What this covers
Ben Thompson of Stratechery works through Aggregation Theory in conversation, applying it to publishing, music, video, e-commerce, and PC hardware. The first stretch is definitional. Microsoft's Windows and Apple's App Store are platforms whose network effects sit outside the product, among developers, OEMs, and users. Aggregators internalize those effects. Google is the example: an entire SEO industry pays to become better suppliers, while Google spends nothing on supply and monetizes with ads.
The range is wide. Thompson argues that falling cost per ad or per click can signal expanding inventory rather than weakness. He uses Shopify to argue that high merchant churn is healthy, and Netflix to show a marginal customer growing more valuable as the catalog grows, which runs against the usual customer acquisition cost pattern. Spotify is his case of a weak aggregator, since record labels hold a monopoly back catalog. He also applies the Smiling Curve to the PC industry and to publishers stranded in the middle of the value chain.
The contested turns come in policy and capital. He says regulation aimed at Google and Facebook misreads where their power comes from, which is customer choice. He rejects the tech-disruption narrative in favor of one continuum, and forecasts that the smartphone and the cloud will outlast specialized devices like watches and AR glasses. He also doubts that venture capital is the right vehicle for physical infrastructure, and he draws on early car-industry consolidation and the origins of cable TV bundles.
Thompson argues that internet businesses are organized around aggregation, where control of consumer demand lets a company commoditize suppliers, and that platform-based regulatory frameworks misread this. He also argues that the internet removes the physical scarcity that once defined media and retail, so incumbents' models and regulatory assumptions are outdated.
- Aggregators win by owning consumer demand and commoditizing suppliers, while platforms win by making suppliers differentiated and dependent
- Internalized network effects and zero marginal cost make aggregators self-reinforcing, and customer choice drives their power, so platform-style regulation fails
- Removing physical constraints (shelf space, broadcast time, distribution) flips old profit centers and scarcity-based assumptions
Aggregators gain power by owning consumer demand and controlled connections, not supply.
- Many companies base projections on the acquisition cost of their earliest customers, who are their ideal and easy-to-convert buyers, but marginal customers become harder and costlier to acquire, so the projections fail and Facebook and Google end up taking the profit.
“Facebook and Google actually end up taking all your profit over time”
- An aggregator is defined by three characteristics: a direct relationship with users, zero marginal cost of serving users, and decreasing customer acquisition costs over time as scale increases.
“Number two is zero marginal cost of serving users”
- Google's aggregation creates a virtuous cycle in which websites compete to rank better on search, so an entire SEO industry pays money to be better Google suppliers while Google spends nothing on supply and can then monetize users with ads.
“the SEO industry is in many respects an industry where people pay money to be better Google suppliers”
- For an aggregator like Netflix, the marginal customer becomes more valuable as the catalog grows, because evergreen content stays relevant, so customer acquisition costs fall with scale, while most companies see marginal customers become harder and more expensive to acquire.
“the marginal value of Netflix is increasing as their customer base increases”
- The key takeaway of aggregation theory is that on the internet, controlling demand matters more than controlling supply, and the relative power of suppliers determines how large and profitable the demand-controlling company can become.
“controlling demand is more important than controlling supply is the big picture takeaway”
- A platform moat is built on APIs and developer infrastructure, while an aggregator moat is built on connections that the aggregator controls and that a supplier cannot easily reach alone, so beating an aggregator requires something easier, better, or highly differentiated.
“To beat that, you have to have something that's easier and better, or so highly differentiated that customers will go through the effort to go around it”
- Facebook's shopping partnership with Shopify shows that the gravity of the internet flows to whoever owns consumer attention, so Shopify's weakness is that it is not a natural owner of demand and does not get Amazon-scale valuations.
“the gravity on the internet always flows to whoever owns consumer attention”
- Netflix is asserting more control over suppliers by owning content outright, so it can capture more upside from its control of demand, a strategic point that most investors miss when they focus on free cash flow.
“they're actually asserting more control over supply so they can garnish more upside from controlling demand”
- The pandemic did not redirect technology but accelerated trends already underway, so if foundational companies are becoming more critical to everyday life, the pandemic should help their valuations by completing that shift.
“What the pandemic, I think, did was accelerate trends that were happening anyways”
Aggregators commoditize suppliers by internalizing network effects, unlike platforms that differentiate suppliers.
- Merchants on Amazon accept a poor deal of having Amazon label, house, and store their goods and hold their inventory, because the customers are on Amazon, which is a clear example of demand control letting an aggregator control supply.
“It's a great example of the aggregator concept where controlling demand lets you control supply.”
- Platforms have externalized network effects, where value comes from interactions between developers, OEMs, and users on the outside, so they want suppliers to be highly differentiated, whereas aggregators have internalized network effects that commoditize suppliers because the aggregator captures value from its own data and usage loop.
“the network effect is externalized”
- Shopify is an Amazon competitor with a different model that can be very profitable and have a greater impact on suppliers, but it will not achieve Amazon-style scale or valuation because it does not follow the aggregator model.
“It doesn't follow that they're going to in the long run have Amazon valuation”
- Uber's network effects are externalized between drivers and riders, but it commoditizes drivers instead of differentiating them, creating an oppositional relationship with suppliers; its heavy spending on recruiting drivers signals that it is not a true aggregator.
“That's a signal that they're not quite an aggregator.”
- Shopify could build a platform-like moat by intermediating between merchants and third-party logistics providers, since no single 3PL can serve tiny merchants, while Shopify can connect both sides at scale, creating connections that cannot be formed one-on-one.
“Shopify should have merchants on the top, all the infrastructure to support eCommerce on the bottom”
- Shopify enables merchants to build direct relationships with customers and bypass intermediaries like Amazon, Google, and Facebook, just as independent creators can use email as an open standard to reach audiences without aggregators, making Shopify an orthogonal competitor to Amazon.
“It's an orthogonal competition because your entire point is to support and feature suppliers”
Dominant tech firms are durable because disruption narratives overstate fragility and consolidation recurs.
- Amazon, Microsoft, and to an extent Google are best placed in cloud because the current general-purpose paradigm will last longer than people expect, and specialized edge applications are too small to be disruptive the way cloud and mobile were.
“they're best placed for these specialized applications because those specialized applications are not as large of an opportunity”
- The car industry went from hundreds of firms per decade to three dominant producers by 1930, and then almost no new entrants until Tesla 60 to 70 years later, showing that tech's market structure will likely follow the same consolidation pattern once economies of scale and paradigms are set.
“by 1930, it was over”
- iOS and Android are in a stronger position than Windows ever was, because the smartphone was a natural hardware-enabled step beyond the PC that did everything the PC did but more continuously and in more places.
“IOS and Android are actually far stronger and in a better much position than even Windows was”
- The smartphone is the logical endpoint of the shift from destination-bound to always-connected computing because it combines a usable screen, pocketability, and constant connectivity in a general-purpose device, and specialized devices like watches and AR glasses cannot replace that general-purpose role.
“we've reached a logical endpoint of that revolution”
- Tech's supposed series of disruptive eras (mainframe, desktop, mobile, internet) are actually one continuum of two dynamics, batch-to-continuous computing and centralized-to-distributed data, so dominant firms are less fragile than the disruption narrative implies.
“all these eras from another point of view are actually all on one continuum”
Platforms and aggregators must be distinguished, or regulating them as one misreads customer choice.
- Platforms are facilitators with APIs and developer ecosystems that can squeeze complementors because they have direct components in the value chain, whereas aggregators leverage abundance and consumer overwhelm to become the place users go first, and conflating the two leads to bad regulation.
“they're also dangerous because they have direct controls or components in that value chain”
- Aggregation theory is largely a descriptive account of how internet companies already work, and Thompson's main contribution is articulating why aggregators differ from platforms and why that difference matters.
“there's certainly a bit of a descriptive nature to this”
- Thompson acknowledges Google has committed abuses, but argues Google's dominance is not forced through control of the access layer, which is what differentiates it from platform power abuses such as blocking competitor apps.
“If Androids only made it so the Yelp app didn't work on Androids, that's an obvious abuse of platform power”
- Regulations aimed at Google and Facebook fail because their power comes from customer choice, not forced usage; companies like Yelp losing traffic to Google is a marketing problem, and regulators punishing Google for surfacing shoes instead of shopping comparison sites misunderstands the aggregator.
“if you don't understand the nature of their power, which is customer choice driven, your regulations aren't going to work”
Value concentrates at the ends of the value chain, leaving middle intermediaries with little.
- Companies that sit in the middle of the value chain assembling other firms' differentiated inputs, like Acer, Foxconn, and traditional publishers, capture little value, while firms that own differentiated inputs (Largan Precision, TSMC) or consumer attention capture most of it.
“To be in the middle putting pieces together is not a particularly great place to be.”
- In publishing, value flowed to aggregators that capture consumer attention (Facebook, Google) and to individually differentiated creators, while traditional publishers in the middle lost power because worldwide distribution erased their local monopolies.
“all of the power and value was accrued to the ends and they were losing their position in the value chain”
- In the PC industry, value sits at both ends of the value chain (brand and marketing on one side, R&D and patent-protected technology on the other), while the middle fabrication step has low value and low margins.
“The value of those components looks like a smiling curve where it's high on the ends”
Netflix wins through scale, upfront content investment, and internet assumptions rivals like Starz never considered.
- Over the long run, many media companies that fail at running their own streaming services will return to selling content to Netflix, making Netflix the future cable bundle for nearly all non-live content.
“Netflix ends up being the future cable bundle for basically anything that's not live”
- Competing head-on with Netflix on subscriber-based streaming is a losing strategy because Netflix spreads content cost across more subscribers, so rivals face a structurally higher cost per subscriber, and they need a different business model to compete.
“I think it's very stupid to attack them head-on.”
- Netflix's model of producing content well ahead of release, sometimes a year of inventory, left it better placed during the pandemic than traditional networks that operate with only three to six months of just-in-time content.
“they've ended up being better placed because they shifted to a higher inventory model”
- Netflix's shift from licensing shows to producing and owning originals means it pays upfront for worldwide rights and residuals, pulling costs forward into years of negative free cash flow, while later the revenue upside is captured without future costs.
“that's why for about three years, Netflix's free cashflow was massively negative because they were pulling forward all these costs”
- Disney's streaming service is valuable less as a standalone subscription business than as a top-of-funnel tool that gives Disney direct customer relationships, email, and payment data, which it can monetize through theme parks, cruises, and other businesses.
“you have to think about Disney Plus not just as being a potential moneymaker but as a phenomenal top of the funnel tool”
- When Netflix bought the Starz catalog of about 11,000 movies, its effective catalog was 11,000 titles on demand, whereas Starz's effective catalog was only what was airing on its channel, giving Netflix a superior product on a dimension Starz never considered.
“Whereas the effective catalog size on Netflix was 11,000.”
- Netflix's strategic history shows it repeatedly embraced internet assumptions, from DVDs with no late fees that dropped inventory and time constraints, to a streaming catalog that was effectively 11,000 movies versus Starz's channel-limited schedule, which Starz never thought to offer.
“Whereas the effective catalog size on Netflix was 11,000.”
Zero marginal cost makes high failure rates, churn, and abundant content strategically sound rather than wasteful.
- For Shopify, high merchant churn is a positive signal rather than a negative one, because the marginal cost of serving another merchant is near zero, so many failed shops still mean more top-of-funnel attempts and more successful merchants.
“So, it's actually high churn on Shopify is a positive signal, not a negative signal”
- Trying to sell content on the internet is a losing game because content is just bits with zero marginal cost, so creators must instead sell an ongoing service, such as the feeling of being informed or continuous analysis.
“Trying to sell content on the internet is a loser's game. It's bits on a screen. Zero marginal cost.”
- On the internet, winning requires either being the place where people start their journey, like Google or Facebook, or being highly differentiated and exploiting zero distribution cost to reach anyone, because the internet also allows drastically lower cost structures.
“The way you win on the internet is you be the starting place where people go to start”
- Falling cost per ad or cost per click is a positive signal for internet companies because it shows they are expanding inventory faster than it is consumed, as with Google early on and Facebook Stories, rather than a sign of weakness.
“the cost decreasing is a positive signal because it shows they're increasing inventory faster than it's being consumed”
- The scarcity constraints that organized the analog economy, including broadcast licenses, airtime, shelf space, land for stores, and consumer money, are disappearing on the internet, flipping the economy from scarcity-based to abundance-based.
“All of those constraints are going away on the internet.”
- Epic Games charges little up front for its engine to maximize the number of games that might become huge, since its marginal cost of a failed game is zero and it only needs a few hits to capture large returns, the same logic that makes Shopify's churn positive.
“in a world of zero marginal costs, you want to increase your rolls of the dice as much as possible”
Entrenched scale advantages fail in digital markets except where specialized infrastructure or niche focus applies.
- In the old publishing model, distribution assets like delivery trucks and printing presses made the money, while on the internet the writers become the only profit center because subscriptions depend on them, so support functions must be made more efficient.
“It was delivery trucks and distribution deals, printing presses. That's what actually made the money.”
- Walmart's attempt to beat Amazon at its own e-commerce game failed because Amazon already held scale and customers, forcing Walmart to pay high customer acquisition costs to Facebook and Google while carrying a worse cost structure.
“they're trying to beat Amazon at Amazon's own game”
- Walmart does better than Amazon in grocery because perishables require a different supply chain that Walmart's existing infrastructure handles well and Amazon's e-commerce logistics does not, while Walmart fails at traditional e-commerce where Amazon is already optimized.
“Grocery, for example, Walmart is doing much better than Amazon because they already have existing grocery infrastructure”
- Future media publications will be very narrow, horizontal, and deep in specific topics because their addressable market is the entire internet rather than one city or country, so a niche publication need only reach enough subscribers worldwide.
“future publications will be very, very narrow and they'll take advantage of the fact that their addressable market is not just New York”
- Local journalism is an underexploited opportunity because readers want to feel informed, so a daily email that says what happened, or simply that nothing happened, delivers more value than the ad-stuffed wire-story content local papers produce.
“send an email that said, 'Nothing happened.' That's actually way more valuable to me”
Spotify's podcast strategy is an advertising play that logged-in targeting could monetize far better.
- Podcast advertising is badly under-monetized: host-read ads are hard to scale, and IP-based dynamic insertion is the dumbest targeting possible, while Spotify could monetize far better by using logged-in identity and streaming data to target ads individually.
“It's the dumbest targeting possible.”
- Spotify could become the center of podcasting by using its large logged-in user base and better targeting to command higher CPMs, which would let it pull podcasters onto its platform and eventually not need to pay them, though converting habitual podcast listeners is harder than converting new ones.
“their growth in the area has mostly been converting new podcast listeners”
- Spotify's acquisition of Joe Rogan and its podcast exclusivity is primarily an advertising strategy, as shown by Spotify moving content costs into its advertising division, rather than a subscription play.
“this is pretty clearly an advertising play. That's why, for example, Spotify recently shifted their accounting of costs for content completely to the advertising division”
- Spotify is a weak aggregator in music because record labels control a monopoly back catalog that Apple could also offer, so labels capture subscription revenue share and can renegotiate power against Spotify, unlike in fragmented podcasts.
“So, Spotify is a very weak aggregator in the case of music.”
- Apple, through its iTunes podcast directory and dominant Podcasts player, controlled the end-user relationship and could have built an ad network and rolled up the podcast space, but chose not to.
“they could have rolled up the podcast space ages ago”
Venture capital's upfront-cost model misfits infrastructure with ongoing costs, so lower-risk financing is needed.
- The venture capital model with high up-front cost and infinite upside is a poor fit for real-world infrastructure with ongoing costs, so physical projects like canals and turnpikes need a lower-risk investment model with more guaranteed returns.
“it follows though that that is a very poor model for real world infrastructure spending”
- Infrastructure providers like AWS are best understood as platforms, because they need not interact with end users at all, as shown when users' applications broke after an attempt to IP-block Amazon's addresses.
“a platform does not need to interact with customers at all. It can just be the layer that stuffs just on top of”
- Tech is wrongly expected to finance physical infrastructure, since a typical Fidelity fund manages as much money as the entire US venture capital industry combined, so VC is a small part of total capital.
“your typical fidelity fund manages as much money as the entire US VC industry combined”
- Venture capital's model of big upfront investment followed by infinite upside was built on the economics of chip production, and it has since transferred to software, where spending on a product is followed by zero-cost scaling to each new user.
“Venture capital was formed around the economics of chip production.”
- Stripe's moat is built on infrastructure investment and scale, giving it a superior cost structure that lets it fund developer tools and documentation and eventually move into merchant and financial services, living at the bottom of the smiling curve.
“It's actually living at the bottom of the smiling curve”
Niche publications grow by converting large internet audiences and winning reader consistency and endorsement.
- Concern about subscription fatigue underestimates how large the internet market is, since billions of people could be potential subscribers with varied interests, so niche writers only need to convert the subset willing to pay.
“there are actually billions of potential subscribers in the world and a lot of them have different interests”
- Free-to-win games monetize through cosmetic purchases that signal status without affecting competitive outcomes, which mirrors how people spend on luxury goods in the physical world, and virtual worlds deserve the same monetization logic.
“No one bats an eye at someone buying a luxury bag to signify their status in the real world”
- John Gruber's full-length endorsement of Stratechery, despite disagreeing with one point, took Thompson from about 500 to 1,500 Twitter followers overnight and was the biggest step change in the site's growth.
“I went from 500 to 1500 overnight”
- For a new publication, the most important story is the second one a reader clicks, because it signals consistency and turns a one-article visit into a reader who sticks around.
“the most important story you write is the second story someone clicks on”
Writer bundles form only when shared infrastructure exists and a common platform precedes bundling.
- A writer bundle is difficult to form because a bundle must pay enough to induce high-value incumbents like Stratechery to join, and the cost of buying them off makes the bundle's economics fall apart, as seen with Spotify paying Joe Rogan.
“Given that I'm in a strong position, I was fortunate to be early to this space, that's going to be a very difficult proposition”
- Bundles usually form because of infrastructure needs rather than for commercial reasons, as the cable TV industry did when rural communities built shared antennas and cable, and later satellite-fed networks such as HBO and TNT turned that infrastructure into a bundle.
“they usually end up forming for infrastructure reasons”
- A bundle of independent writers would need a common platform that everyone is already on, such as Substack, before any single provider gains enough economic power that the bundle must buy them off and the economics collapse.
“On the internet, where everyone has distribution, it's a lot harder to figure out what is the cable”
Tech must engage politics and industrial policy as China's influence on US companies grows.
- Tech needs to be more involved in politics and government industrial policy, since the internet lets innovators avoid the regulatory friction of building in the physical world, and China's influence on US companies now requires a coordinated response.
“tech probably needs to be more involved in politics”
- Spreading tech work away from San Francisco through remote work and distributed companies creates competition between cities and politicians, which pressures local governments to improve infrastructure and governance.
“the more there's driving competition between cities and between politicians in spreading the influence”
- US policy toward China is incoherent: it is good to stand up to China but it would be more cogent to form deals with allies such as Mexico and make Mexico a manufacturing alternative to China.
“We should be trying to make Mexico into the next China”
- Opening China was assumed to spread liberal values from West to East, but influence is two-way, as shown by China pressuring the NBA and US companies over a team general manager's comments on Twitter, which requires a collective Western industrial and government response.
“Chinese authoritarianism can flow out where they are demanding a NBA GM be fired”