Between the wafer and the rack, everything is a lead time.

Accelerators, optics, switchgear, transformers and cooling plant are what a compute schedule is actually made of. Nuclir tracks them as delivery dates, not datasheets.

The scarcest part in an AI build is rarely the chip.

A rack needs power delivered to it, cooled around it, and switched in front of it. Large power transformers, medium-voltage switchgear and industrial chillers are all built by a small number of qualified plants on lead times measured in years, and demand for them is now set by a market that did not exist when those plants were sized. That is a textbook long-lead item problem arriving in an industry unused to it.

Optics and networking behave the same way one layer up. Transceiver supply, fibre plant and switch silicon each have their own qualification cycles, and a cluster is only as deliverable as the slowest of them. The result is that a data center schedule is decided by four or five components nobody puts in the announcement.

A part number is not an entity. A production line is.

Tracking hardware by SKU tells you what exists. Tracking it by the line that makes it tells you what can arrive, and when. Nuclir resolves manufacturers, plants, qualified alternates and observed shipments into one model, so a component with three nominal suppliers and one qualified plant is visible as what it is - a sole source.

Each piece of equipment stays connected to the sites waiting on it and the routes carrying it. That is what turns a shipping delay into a dated change on a specific build rather than a line item in a freight report nobody reads.

Four classes of equipment that set dates.

Each is resolved to the plants that build it and the sites waiting on it, which is what makes a delay a dated event rather than a rumour.

Accelerators and systems

Compute hardware from qualified assembly through shipment, tied to the campuses it is destined for.

Optics and networking

Transceivers, switch silicon and fibre plant - the layer that decides whether a cluster is one cluster.

Power equipment

Large transformers, switchgear and generation, built by few plants on lead times measured in years.

Cooling plant

Chillers, heat rejection and liquid cooling, where density targets meet what can actually be procured.

Orders, shipments, and the ground itself.

Equipment leaves a trail before it arrives: procurement notices, customs and manifest records, port movements, and the physical footprint of a plant expanding to meet demand. Read separately these are four datasets; resolved against one another they are a delivery forecast with evidence attached.

None of it is secret and none of it is predictive. It is the observable record, assembled while it is still current and delivered from the edge closest to the requesting node - which is the only form in which a lead-time signal is worth anything. What is genuinely unknowable, we say is unknowable.

Evidence that keeps pace with the decision.

Evaluate Nuclir against the systems, markets, and decisions that matter to your organization. The result is current intelligence with the context needed to use it responsibly.

Request access Speak with the team