Why software is becoming infrastructure
Software used to be a product you bought. Then it became a service you rented. The next stage is stranger and more interesting: software as a layer of settled infrastructure that nobody owns and everybody builds on.
Three stages of software
In the first stage, software was an artifact: a box, a license, a version number. You bought it once and it ran until your hardware died. The second stage made software a service: subscriptions, meters, and — the defining property — a relationship. The vendor kept a connection to the running system, and that connection became the business.
Each stage moved value from the artifact to the substrate. The box was the product; the service was the relationship; and the substrate is what remains when both the box and the relationship become commodities. We are entering the third stage now, and its shape is only partially visible.
What infrastructure means
Infrastructure is software that has stopped being a decision. Nobody chooses whether to use TCP/IP, the file system, or the relational database as a category. The interesting question moved up a level — not whether to use these things, but what to build with them. When a layer settles, competition on that layer ends and competition on the layer above begins.
Machine intelligence is settling in exactly this sense. A capable model is becoming something you assume, like storage or bandwidth — a commodity input whose price falls while its quality rises. The evidence is not in any single announcement but in the trajectory: every capability that commands a premium this year is a commodity API call within three.
The consequences
If intelligence is infrastructure, then the scarce resources are the ones infrastructure cannot supply: distribution, trust, ownership of the customer relationship, and the accumulated context of a real business. Software companies that are thin wrappers over rented infrastructure will be competed away, because their entire cost structure is somebody else's price list.
The durable position is to own a layer. Not the intelligence layer — that is settling — but something built on it that is hard to settle: a corpus of proprietary data, a mechanical advantage in a specific domain, a system whose value grows with use in a way a competitor cannot copy by renting the same APIs.
Why this shapes the lab
Fed Labs was organized around this observation. Research into what is settling, development on the layers that are not, and ownership of the result — because infrastructure rewards exactly the entity that holds the asset while the layer above it gets rebuilt every few years.
The registry on this site is the honest, partial view of what that looks like in practice.