The intelligence layer for AI infrastructure.

What Clarke is

Clarke is a structured, decision-grade view of AI compute supply. We track facilities, normalize their power, estimate their compute in one unit, dense FP8, and forecast their delivery — facility by facility, market by market, operator by operator.

Every figure is paired with the source behind it. Where official data is missing, we publish a Clarke Estimate, say what it rests on, and show the working step by step. Future capacity is drawn as steps toward the planned total, each labelled for how firm it is, and a date Clarke estimated is marked as an estimate.

Why it matters

AI infrastructure is being built at speeds no analyst stack was designed to cover. Utility interconnection backlogs now hold 5–10× more capacity than will ever be built. Capacity model projections for 2030 diverge by a factor of 40. Site power contracts get confused with actual IT capacity. Marketing announcements get confused with operational reality.

The result: billion-dollar planning decisions made against intelligence that doesn't separate signal from speculation. Clarke exists to fix that — with a shared vocabulary (Clarke Standards), a transparent methodology, and a record that's defensible at a board meeting rather than a LinkedIn post.

Who Clarke is for

How we work

Three principles run through everything Clarke publishes:

Sources are mandatory. Every figure carries the sources behind it. Where a source is no longer online, Clarke holds a copy and says so.

Uncertainty is shown, not hidden. Every figure carries a label saying how well sourced it is — Confirmed, Operator Stated, Clarke Estimate, Planned or Potential — and every Clarke Estimate states its basis. Withheld means withheld — we don't fill gaps with guesses.

Standards are open. The Clarke vocabulary is free to reference and build on with attribution. The scoring models, grid feasibility models, and pipeline architecture stay proprietary. The definitions don't.

The Clarke team

Clarke is being built in New York. The platform is currently in Alpha; the public index is open, deeper records are admitted through alpha applications. Clarke is a NVIDIA Inception Program member.