Network effects and ecosystem

The mechanism

The engine here is that other people's work piles up on top of yours, so every new participant makes the platform more useful to the next one and the advantage grows without the company paying for each increment. NVIDIA spent years giving CUDA away to researchers, and once roughly five million developers had written code against it for everything from molecular dynamics to weather simulation, that accumulated software became the reason the next lab buys an NVIDIA chip instead of a cheaper one. TSMC built the same loop on the supply side through its Open Innovation Platform: a better process lets its fabless customers ship better products and grow, which funds the next process node, which pulls in more customers, and the hosts describe this flywheel as one that "goes slow," because the compounding takes years of effort before it turns on its own.

The tension

An ecosystem is the one asset a company cannot simply buy or rush, which is its strength while it is being built and its trap once the company depends on it. The same five million CUDA programs that lock customers in also lock NVIDIA in, because breaking compatibility would strand the very work that created the moat, so the platform owner ends up serving the ecosystem as much as it is served by it. A flywheel that took years to start can also coast for years on momentum after the underlying edge has eroded, which hides a decline until it is already far along. And the compounding runs backward on the same logic it ran forward: if enough developers or customers leave, the thing that made the platform valuable to everyone else leaves with them.