Network Effects
Network effects occur when a product, platform, or ecosystem becomes more valuable as more participants join it. They are the single most powerful driver of market concentration in technology and the primary moat for platform businesses — but their character varies enormously depending on the architecture of the network itself.
Laws of Network Value
The simplest formulation is Metcalfe's Law: the value of a network scales with the square of the number of connected users (n²). Each new node adds value to every existing node, creating a geometric growth curve that explains why early investment in user growth dominates platform strategy.
Reed's Law goes further, arguing that Metcalfe understates value because it ignores subgroup formation. The number of possible subgroups in a network of n users scales as 2ⁿ — meaning networks that enable group formation (guilds in games, Discord servers, Slack workspaces) grow value even faster than Metcalfe predicts.
But even Reed's Law misses something: not all subgroups are equally valuable, and not all networks enable the same kinds of interaction. A more complete view considers the degree to which a network facilitates emergent interconnections — the freedom for participants to create, trade, collaborate, and build novel structures the platform designers never anticipated. The richer these emergent connections, the greater the resulting creativity, innovation, and wealth generation. This is what distinguishes a transformative platform from a merely large one.
Emergent vs. Constrained Networks
Network architectures fall on a spectrum. At one end are hub-and-spoke (or "star") networks — centralized systems where most interactions flow through a single coordinator. Traditional broadcast media, early web portals, and many walled-garden platforms fit this pattern. Adding nodes increases the hub's load but contributes little new value between the spokes.
At the other end are scale-free networks — distributed systems where nodes connect freely, power-law distributions emerge naturally, and subgroups self-organize. The open internet, open-source ecosystems, and peer-to-peer markets follow this pattern. These networks produce emergent behavior: user-generated content, bottom-up virtual economies, community-driven innovation, and novel use cases the original architects never anticipated.
Most real platforms sit somewhere between these poles, and where they land determines the character of their network effects.
Internalized vs. Externalized Network Effects
A critical distinction is whether a platform's network effects are internalized or externalized. Internalized effects occur when value accumulates inside the platform's boundaries — social graphs locked in a walled garden, content that exists only within one app, identity tied to a single service. Facebook and early Roblox exemplify this: user investment makes exit costly, concentrating value inside the platform.
Externalized effects occur when a platform makes its participants more successful in the broader ecosystem. Microsoft's developer tools, Shopify's merchant infrastructure, and the Model Context Protocol ecosystem exemplify this: they create value that flows outward, making participants more capable beyond the platform itself. As Bill Gates observed: a true platform is one where the economic value of everybody who uses it exceeds the value of the company that creates it.
Externalized network effects are harder to build but more durable. They create loyalty through empowerment rather than lock-in — a strategic distinction increasingly relevant as regulations target walled gardens and interoperability becomes a competitive advantage.
Commoditized vs. Differentiated Supply
The nature of the supply side matters enormously. Platforms with commoditized supply — where suppliers are interchangeable and compete primarily on price — tend toward the Amazon model: the platform aggregates demand, controls discovery, and extracts value through take rates and advertising. Suppliers become dependent because the platform owns the customer relationship.
Platforms with differentiated supply — where each creator, developer, or merchant brings unique value — follow a different pattern. Here the platform succeeds by helping suppliers build direct relationships with their audiences. Shopify, Substack, and Unity follow this model: they provide infrastructure that amplifies what makes each supplier distinctive, rather than homogenizing supply into a searchable commodity.
This distinction maps directly to discovery: commoditized supply favors algorithmic ranking and attention-economy dynamics; differentiated supply favors recommendation, curation, and deep intent matching — increasingly powered by AI search and agentic commerce.
Network Effects in the Agent Economy
The emergence of AI agents introduces network effect dynamics unlike anything we have seen before. The Model Context Protocol ecosystem exhibits classic cross-side effects: more MCP servers attract more agent developers, which attracts more tool builders. But there is a deeper dynamic at work — agents themselves become network participants, capable of discovering services, negotiating transactions, and composing multi-step workflows autonomously.
This accelerates the transition from internalized to externalized network effects. An agent does not care about a walled garden's social graph; it cares about structured APIs, clear tool descriptions, and composable services. Platforms that expose their capabilities as agent-accessible tools gain network effects from the entire agent ecosystem, while closed platforms get bypassed. The agentic web is itself a massive network effect play: the more tools, APIs, and structured data available on the open web, the more capable agents become, which drives more investment in agent-accessible infrastructure.
Open-source AI models gain their own form of network effects through community fine-tuning, shared benchmarks, and collective feedback loops. Each contributor makes the model better for everyone else — a pattern that compounds as deflationary technology drops the cost of training and inference, pulling more participants into the ecosystem.
The long-term implication: the most valuable networks of the next decade will be those that facilitate emergent connections not just among humans, but among humans, agents, and the structured services they interact with — creating compounding returns that no single company can replicate.