Indices/AI & Compute
LIVE

SPREAD — The Silicon Spread

What does the market pay above the electricity?

An H100-hour is electricity transformed into work — compute's version of the spark spread that prices every gas power plant. We chart the transacted GPU price (Ornn, the ICE futures reference) over its industrial-power floor (EIA, datacenter-region). Everything above the floor is scarcity, capital recovery and margin — the premium the market puts on intelligence.

Silicon spread · price ÷ floor
×
H100 transacted $/hr
Power floor $/hr
Spread $/hr
The analogy
Power markets price generators by the spark spread — electricity out minus gas in. Compute has the same physics: kilowatt-hours in, GPU-hours out.
KOPS SPREAD
Transacted H100 price over the raw power floor — the scarcity premium of compute, one number, daily. When it compresses, compute is commoditizing.
Valuable for
Neoclouds pricing capacity, lenders underwriting GPU debt, and anyone asking when the AI margin stack normalizes.
Live index

The premium over physics.

The multiple of the transacted H100 price over its power floor, daily from Ornn's prints. The floor is nearly constant — this chart is the scarcity premium itself.

Silicon spread · price ÷ power floordaily · Ornn prints 20:00 UTC
H100 · transacted $/hr vs the floor91d archived window
The margin stack

Electricity in, intelligence out.

The full transformation chain the AI economy runs on — each layer priced live by a KOPS index.

Electricity 0.7 kW card × 1.3 PUE × EIA industrial (TX+VA composite)
GPU-hour H100 transacted (Ornn · ICE futures reference) · COMPUTE
Tokens blended market $/M tokens · KOST
each layer's margin over the one below = where the AI build-out's economics live
Reading it. The floor is physics and doesn't move much. The GPU layer's premium over it — currently × — is what pays for silicon, datacenters, debt and margin. If Jevons wins, volume grows while this multiple grinds down; if scarcity wins, it stays wide. The commoditization scoreboard for the whole stack is CIB.
Index methodology

One formula, all public inputs.

01

The floor

An H100 SXM draws ~0.7 kW; datacenter overhead adds 1.3× PUE; industrial power comes live from EIA for the biggest datacenter states (TX / VA). Floor = 0.7 × 1.3 × $/kWh — the price of the raw electricity in a GPU-hour.

02

The price

The numerator is transacted, not posted: Ornn's H100 index, built from printed transactions and referenced by ICE's compute futures. We archive its rolling window daily so the history stays free.

03

The spread

SPREAD = price − floor, quoted as $/hr and as a multiple. The multiple is the headline: it's dimensionless, comparable across power regimes, and falls as compute commoditizes.

SPREAD = H100transacted − ( 0.7 kW × 1.3 PUE × $/kWh )
compute's spark spread — the scarcity premium of intelligence over its electricity.
Sources
Ornn transacted H100 (ICE futures reference)
H100 SXM 0.7 kW · PUE 1.3
EIA industrial power (US / TX / VA)
KOST token price (downstream layer)
The slate

The rest of the compute stack.

The spark spread of the AI age.

Electricity in, intelligence out — the premium in between, measured daily.