Note: This is a research note supplementing the book Unscarcity, now available for purchase. These notes expand on concepts from the main text. Start here or get the book.

Network Effects: Why the Rich Get Networked and the Networked Get Rich

In 1876, Alexander Graham Bell held a device that could talk to exactly one other device. Useful? Barely. You could call your neighbor—if they also owned one. Which they didn’t.

Fast forward to today: 8 billion people carry phones that can reach anyone, anywhere, instantly. Each new phone didn’t just add value for its owner—it made every other phone more valuable. That’s not normal economics. That’s network effects, and they’ve quietly become the most powerful force shaping modern civilization.

Here’s the thing about network effects that nobody told you in Econ 101: they’re not just a growth hack for tech companies. They’re a gravity well. Once a platform achieves critical mass, escaping becomes nearly impossible—for users, for competitors, and increasingly, for entire economies. Facebook didn’t beat MySpace because it was better. It beat MySpace because everyone else was already on Facebook.

And now, as AI and robotics reshape the economy, network effects aren’t just making some companies big. They’re creating dynamics that could lock in power structures for generations.

This matters for anyone thinking about abundance economics, because the transition from scarcity to abundance doesn’t happen in a vacuum. It happens on platforms. And platforms with strong network effects tend toward monopoly, not democratization.

Let’s unpack why.

What Are Network Effects, Actually?

Network effects occur when a product becomes more valuable as more people use it. This sounds obvious until you realize most products don’t work this way.

A sandwich doesn’t taste better because more people eat sandwiches. Your car doesn’t drive smoother because your neighbor bought one too. But a phone? Every new phone owner makes your phone more valuable, because now there’s one more person you can call.

The formal definition: “Network effects are a phenomenon where the value of a product or service increases as more people use it.”

But there are different flavors of this phenomenon, and understanding the differences is crucial.

Direct Network Effects: More Users = More Value (Duh)

Direct network effects are the intuitive kind. Every new user directly increases value for existing users.

Examples:

Telephones: One phone = paperweight. Two phones = expensive conversation. A billion phones = the collapse of distance as a meaningful barrier to human communication.
Social networks: Facebook with 10 users is a glorified email list. Facebook with 3 billion users is where your grandmother shares conspiracy theories about 5G towers.
Messaging apps: WhatsApp is useless if your friends are on Telegram. WhatsApp is essential if everyone you know uses it.

The math is elegant. Robert Metcalfe (inventor of Ethernet) noticed that the value of a network grows proportionally to the square of its size. Metcalfe’s Law says if you have 10 nodes, you have 45 possible connections (10 times 9 divided by 2). Double the nodes to 20, and you get 190 connections—more than four times as many.

This creates explosive growth curves that look like magic until you understand the math. It’s not magic. It’s just multiplication doing what multiplication does.

Indirect Network Effects: The Platform Sandwich

Indirect network effects are more subtle and arguably more powerful. They occur in two-sided (or multi-sided) markets where different user groups benefit from each other’s participation.

Examples:

Uber: More drivers mean shorter wait times for riders. More riders mean more business for drivers. Neither group cares directly about their own group’s size—they care about the other group.
App stores: More iPhone users attract more app developers. More apps attract more iPhone users. Apple sits in the middle, collecting 30% of every transaction.
Video game consoles: Nobody buys a PlayStation for the hardware. They buy it for the games. But developers only make games for platforms with players. Chicken, meet egg.

The brilliance of platform businesses is that they can create value just by connecting groups that need each other. Airbnb owns no property. Uber employs no drivers. They’re just the matchmaker, and matchmaking at scale is worth hundreds of billions of dollars.

Data Network Effects: The AI Flywheel

And now we arrive at the variant that matters most for our AI-abundant future: data network effects.

Data network effects occur when a product, generally powered by machine learning, becomes smarter as it gets more data from users. More users generate more data. More data improves the algorithm. A better algorithm attracts more users. Rinse, repeat, until you’ve built an unkillable competitive moat.

This is often called the data flywheel, and it’s why AI companies obsess over user engagement:

“The data flywheel is the idea that more users get you more data which lets you build better algorithms and ultimately a better product to get more users.”
— jxnl.co on the Data Flywheel

The poster child for data network effects is Tesla’s Full Self-Driving program. While competitors test autonomous vehicles with fleets of hundreds of specially instrumented cars, Tesla has over 5 million vehicles collecting driving data—an estimated 50 billion miles per year, or 100,000 miles per minute.

By Q1 2025, Tesla’s FSD fleet had accumulated nearly 7 billion cumulative miles, including 2.5 billion miles on city streets. Every mile teaches the AI something. Every edge case improves the model. Every customer is an unpaid data collector, and they paid Tesla $12,000 for the privilege.

This creates a “virtuous cycle” that’s almost impossible for competitors to break: more data leads to better AI, better AI leads to more customers, more customers lead to more data. RBC projects Tesla will generate $53 billion annually from FSD and licensing by 2035. Not bad for a car company that stumbled into being an AI company.

The Winner-Take-All Problem

Here’s where network effects get troubling for anyone who likes competition, democracy, or distributed power.

Network effects don’t just create advantages—they create compounding advantages. Each new user doesn’t add value linearly; they multiply existing value. This means the gap between leader and challenger widens over time, not narrows.

Imagine two social networks launching the same day. Network A gets 1,000 users. Network B gets 1,001 users. With linear economics, they’re basically identical. With network effects, that one extra user might cascade into dominance—users invite friends, friends invite friends, and before you know it, Network B has 100 million users while Network A struggles to hit 50,000.

This is why platform markets tend toward monopoly. According to NFX’s research on network effects, network effects are responsible for approximately 70% of the value created by tech companies since 1994. The internet economy isn’t just slightly concentrated—it’s almost feudal.

The AI Platform Power Play

Which brings us to 2025 and the question nobody in tech wants to answer honestly: Is AI going to democratize power or concentrate it?

The optimistic view says AI is fundamentally different. The models are converging in capability. Open-source alternatives exist. No fewer than 14 different companies have produced models surpassing the original GPT-4. Competition is fierce.

The pessimistic view says this is temporary. Foundation models exhibit massive economies of scale—training a frontier model now costs hundreds of millions of dollars, with billion-dollar price tags projected within years. The capital requirements alone create a “triopoly” where only a handful of organizations can play.

And there’s a new form of lock-in emerging that’s even scarier than traditional network effects: memory networks.

Memory Networks: When AI Remembers Everything

Traditional network effects favor dominant platforms, but memory networks make this dynamic far more extreme. Why? Because traditional networks accumulate connections, while memory networks accumulate intelligence about you.

Think about your relationship with ChatGPT, Claude, or Gemini over months of use. The AI learns your writing style, your preferences, your projects, your thought patterns. It gets better at anticipating what you need. Stanford’s Robert Siegel argues this creates “stickiness”—the longer you use it, the less likely you are to switch.

“If you keep putting questions into ChatGPT, which learns your behaviors better, and you like it, there’s no reason to leave as long as it’s competitive.”

This isn’t just inconvenience-based switching costs (like moving your contacts to a new phone). It’s intelligence-based switching costs. The new platform doesn’t know you. It takes months or years to rebuild that relationship. Your accumulated context—your digital self—becomes hostage to the platform.

OpenAI is explicitly building around this. Memory and personalization create vendor lock-in. Once you start depositing your cognitive history into a platform, moving that data becomes practically impossible. It’s not a file you can export. It’s a relationship you’d have to rebuild from scratch.

The endgame? A world where switching AI platforms feels like switching identities—so painful that even a superior competitor can’t pry users away.

Network Effects in Robotics: The Physical Flywheel

If data network effects are scary in software, they’re terrifying in hardware.

The global robotics market reached nearly $50 billion in 2025, growing at 14% annually toward a projected $111 billion by 2030. And robotics companies are racing to build the same kind of data flywheels that made Tesla and OpenAI dominant.

NVIDIA’s Project GR00T aims to create general-purpose foundation models for humanoid robots—essentially an operating system that learns from every robot running it. The more robots deployed, the smarter the model. The smarter the model, the more valuable new robots become.

Investment is following the logic: Apptronik raised $350 million in Series A funding. Figure AI is seeking a $39.5 billion valuation—15 times its previous round. Chinese startups Robot Era and EngineAI have secured millions.

The vision is explicit: “Seamless collaborations between robotics ecosystems are about more than technical compatibility; they may also reshape how industries leverage global innovation.” These interconnected systems could lay the groundwork for a new era in which AI-driven robotics becomes a universal productivity multiplier.

But “universal productivity multiplier” cuts two ways. If one company’s robots dominate, their data flywheel becomes unstoppable. Every task performed, every object manipulated, every movement refined becomes training data for the next generation. Late entrants face a competitor that’s accumulated billions of hours of real-world experience.

Regional clusters are already forming. China accounts for 54% of global robot installations—approximately 295,000 robots in 2024 alone. Pittsburgh’s robotics ecosystem includes 250+ companies and 7,300+ jobs across 18 industry verticals. The infrastructure is locking in.

Network Effects vs. Abundance Economics: The Core Tension

Here’s the uncomfortable truth at the heart of the Unscarcity framework: network effects and abundance economics point in opposite directions.

Abundance economics says that as technology makes goods cheaper to produce, they should become accessible to everyone. Solar energy, robot labor, and AI cognition are all racing toward near-zero marginal cost. In theory, this means universal access.

Network effects say that as platforms become more valuable, they consolidate power. The winner takes not just market share but the data, the users, and the infrastructure that future competitors would need. In practice, this means gatekeepers.

These forces collide directly in the AI transition.

Consider housing (a Foundation essential in Unscarcity terms). Construction costs are plummeting as modular building and robotics advance. But access to housing is mediated by platforms—real estate databases, mortgage algorithms, zoning systems—all of which exhibit network effects. Zillow and Redfin didn’t just digitize listings; they accumulated data on prices, preferences, and markets that new competitors can’t replicate.

Or consider healthcare. AI diagnostics are becoming incredibly capable. But the companies training these systems need patient data—lots of it. Whoever gets the data first builds the flywheel first. Mass General Brigham and Mayo Clinic are racing to partner with AI companies precisely because they understand that their data is a diminishing advantage—use it or lose it.

The pattern is consistent: abundance in production collides with concentration in distribution. Technology makes things cheap. Platforms make access expensive.

The Scarcity Reorganization Thesis

Recent academic work suggests that AI doesn’t eliminate scarcity—it reorganizes it. Even when AI delivers technological abundance, scarcity shifts to wherever “constraints bind and rents are generated.”

The framework identifies five layers where scarcity and rent formation arise:

Physical resources and space (land, minerals, water)
Infrastructure (energy, compute, data centers, fabs)
Capabilities (high-trust organizational knowhow)
Institutions (governance of access)
Jurisdictions (enforcement capacity and security)

Notice what’s not on that list: production of goods. AI can make producing things essentially free. But controlling the platforms, infrastructure, and institutions that distribute those things? That’s where new scarcities—and new monopolies—emerge.

This is why the Unscarcity framework emphasizes infrastructure over income. UBI gives people money to spend in markets dominated by network-effect monopolies. Those monopolies capture the cash. Direct infrastructure provision—housing, food, energy—bypasses the gatekeepers entirely.

Breaking Network Effects: Is It Possible?

Network effects aren’t literally unbreakable. History offers examples of dominant networks getting displaced.

MySpace had network effects. Facebook won anyway—partly through better product, partly through viral mechanics, partly through timing (mobile). BlackBerry had enterprise lock-in. iPhone and Android won anyway—by redefining the product category entirely.

The research suggests several factors can weaken network effects:

Multi-homing. If users can easily use multiple platforms simultaneously, no single platform captures all the value. People have accounts on LinkedIn, Instagram, Twitter, TikTok, and threads—each serves different purposes.

Localized learning. Tesla’s data flywheel helps it in the geographies where Teslas are common. But driving conditions in Mumbai differ from Miami. Local competitors might build better local models.

Diminishing returns to data. After some point, more data doesn’t proportionally improve algorithms. The millionth driving hour matters less than the first thousand.

Open-source alternatives. Llama, Mistral, and other open models provide a floor beneath which closed models can’t fall. You can always switch to something that’s 80% as good for free.

Regulation. The EU’s Digital Markets Act explicitly targets network-effect monopolies. Interoperability requirements, data portability, and anti-self-preferencing rules can reduce switching costs.

But notice what all these mitigations have in common: they require intentional design. Markets left to their own devices don’t naturally resist network-effect concentration. You need either lucky market conditions (multi-homing opportunities, local differentiation) or deliberate intervention (open-source movements, regulatory action).

The Foundation Approach: Abundant Infrastructure, No Gatekeepers

The Unscarcity framework tackles network effects by removing essential services from network-effect dynamics entirely.

The Foundation (90% of resources) provides housing, food, energy, healthcare, and coordination as infrastructure—not products mediated by platforms. You don’t shop for housing on Zillow; you get housing. You don’t compare health plans on a marketplace; you get healthcare.

This matters because network effects require markets. They’re a property of competition—of platforms fighting for users, of users choosing between options. If there’s no competition because everyone gets access automatically, network effects have nothing to latch onto.

Consider the analogy to roads. Driving exhibits network effects—more drivers justify more infrastructure, more infrastructure attracts more drivers. But we don’t usually let one company own all roads and charge monopoly tolls. We treat roads as public infrastructure, accessible to all, funded collectively.

The Foundation applies that logic to survival needs. Abundance makes these goods cheap to produce. Direct provision ensures they’re cheap to access. No platform sits in the middle extracting value.

The Ascent (10% of resources)—the layer of genuine scarcity like frontier research, space exploration, and consciousness expansion—does operate through competitive mechanisms. But even here, the Impact system builds in decay (achievements fade over time) and diversity requirements (the Diversity Guard prevents any single faction from capturing control). Network effects are channeled rather than eliminated.

What This Means for the Transition

If you’re reading this in 2025, the network-effects battles are already underway. OpenAI, Anthropic, Google, Meta, and a handful of others are racing to establish AI platforms that could define the cognitive infrastructure of the 21st century.

The stakes are civilization-scale:

Scenario A (Star Wars): One or a few AI platforms achieve unassailable network effects. They control not just software but the data, the models, and eventually the robots and energy systems. Abundance exists, but access is gated. A technological aristocracy emerges, and the masses are economically irrelevant but biologically alive—kept docile with subsistence-level benefits.

Scenario B (Star Trek): Abundance technologies are deliberately structured to prevent network-effect monopolies. Open-source AI, interoperable platforms, public infrastructure for essentials. Network effects still exist in luxury and frontier domains, but the Foundation is protected.

We’re currently on a trajectory toward Scenario A. The capital, the talent, and the regulatory capture all favor concentration. OpenAI attracted 78% of daily unique visitors to core AI model websites in recent surveys. ChatGPT has 700 million weekly active users.

But trajectories can change. Open-source models are closing the gap. Regulatory frameworks are emerging. And more people are starting to understand that the structure of AI platforms will determine whether abundance technology liberates humanity or creates the most sophisticated system of concentrated power in history.

Network effects are physics. How we respond to them is politics.

Connection to the Unscarcity vision: Network effects explain why simply making things abundant isn’t enough—distribution channels can become extraction channels. The Foundation bypasses network-effect dynamics for essentials by providing infrastructure directly rather than through markets. The Ascent preserves competition for meaning and frontier achievement but uses Impact decay and the Diversity Guard to prevent any faction from capturing control permanently. This is how you get abundance without feudalism: by recognizing that network effects are real, powerful, and dangerous—then architecting around them deliberately.

Network Effects: Why Facebook Won and Why AI Monopolies Loom