in 2016, Sundar Pichai declared the world was moving from mobile-first to AI-first and introduced the Google Assistant. What Thompson wrote then he has since re-quoted at least three times:
the "I'm Feeling Lucky" button guaranteed that the search in question would not make Google any money. After all, if a user doesn't have to choose from search results, said user also doesn't have the opportunity to click an ad… Google Assistant has the exact same problem: where do the ads go?Google and the Limits of Strategy (2016), quoted in AI and the Big Five
In 2016 that was an elegant observation. In January 2023 it became a matter of survival.
After ChatGPT he wrote "AI and the Big Five", going company by company and asking whether AI was sustaining or disruptive for each. The test is not technology; it is whether the business model lines up. Microsoft fits best: it sells subscriptions, so adding capability for a fee is natural, and it is OpenAI's exclusive cloud. What about Bing? His image: Bing "is like the Mac on the eve of the iPhone" — it contributes real revenue but a fraction of the leader's, so risking that model for a shot at massive share is a bet well worth making.
Google is the opposite. It invented the transformer, is rumoured to have a chat product far superior to ChatGPT, and has nothing on the market. Why? Because search has from day one handed the final verdict to humans — Google offers links, the user decides by clicking, and advertising is priced per click. AI is the ultimate form of "I'm Feeling Lucky": it just answers.
And here he does something few are willing to do: he defends Kodak. In 1997, the year The Innovator's Dilemma was published, Kodak's stock peaked at $94.25 and the company was not technologically behind — it had invented the digital camera. The problem was the business model: film carried very good margins, and digital cameras needed no film. So Kodak's management was strongly incentivised to convince itself that digital was for amateurs and would stay that way until it got drastically cheaper, which would take a very long time.
Then the key line: Kodak's management was right. It took over twenty-five years from the digital camera's invention for digital sales to surpass film, and longer for professional use. Kodak made a lot of money in the meantime and paid out billions in dividends. It went bankrupt in 2012, and the reason was that consumers had access to better products.
The idea that this is a happy ending is, to be sure, a contrarian view: most view Kodak as a failure, because we expect companies to live forever.AI and the Big Five, 9 Jan 2023
This isn't apologetics for anyone; it is prying out a hidden premise — that companies ought to be immortal — and holding it up to the light. And he states his own position, including the bet he has already lost:
I'm not calling the top for Google; I did that previously and was hilariously wrong. Being wrong, though, is more often than not a matter of timing.ibid.
Two years later he patched the framework again — this time with a limit on the framework itself. In 2025 he had to explain something it could not: if AI is disruptive to Google, why has Google invested earlier and harder than anyone?
His answer goes back to the button. "I'm Feeling Lucky" was unclickable by 2010 — instant results meant you never got the chance — yet it stayed on the homepage for years. That, he argues, was not laziness but a statement of purpose: AI is the ultimate manifestation of "I'm Feeling Lucky", and it is why Page and Brin started the company in the first place.
business models matter, but they aren't dispositive.Tech Philosophy and AI Opportunity, 8 Jul 2025
A man whose calling card is "your business model is your destiny", publishing the words "but they aren't dispositive" — this page thinks that is the single most worthwhile move to steal. The second axis he supplies is tech philosophy, from his 2018 piece "Tech's Two Philosophies". One camp holds that computers should do things for you (Google, Facebook); the other that computers should let you do your own work better (Microsoft, Apple). The prophet of the second is Steve Jobs, whose analogy was the bicycle: humans rank unimpressively for efficiency of locomotion, about a third of the way down the list, until you test a human on a bicycle, who blows away the condor and goes clean off the top. Plot opportunity against philosophy and Apple's stinginess on AI, Meta's hiring frenzy and Google's charge ahead all land in their proper quadrants.
The third thing is not a conclusion but a move, and one he has been leaning on harder every year: when someone makes a normative claim — what counts as good journalism, who owes whom, whether we ought to slow down — don't start by arguing the merits. Start by opening their income statement.
The first target was his own trade.
In February 2015 Andrew Sullivan shut down his blog and "blogging is dead" became conventional wisdom overnight. Vox's Ezra Klein wrote that "we're getting better at serving a huge audience even as we're getting worse at serving a loyal one," and declared that Sullivan had proved blogging "resistant to scale".
Thompson replied four days later, and pointed out the delay itself: by Sullivan's own definition of blogging — "daily, hourly, always-on-deadline stress" — spending a weekend with his family while he turned the debate over in his head meant he was doing it wrong.
His rebuttal is a distinction: Klein had confused scale with reach.
it cost him the same amount of both time and money to serve 1,000 subscribers as it would have to serve 100,000, or 1 million, and he didn't need to change a thing about himself or his content to do it. No, it's not scale that is the problem, but rather reach.Blogging's Bright Future, 2 Feb 2015
And limited reach is the conscious price of a subscription:
Stratechery, though, serves a niche, and niches are best served by making more from customers who really care than from milking pennies from everyone.ibid.
Then the key that unlocks the whole essay, a conclusion he says he first reached writing about newspapers: your business model is your destiny. An advertising model demands huge inventory, large audiences and massive targeting data — and "huge", "large" and "massive" are antithetical to something defined by one individual. Vox, chasing reach, asks its readers only to tolerate posts they don't care about, so no one should be surprised the site experience keeps getting more annoying. He asks for money instead, and is therefore heavily motivated to deliver value: don't waste the reader's time, quality over quantity, wait four days to write something considered instead of trying to be first.
Six years on he swung the same blade harder. In 2021, a16z announced it was building its own media property and journalists were not amused; a Washington Post columnist said the industry needed more independent views, not more sycophancy. Thompson agrees the conflict of interest is real, but says the complaint conceals a larger defect belonging to journalists themselves. He quotes The Verge's editor-in-chief Nilay Patel describing his own position — "I have almost nothing to do with the revenue of The Verge… I'm very insulated from sales" — and then lands this:
After all, journalists don't know how their businesses work, and they are proud of it!Publishing is Back to the Future, 27 Jan 2021
He does not treat this as a moral failing. He treats it as a luxury that a particular structure used to pay for — and that stopped paying long ago. The evidence is Warren Buffett's 1991 shareholder letter: media properties used to possess the three characteristics of a franchise (needed or desired, no close substitute, not price-regulated), and a franchise tolerates mismanagement; but by 1991 choice had multiplied and media was degrading from franchise into mere business. Note the year: the World Wide Web did not yet exist.
Which makes Australia's law requiring Google and Facebook to pay for news, in his reading, legislation built on a myth:
This is what happens when you don't understand how your own business works: you create a myth wherein Google and Facebook decimated the news business, when in reality they came along years after the business — already not a franchise — had been decimated by the Internet.ibid.
He adds an even less forgiving line: journalists never realised that people were paying for paper, newsprint and delivery trucks rather than for their reporting, and that advertisers were paying for the people.
In September 2026 the same move landed on an entirely different subject: AI safety.
The occasion was Anthropic and CEO Dario Amodei's insistence that we must pace the frontier. Thompson states his position first: he rejects the philosophical premise that grants equivalent moral weight to every being who may ever exist, which he says tilts the scales absurdly towards safetyism; he also says the discourse "taps into the psychology of religion, where dissent is not brooked and questioning the premise is heresy", and declares himself a heretic.
But the actual move is the next sentence:
"Pacing the Frontier" is framed as — and I believe motivated by — a concern about safety, but it also happens to address several distinct problems faced by the frontier labs.Frontier Overhangs, 21 Sep 2026
Note the clause "and I believe motivated by". He is not alleging hypocrisy. He is saying something more awkward: when a moral claim happens to point in the same direction as every one of the claimant's commercial interests, the coincidence is itself evidence that belongs on the table.
He then breaks the coincidence into four overhangs and prices each. Capability: models are now good enough to build products with real moats, and harness and model have been shown to be separable — Microsoft's Copilot now lets you pick a model — so "pure capability no longer translates directly into a moat." Product: the labs need to build user touchpoints while their capabilities still lead, yet Meta built the best personal-agent product he has tried on a model that is not state of the art, which is a bearish signal. Pricing: today's high prices are a price umbrella held up by compute scarcity; slow down training and you could reallocate compute to inference and take the whole market. Capital: the funding structures now reach into debt markets and even insurance floats and pension funds, so AI had better deliver before it's too late.
And then safety, where he arrives at a counterintuitive conclusion: as to the risk that actually exists today, pacing the frontier makes things worse. His argument is crisp. An attacker's automation always has positive expected value — a failed exploit leaves the status quo, a successful one takes the system. A defender's automation has negative expected value — success merely preserves the status quo, while failure breaks the software or introduces new vulnerabilities. So attackers will go fully automated while defenders keep a human in the loop, and no human in the loop can keep up. Effective defence requires models trusted enough to act autonomously without bringing down the infrastructure they are protecting — and we're not there yet. He also separates two things routinely conflated: a model directed to do bad things is aligned; whether models ought to know good from bad is an entirely different question.
The article's final line is this method in its purest form:
Anthropic is fine with Anthropic being in the lead; anyone else requires government intervention.ibid.
With someone who has written for thirteen years and close to three thousand posts, the transferable part is rarely the conclusions. All six of these are induced from the pieces above, not from general impressions.
One: to break a theory, find its sample, not a counterexample. This is his most distinctive move. Producing one case Christensen can't explain only earns you "there are always exceptions". What he does instead is ask whose purchasing behaviour the theory was generalised from, and discover the answer is disk drives, PCs, mortgage banking, microprocessors and software — enterprise procurement, all of it. Once that holds, all three premises fail at once in consumer markets and the building collapses on its own.
Two: make "where does the money come from" the first question, not the last. To judge whether a company will do something he looks neither at its technical capability nor at what executives say on stage, but at whether the thing runs with or against its revenue structure. Google inventing the transformer and shipping nothing, Kodak inventing the digital camera and talking itself into calling it a toy, Microsoft happily betting Bing — three unrelated-looking facts that are one fact under this question.
Three: use objects so specific they hurt. He explains publishing with a Taiwanese lens maker; the structure of tech competition with the count of American automobile companies founded per decade from 1895 to 1980 (233 in the 1900s, zero in the 1980s); the risk in AI financing with Jay Cooke's ill-fated 1870 agreement with Northern Pacific. An abstract structure has to hang on a physical object you can remember — you may forget "value flows to the edges", but you will not forget that Largan and Foxconn are only five times apart.
Four: maintain your concepts like software. He built a Concepts index on the site and hangs a decade of articles under headings like Aggregation Theory, Disruption Theory, Incentives. Aggregation Theory itself carries clear version numbers: v1 in 2015; in 2017 an explicitly atypical article that adds definitions and a taxonomy (and announces that it has no narrative and no news hook, its purpose being to provide a future point of reference); in 2018 the Gates line. He doesn't repeat himself, he upgrades himself.
Five: date your errors, and classify them. "I did that previously and was hilariously wrong" — note that he doesn't stop at the apology but adds a diagnosis: "being wrong is more often than not a matter of timing." The 2018 case goes further: he replaced his own five-year-old model (an asymptote) with one that yields a different conclusion (expectations themselves rise). For someone who makes a living on frameworks, publicly patching your own framework is the expensive move, and he makes it often.
Six: quote at length; don't paraphrase. This is the habit easiest to overlook and hardest to overstate. Christensen's books, Bezos's shareholder letters, Buffett's 1991 letter, the explanatory memorandum to an Australian bill, the philosopher Nadella cited at a developer conference, the 1980 Jobs clip in which the bicycle story has not yet acquired the phrase "bicycle of the mind" — he blocks them in whole and then takes them apart. The cost is very long articles. The benefit is that the reader always holds the primary evidence and can judge whether the dismantling was fair.
First: Aggregation Theory may be old wine in a new bottle, and the bottle may be harmful.
This is investor Lyall Taylor's 2019 critique, and the most substantive one this page was able to read:
The problem I have with the theory is that it implies there is something fundamentally new or unique about the economics of the brave-new-world of tech, when in reality, the old economic rules still work just fine. This, in turn, creates the raw material to rationalize bubble thinking/valuations, instead of more level-headed analysis.Lyall Taylor, The problem with Ben Thompson's 'aggregation theory', 20 May 2019
Taylor's substantive point is that what actually happened is not that demand became more important than supply, but that the demand side became more concentrated than the supply side — which is not new: Carnegie and Rockefeller got rich controlling the infrastructure through which others reached the market. He offers a neat counterexample: music streaming, where supply is more concentrated than demand (three majors against Spotify, Apple Music, YouTube, Amazon and more), with the result that the streamers don't make money and the labels do.
In Ben Thompson's parlance, it could be said that the reason is 'supply aggregation theory'. In conventional parlance, it is because the supply-side is the point in the supply chain with greater market power.ibid.
He also argues that the word "aggregation" implies a structural permanence that competitive economies do not offer: the squeezed end has an incentive to consolidate, and the over-earning end eventually attracts attack. That charge has an echo inside Thompson's own text — he classified Netflix as a Level 1 aggregator back in 2017 and said such companies are "susceptible to competitors with deeper pockets or orthogonal business models". So the disagreement may be closer to a fight over naming than over facts. But Taylor's warning that the vocabulary gets used to rationalise valuations is not a problem definitions can solve.
Second: a critique I could not read, recorded as bibliography only. The antitrust scholar Tim Wu published a two-part review of Stratechery on Medium (superwuster, "Reviewing Ben Thompson's Stratechery" and its part two). At the time of writing both were behind a Cloudflare block and would not open, so I have not read them and will not restate their arguments. I note only that they exist; interested readers should check for themselves.
Third: his single biggest prediction now looks wrong, and wrong in an interesting way.
In January 2020 he wrote "The End of the Beginning", whose conclusion he himself called "a tad bit heretical":
there may not be a significant paradigm shift on the horizon, nor the associated generational change that goes with it… today's cloud and mobile companies — Amazon, Microsoft, Apple, and Google — may very well be the GM, Ford, and Chrysler of the 21st century.The End of the Beginning, 7 Jan 2020
Three years later the same man opened "AI and the Big Five" with: "It seems clear to me that this is a new epoch in technology."
In fairness these are not a flat contradiction, and half of the 2020 claim has been vindicated: the biggest winners of the AI era include a great many incumbents — Microsoft, Google, Meta, Nvidia, TSMC — rather than a cohort of challengers starting from zero, which is precisely what he was arguing. But the strongest, most arresting version of that article — that paradigm shifts themselves were over — did not hold. It is worth recording separately because it exposes a systematic tendency in this mind: he is superb at explaining structures that have already settled, and the stability of structures is exactly what he is most prone to overestimate. Taylor's critique points at the same thing.
Fourth: the slant that comes with position — half of which he concedes himself.
He lives on subscriptions, and his readers are the executives, venture capitalists and practitioners of the industry he covers. He conducts CEO interviews, and those interviews yield material no one else has (in the 2026 piece read for this page, Nadella breaks the news there that Microsoft will decouple harness from model). His ethics statement discloses everything disclosable and sets a bar far above his peers. But disclosure is not dissolution: if an analytical framework systematically concludes that aggregator dominance is the natural result of technical structure and that conventional antitrust remedies are largely futile, that framework is going to be popular in this industry, and the author not owning shares does not change it.
His own words are not ambiguous. In the 2017 article defining aggregators he writes three limits for regulators: an aggregator's user relationship is chosen by the user (because the service is superior), which makes consumer-welfare antitrust arguments hard; the nature of digital markets may make aggregators inevitable, so breaking companies up or capping their addressable markets will likely just install a new aggregator; and aggregators make it dramatically cheaper for suppliers to reach customers, so regulators should take care to preserve — even protect — those new opportunities. Each is defensible. Together they amount to a brief for leaving things alone. He is not carrying anyone's water — but a framework built on structural determinism arrives most naturally at "this is structure, don't bother", which is exactly what the companies being analysed most want to hear.
Finally, he has written a fairly brutal piece of self-criticism, best reproduced as-is. In the 2026 article, discussing his own read on agentic coding, he inserts this:
I do, as a matter of my job, talk about things that I do not necessarily experience personally; the canonical example is my long-running analysis of and advocacy for advertising as a business model, despite the fact I run a subscription business.Frontier Overhangs, 21 Sep 2026
A man whose calling card is "your business model is your destiny", admitting that his most famous body of analysis concerns a business he has never run. That is not self-deprecation; it is putting his own sampling bias on the table — exactly the move the third idea above demands of everyone else who speaks.
This page also draws on: Publishers and the Smiling Curve (2014-10-28), Blogging's Bright Future (2015-02-02), Defining Aggregators (2017-09-26), Divine Discontent: Disruption's Antidote (2018-05-02), Tech's Two Philosophies (2018-05-09), The End of the Beginning (2020-01-07), Publishing is Back to the Future (2021-01-27), Tech Philosophy and AI Opportunity (2025-07-08), and About / Ethics Statement. His own concept index is at stratechery.com/concepts. The critique quoted in section six is Lyall Taylor, The problem with Ben Thompson's 'aggregation theory' (2019-05-20).