Every fifteen years or so, a new substrate appears underneath the software economy. Mainframes gave way to PCs. PCs gave way to the web. The web gave way to mobile and the cloud. Each time, the firms that captured the most durable value weren't the ones building on top of the new layer — they were the ones building the layer itself.

We're inside one of those transitions right now. The cost curves of compute, the architecture of model training, and the topology of where data and inference actually live are being rewritten in real time. Most of the public conversation is fixated on applications. The interesting work is happening one or two levels deeper.

Lightside concentrates on three intersecting surfaces: the systems that train and serve models efficiently, the protocols that move data and coordinate compute across networks, and the primitives that let small teams ship infrastructure that previously required hyperscaler budgets.

These aren't separate categories. The most interesting companies sit at the seams — a routing layer that doubles as a market for compute; a runtime that becomes a distribution channel; an inference engine that quietly turns into the system of record. We pay close attention to the seams.

Pre-seed and seed, before the company has a name in the market. The decision to back a foundational technology is almost always made on technical taste, not traction. By the time the metrics arrive, the round is crowded and the price has decoupled from the work.

We are comfortable holding conviction for years before consensus catches up. Our checks are sized to be early, not large. The job is to be the first institutional believer, then to compound usefulness from there.

We are operators first. The most useful thing we can offer between board meetings is sharp feedback on the technical roadmap, introductions to the next ten engineers, and a direct line when something is on fire at 11pm. We do not run a platform team. We answer the phone.

We hold a small, deliberate portfolio. Each company gets real attention — not a logo on a slide. That constraint is intentional and it is the discipline that makes the rest of the model work.

Infrastructure is never neutral. The architectural choices made now — about who can train, who can serve, who can audit, who can fork — will compound for a generation. We'd rather back the founders making those choices on purpose than watch them get made by default.

If you are building at that layer, with that intent, we would like to meet you early.