CIB · 0–100
–
Is compute a commodity yet? One number, updated every minute.
A live 0–100 commoditization score built from the four basis risks of compute derivatives — settlement, obsolescence, quality and calendar — each measured and charted below from independent GPU benchmarks: Ornn's transacted index, CCIR's reference rates and term structure, ComputePrices' 91-provider list, and Vast.ai's live marketplace.
The pin is compute today; the right edge is gold. The score re-computes every minute and archives a tick every ~5 — Vast reprices by the minute, ComputePrices through the day, Ornn prints 20:00 UTC, CCIR by 07:30 ET.
A cash-settled contract converges to its index — and nothing else. If independent benchmarks disagree on what an H100-hour costs, that disagreement is the risk a hedger eats. We chart the max−min spread across the prints, as % of the median.
No trade converts an H100-hour into a B200-hour — so when the vintage ratios drift, an old-SKU hedge stops tracking a new-SKU book and nothing arbitrages the gap shut. This is the one basis with no anchor. We chart the cross-vintage price ratios from Ornn's transacted prints.
If SKUs moved together, one liquid contract could hedge every fleet — like grades of crude against WTI. We chart the average pairwise 30-day correlation of daily moves across H100, H200, B200 and A100. High = one benchmark carries the market; low = every SKU is its own island.
CCIR publishes committed asks by tenor daily. Once our archive is deep enough to compare a 3-month quote to the spot that actually printed 3 months later, this leg scores the curve's realized error — until then it shows the live term structure and counts down.
Four basis risks, four charts, one live score — how close compute is to being a real commodity.