Chris Dixon
What this covers
Chris Dixon maps the internet's trajectory from decentralization to consolidation and back again. The central claim is structural: the open web lost to corporate networks not because of inferior technology but because companies could subsidize costs (video hosting, search infrastructure) that decentralized protocols like RSS could not. Now two forces compete for the internet's future. Blockchains, Dixon argues, can restore the promise of low take rates and creator control by combining protocol networks' societal benefits with corporate networks' ability to offer incentives through smart contracts and token treasuries. Conversely, AI threatens to accelerate consolidation by breaking the implicit covenant of the web—the exchange where platforms link back to content creators—and concentrating power in the hands of a few companies with large datasets and capital.
The conversation spans several distinct territories. Dixon discusses first-order and second-order effects as a lens for understanding technology's political consequences, illustrates composability as software's version of compounding interest, and traces barbelling across retail, media, and other industries where the middle collapses while extremes thrive. On blockchains, he examines stablecoin infrastructure reaching payment parity with traditional systems, the distinction between brownfield and greenfield blockchain applications, and read-write-own as the internet's third era. On AI, he names Stack Overflow as a canary in the coal mine for content atrophy, argues open weights models are crucial to preventing centralization, and compares AI's commercial structure to whether it will resemble the internet or mobile—a choice with vastly different implications for startups and incumbents. Across these territories, Dixon foregrounds how architecture shapes sovereignty and power.
Dixon argues the internet consolidated into a handful of high-take-rate corporate networks because of subsidization and network effects, and that blockchains can restore the decentralized internet's promise of value flowing to creators—while AI, left unchecked, will accelerate centralization and break the implicit content covenant that sustains the open web.
- Protocol networks lost to corporate networks largely because companies could subsidize costs (e.g., YouTube hosting) that decentralized standards like RSS could not
- Blockchains can combine protocol networks' societal benefits (low take rates, community control) with corporate networks' competitive advantages via smart contracts and token incentives
- AI breaks the search-era 'link-back covenant,' threatening to atrophy the long tail of the internet (Stack Overflow as canary), and open-source/open-weight models are the key counterbalance to AI-driven centralization
Consolidation of internet economics through intermediaries has replaced the web's original decentralized structure.
- The structural problem with the modern internet is primarily economic: YouTube takes about 50 percent (generous by internet standards but high by general economic standards), Spotify takes 30 percent plus additional label fees, and only about 8,000 of Spotify's 8 million artists make more than $50,000 a year—not because consumers aren't paying for music, but because intermediaries capture the economics.
“it's like 8,000 of the 8 million artists make more than $50,000 a year, so a very, very small percentage make something like the average American living.”
- The internet began as a decentralized network where creators could sell directly to consumers with no intermediary, but it has consolidated such that roughly 90 percent of internet traffic runs through fewer than 10 services and companies, which extract high take rates.
“Ninety percent of the internet traffic runs through less than 10 services and companies. They have very high, what we call take rates on the internet business.”
- The internet has operated for 25 years on an implicit covenant where distribution platforms (Google, Facebook) take snippets of content from the long tail under fair use and provide a link back driving traffic, in exchange for which content providers monetize via ads or subscriptions; this equilibrium evolved rather than being designed and worked reasonably well, if more in the platforms' favor.
“for 25 years or so, there's been this implicit covenant between the distribution centers like Google and the content providers like us.”
- A single media company like The Atlantic cannot do much about AI breaking the web's content economics because it is a systemic, structural technology problem rather than something an individual node in the system can solve.
“I'm not sure that a single node in the system, like The Atlantic, has a lot to do. I think that it's really structural.”
- In a world where AI gives you the answer directly with no need for a link, the implicit covenant collapses, threatening to atrophy the long tail of the internet; Stack Overflow is the canary in the coal mine, with traffic down roughly 80 percent after AIs trained on it and on GitHub to create coding tools like Copilot and Cursor, risking a future internet structure resembling 1970s broadcast TV with only a few channels.
“The traffic is down — I don't know — 80 percent now because of these new... My concern is Stack Overflow is the canary in the coal mine. This is the first thing to go.”
Blockchains offer a new architecture for building internet services with no intermediary, low take rates, and user asset ownership.
- Blockchains are a new architecture for building internet services with no intermediary and low take rates; using smart contracts (code that owns tokens and can spend money), blockchain networks done right can offer the societal benefits of protocol networks (money flowing to the edges, community control) while also having the competitive advantages of corporate networks, including the ability to offer incentives and subsidies like a treasury.
“Blockchain networks done right can ideally be the best of both worlds. They can have the societal benefits of protocol networks but the competitive advantages of these corporate networks.”
- Stablecoin infrastructure has reached the long-sought target of 'one second, one penny' transfers on Layer 2 networks like Base and Solana, enabling use cases like treasury management, cross-jurisdiction money movement, and fully automated international invoicing that reduces invoice fraud via reputation networks—which is why companies like Stripe (via its Bridge acquisition) call stablecoins the 'room temperature superconductor' of payments.
“have hit what we've thought of for a long time as a target, which is one second-one penny to transfer things. That's where we are now technically.”
- Blockchains let users own digital assets—fungible (interchangeable, like currency) or nonfungible (distinct, like a movie ticket, housing deed, in-game item, or representation of a physical object)—and the better architecture is to hold these on a blockchain where the user can move them across services, rather than locked inside a single provider's system; this is the 'own' era following the 'read' and 'write' eras of the internet, shifting power so users have a persistent, portable inventory unbundled from any single service.
“you can have those either locked inside of the game owned by the game maker, or you can have them on a blockchain, where the user can move them around and control them. I think that's a better architecture.”
Open-source composability drives technology adoption by enabling software reuse at scale despite coordination costs.
- Composability—the ability for anyone to write software once, put it on GitHub as a public resource, and have others reuse it as a Lego brick to build other things—is to software as compounding interest is to finance, and it is one of the main reasons open source (Linux) has taken over 90 percent market share despite the coordination costs of decentralized development.
“composability is to software as compounding interest is to finance. It's this thing where, basically, one person builds something, and you never have to build it again, and you reuse it, and you build the Lego bricks up.”
- Decentralized protocol networks like RSS lost to corporate services like YouTube not just because of user experience but because corporate networks could subsidize expensive costs (like video hosting) using raised venture capital, whereas protocol networks have no company behind them and cannot offer such subsidies in winner-take-all markets with network effects.
“These companies like YouTube and Twitter — they raise a lot of money. What do they spend that money on? A lot of it is for subsidizing the hosting costs because the basic idea is, these are winner-take-all markets.”
- The emerging consensus is that AI foundation models may end up as commodities—essentially large-scale statistical analysis whose tricks and secrets propagate widely, like math libraries (PyTorch)—with durable businesses forming instead at the cloud-provider layer (AWS) and the end-user application layer (Cursor, vertical products), as evidenced by DeepSeek achieving competitive results with a team of only about 150 people.
“It may just be the foundation models end up being like PyTorch or whatever, like math libraries. Essentially, it's large-scale statistics.”
- AI tends to be centralizing and consolidating of power because the countries and companies that produce these very powerful systems gain immense power, which is why governments and companies will eventually demand open source—making open source one of two dominant models competitive with proprietary AI—whereas crypto is decentralizing, capturing Peter Thiel's framing that crypto is libertarian and AI is communist.
“AI tends to be centralizing. It tends to be consolidating power... the people that produce these things, if they're very powerful things that manage much of the world, being one of the countries and companies that produce them seems to give you immense power.”
Stablecoins have achieved real-world adoption and offer payment, currency, and national interest benefits.
- Stablecoins—digital tokens backed one-to-one by an asset in the bank like a dollar or euro (e.g., Tether, USDC)—have seen massive real-world adoption, with Visa's dashboard tracking roughly $3.5 trillion in stablecoin transactions in the prior month, growth that occurred before regulatory clarity and could accelerate further with pending legislation.
“It's $3.5 trillion in stablecoin transactions last month, and that's been steadily going up.”
- The pending stablecoin legislation that passed the Senate Finance Committee would require 100 percent reserves, making bank runs impossible for compliant asset-backed stablecoins, in contrast to algorithmic stablecoins like Terra Luna that collapsed because they were backed by their own token in a circular structure rather than dollars.
“The hope would be that these existing ones and the new ones and the regulations around them would require full one-to-one asset backed, and therefore make bank runs impossible.”
- Beyond payments, legitimizing stablecoins serves a national security and dollar-status interest because it is another way to popularize the dollar, potentially maintaining or increasing its status as the world's reserve currency given observed global demand.
“it's another way to popularize a dollar, right? We've seen the demand for it, and if we legitimate it, it presumably will make, hopefully, the dollar more popular and increase its status or maintain its status as a reserve currency.”
- US financial services regulations historically tend to propagate to other countries through reciprocation requirements (KYC, AML), so the hope is that a US stablecoin framework would become a global norm that legitimate actors follow, marginalizing rather than eliminating the offshore, higher-yield, run-prone stablecoins at the fringes.
“historically, our financial services regulations tend to propagate to a lot of other countries.”
AI's consolidating tendency can be counterbalanced by open-source models that enable auditing and choice.
- Artificial intelligence, left unchecked on its current trajectory, will likely lead to even further consolidation by rewarding companies with large amounts of data and capital, and blockchains can serve as a counterbalancing force to those consolidation trends.
“left unchecked on its current trajectory, it will likely lead to even further consolidation and reward companies with large amounts of data and capital”
- Whether a country retains sovereignty when adopting foreign AI depends critically on architecture: if a nation like Peru can obtain the open weights of a model like DeepSeek and fine-tune or change those weights, it preserves meaningful control, whereas a closed proprietary model embedded as a service leaves the country effectively dependent on the foreign company.
“I do think it's different if they can get the open weights of DeepSeek, and the Peruvian people and government can fine-tune or change those weights and decide on...”
- Politics is downstream of culture and information flows, so as AI becomes the source from which people get their news and political information, the key question becomes who created that AI and whether it is open source and auditable; strong open-source AI is the single most important factor for making the AI future better because it gives people choice and allows the systems shaping news and culture to be audited.
“If I can pick one issue that will make the AI future better, my own belief is that we have very strong open-source AI so that people can have a choice. It can be audited. It can be open.”
Major technology policy disputes will be ultimately decided by Congress rather than courts.
- When a breakthrough technology like AI or crypto emerges, there's a strategic choice between brownfield (going after existing use cases to make them better) and greenfield (going after new use cases); blockchains are currently most successful in areas that are 'more broken'—payments and financial services—rather than challenging entrenched network effects in social media where the wars are largely over.
“In software, there's a phrase, greenfield, brownfield. Greenfield is, you have some breakthrough — AI or crypto or whatever — do you go after existing use cases and make them better or do you go after new use cases?”
- The key question for AI's commercial structure is whether it resembles the internet or mobile: with the internet, roughly 90 percent of net new market-cap value went to startups that didn't previously exist (Amazon, Google), whereas with mobile roughly 90 percent went to incumbents (Apple, Google, Facebook) who leaned in and grew their user base 10x—so much so that almost no venture firm beat the strategy of just buying Apple stock in 2010; if AI is like mobile, it reinforces big-company strengths (data, capital, distribution) which is bad for VCs but potentially great for consumer surplus.
“With mobile, on the flip side, something like 90 percent — I haven't done an exact study — went to incumbents.”
- A possible future structure is a DAO—an autonomous entity on a blockchain—that functions like a university endowment with two halves: an investing side (in tokens and assets) and a giving side that uses AI to distribute grants (a blockchain version of Fast Grants), creating a self-sustaining, compounding entity that funds projects from its returns.
“could we create a blockchain, an autonomous AI entity that is like an endowment? Half of it is investing... and the other half is giving the money away, ala Fast Grants.”
- The modern naturalistic, physicalist worldview has forgotten that there is more than propositional knowledge ('knowing that')—there is also 'knowing how' and other kinds of knowledge; religious traditions are good at embracing these multiple forms, so the common error is to analyze something like Christianity as a set of propositions when it is actually a form of life involving community, ritual, and co-developed meaning.
“One of the things that we've done in the modern world is we've forgotten that there's more than just propositional knowledge, more than just knowing that. There's knowing how.”
- Venture capital is a bundle of services—raising money, picking investments (where AI could soon beat humans), and a high-touch human aspect of building a culture and a support team that entrepreneurs want to work with—and while AI may excel at the picking aspect, the human advisory role of helping founders build and design (e.g., robots) is likely to persist for some foreseeable period.
“In that picking aspect, I think AI could do a very, very good job and probably beat humans in a lot of ways in the near future. There's another aspect... there's a very human aspect to what we do.”
- Studying analytic philosophy—especially the dialogue between thinkers like Hume, Kant, and Hegel—teaches a kind of intellectual plasticity and meta-moves (such as Kant's transcendental argument that reasons from premises to the preconditions of those premises rather than to conclusions), forcing intellectual honesty and the ability to argue multiple sides, even if it rarely yields firm conclusions.
“what he's saying is — this is what a transcendental argument is — is you're not going from premises to conclusion. You're going from premises to preconditions to those premises.”
- Many of the world's most challenging problems and gating factors to productivity, economic growth, and human well-being are not intelligence problems that more AI can solve, but collective action and coordination problems—getting people to agree—as illustrated by housing, which is a regulatory or collective-action problem rather than something more intelligence would fix.
“a lot of the gating factors to productivity, economic growth, and just human well-being, I believe, are collective action problems, coordination problems, getting people to agree on something.”
- The major disputes around AI (copyright, safety) and crypto are too important to be settled in the courts and will ultimately be decided in Congress, as evidenced by crypto's four years of regulatory struggle in the courts now moving toward congressional resolution.
“I believe these are not going to be settled in the court. They will be settled in Congress. They're too important.”
Technology's second-order effects reshape society in ways not apparent from first-order adoption patterns.
- Technology always has first-order and second-order effects: the automobile's first-order effect was getting from A to B faster (1900-1930), while its second-order effects—suburbs, trucking, highway systems—played out over the following decades; social media's first-order effect was sharing trivia, while its second-order effect is a restructured political dynamic enabling insurgent movements and counter-establishment narrative universes, and we are still early in this development.
“The first-order effect of the automobile was, you can get from point A to point B faster. The second-order effect was suburbs, trucking, highway systems, all the next 50 years.”
- A common pattern with technology is 'death of the middle' or barbelling: when the internet arrived, mid-sized retailers (Kmart, Sears) that were artificial bundles of distribution, merchandising, and service got unbundled and lost, while value flowed to the extremes—scale players (Amazon, Walmart) and high-touch boutiques (Gucci, Hermès, LVMH); the same has happened in media, where 30-second TikTok dopamine hits and three-hour podcasts win while the 30-minute sitcom dies, because the barbell matches genuine human needs.
“you get this barbelling effect where you either want to be really, really big like Amazon and Walmart and take advantage of scale, or you want to be a boutique — high-touch, high-brand-value — Gucci, Hermes”
- Just as photography both replaced representational art (pushing high art into abstraction) and created an entirely new medium (film) that couldn't exist before, AI may replace illustration while enabling a new AI-native form of media that couldn't have existed before; meanwhile, as machine-created content scales, demand will rise for bespoke human things (live concerts, fine art, handcrafted books) because fundamental human nature doesn't change.
“maybe AI will be the camera and it will replace illustration... But on the flip side, maybe it will enable a new AI native form of media that couldn't have existed before.”
- AI is unlikely to slow down because there are so many smart people, so much money, and so many improvement paths; the field has moved from pre-training to reasoning and will likely find another method within a year, with largely untapped frontiers in humanoid robots and video and other media modalities.
“My guess is it's not going to slow down. There're so many smart people and so much money and so many different ways in which it can improve.”
Bitcoin's global neutrality makes it suited for machine-to-machine transactions outside national currencies.
- Bitcoin's volatility, a knock against it as a payment system for humans, becomes a non-issue for machine-to-machine transactions executed in microseconds; and David Marcus's argument (via Lightspark) is that Bitcoin is preferable for AI payments precisely because it is not the dollar—it is the only crypto asset credibly globally neutral, seen by no country as a national instrument, making it suited to be the currency of the internet.
“His argument is Bitcoin is preferable because it's not the dollar, that a lot of countries will want some. It's the only crypto asset that is credibly globally neutral”