When the next site costs almost nothing to launch

The shift that drives the whole economics is that you stop building websites and start building the machine that builds them. Once there's a shared engine that can be pointed at a new dataset, a new theme, and a new domain, a site is no longer a thing you construct — it's a thing you configure. The value has moved from any individual site into the platform underneath, and the sites become expressions of that platform rather than separate creations. The asset is the machine; each site is the machine, aimed at something new.

This is why the marginal cost collapses. The expensive, one-time work — the engine, the templates, the deploy, the data pipeline — is amortized across every site the platform will ever run. Adding one more doesn't re-pay that cost; it just uses it again. So the cost of the next site is only the cost of its configuration and its data, which is small, and gets smaller relative to the platform investment with every site you add. The economics of the network are the economics of a fixed cost divided by a growing number.

I have watched this happen in my own work, and the clearest case is the Atlas Network. I built BBQ Atlas first — a map, entity pages, the crawlable-URL scaffolding — and when I wanted a second, unrelated atlas about military history, I extracted the map-and-entity engine instead of rebuilding it. The War Atlas now runs on that same engine, and a third atlas would be a dataset, a theme expressed as a few OKLCH color tokens, and a domain — not an application. The subjects share nothing; the machine is identical, and that sameness is what makes the platform the asset worth owning.

Stop building websites and start building the machine that builds them. Then the next site isn't a project — it's a configuration, and its cost is a rounding error.

Here's the danger that took me a painful lesson to internalize. When launching a site costs almost nothing, the natural instinct is to launch a lot of them — why not, they're free. But a site that's cheap to launch is not necessarily a site that should exist, and the near-zero cost removes the friction that used to make you ask whether it should. I've generated large numbers of templated sites and pages precisely because I could, and watched them fail and even drag down the good work, because cheap-to-make quietly became should-make in my head. The gun went off in my own foot.

The problem is that the cost didn't actually disappear — it moved. It moved from the launching to the consequences: a thin, redundant, low-value site still costs you, just later and less visibly, in reputation and in the drag it puts on everything around it. Near-zero marginal cost to launch is not near-zero cost overall; it's a bill that comes due downstream, in quality and trust, for every site you launched because you could rather than because you should.

I can put numbers on that bill, because two of my own domains paid it. On NameBayBay I pushed roughly a hundred thousand programmatic URLs at Google and IndexNow as fast as I could; impressions spiked to roughly eighteen thousand almost overnight, then collapsed to a few dozen a month — effectively erased from search, and still not recovered. CivicsMatter is the crueler version of the same lesson: the data there was genuinely real, eighteen thousand actual districts and legislators and offices, and it still climbed past seventy thousand impressions in a month, then fell back under twelve thousand two months later and kept sliding. Real data was necessary. It was nowhere near sufficient.

The reason a near-free site can cost that much is that search engines increasingly grade at the domain level, not the page level. A young domain that absorbs tens of thousands of thin pages overnight looks exactly like the content farm it isn't, and the whole account gets marked down at once. The asymmetry is what makes near-zero launch cost so treacherous: getting flagged is fast and getting un-flagged is slow. A classifier can decide your domain is low-value in a single crawl; convincing it otherwise is months of publishing nothing but good pages while it holds a prior on you. The launch was free. The demotion was not.

So the discipline has to relocate to the one question that's still expensive: should this site exist at all? When the platform made launching trivial, it didn't make deserving-to-exist trivial. Each new site still needs a real reason — a genuine dataset, a genuine audience, a genuine differentiation — and the near-zero launch cost means nothing is stopping you from skipping that question except your own resolve to keep asking it. The gate that used to be enforced by expense now has to be enforced by judgment.

In practice that means a per-site quality bar that the cheap launching can't bypass: a new site gets to exist only if it clears a threshold of genuine value, regardless of how easy it is to spin up. The platform makes it possible to launch fifty sites; the discipline is refusing to launch the forty that shouldn't exist just because you could. The scarcity that used to come for free from cost now has to be manufactured deliberately, because the economics stopped supplying it.

The mechanism I use is deliberately boring, and aimed at pace as much as quality. Before I let a template loose on a whole dataset, I prove that a single page — or a small cluster — can actually rank. I ramp volume against demonstrated indexing, not against how many rows happen to sit in the table. And I treat every domain as having a trust budget I can overdraw: publish faster than the site has earned, and the whole account freezes. The near-zero cost tempts me to skip all three checks; the discipline is running them anyway.

The gate is made of specifics, not slogans. On The War Atlas, every battle traces to a primary source, and force numbers appear as source-cited ranges with a confidence grade rather than invented precision — a page earns the right to exist by being verifiably real, not by filling a slot. The point is to rebuild, by hand, the scarcity the economics stopped supplying. Cost used to say no for me. Now nothing says no except a standard I hold far higher than the launch cost will ever require.

The mature view of a site network is that it's not an achievement of volume — it's an achievement of two things that sound contradictory: sameness and restraint. Sameness, because the whole economic magic comes from every site being the same engine underneath, so the platform investment pays off across all of them. Restraint, because the near-zero cost that sameness buys you is only an advantage if you refuse to abuse it. A network of one great engine running many genuinely-worth-existing sites is a compounding asset. A network of one great engine running a hundred sites that shouldn't exist is a liability with good infrastructure.

So the economics of a network are seductive and double-edged, and understanding both edges is the whole game. The marginal cost going to zero is what makes the model powerful — one person really can run infrastructure that used to need a team. And it's what makes the model dangerous — the same leverage that scales your good judgment scales your bad judgment just as efficiently. The platform is the asset; restraint is what keeps it one. Build the machine, and then be far more disciplined about what you point it at than the near-zero cost will ever require you to be.