Independent Standard · Atlanta, GA

The first real unit
for measuring
compute energy.

CEH™ (Compute Energy Hour) expresses how efficiently
energy converts to AI output — kWh per token, per frame,
per TFLOP. Energy-agnostic. Reproducible. Open.

0
Prior independent standard existed
8×
Efficiency variance across current hardware market
10
Hardware configs benchmarked in v1.0
The Standard
CEH™ = ( TDP × Units × Util% × PUE ) ÷ 1,000 ÷ Output/hr
kWh consumed ÷ compute output units per hour
CEH™ Cost = CEH™ × $/kWh Any energy source — grid, BTM, nuclear, solar
CEH™ Carbon = CEH™ × kg CO₂/kWh Grams CO₂ per token, per frame, per TFLOP
Framework Inputs
⚡
Hardware TDP
Per-unit power draw, applied across unit count. Source: manufacturer spec or measured.
📊
Utilization Rate
Measured average during benchmark window — not theoretical peak.
🏢
PUE (Facility Overhead)
Hyperscale 1.1–1.2 · Edge/CoLo 1.4–1.7 · BTM 1.05–1.1
1
Compute Output Unit (COU)
Tokens/hr (inference) · TFLOPS (training) · Frames (rendering)
CEH™ Benchmark Index · v1.0
LLM inference · Llama-class · batch 8 · vLLM · 8-GPU node · PUE 1.2
Hardware CEH™ Rate Grid $/M tok Grade
NVIDIA B200Blackwell · 2025 1.07 × 10-10 $0.009 S
NVIDIA H200Hopper · 2024 2.21 × 10-10 $0.019 A
Google TPU v4Google · 2023 2.63 × 10-10 $0.022 A
NVIDIA H100 SXMHopper · 2023 3.18 × 10-10 $0.027 A
NVIDIA L40SAda Lovelace · 2023 5.43 × 10-10 $0.046 B
AMD MI300XCDNA3 · 2024 6.11 × 10-10 $0.052 B
NVIDIA A100 SXMAmpere · 2021 8.92 × 10-10 $0.076 C
CPU Cluster (EPYC)x86 Baseline · No Accelerator 6.11 × 10-6 $0.52 F
S Exceptional < 2.0×10-10
A Excellent 2.0 – 4.0×10-10
B Good 4.0 – 8.0×10-10
C Marginal 8.0 – 15.0×10-10
F Unacceptable > 30.0×10-10
PUE 1.2 applied uniformly. Grid = $0.085/kWh US commercial average (EIA 2025). Carbon intensity: 0.386 kg CO₂/kWh (EPA eGRID 2024). Throughput: MLPerf Inference v5.1, CUDO Compute, Koyeb, Spheron (2025–26).
01 — Operators

Adopt the Standard.
Publish your index.

Data center operators, neo-clouds, and BTM developers publish CEH™-certified benchmark data on their own infrastructure. Differentiate on energy efficiency — not just $/GPU-hr. Get listed in the public CEH™ Index with your certification mark.

Certifiable benchmark Quantified PUE contribution Green compute positioning Hardware refresh ROI
Example: BTM-Optimized Facility
$0.035 /kWh
vs. grid avg $0.085/kWh
59% lower CEH™ Cost vs. grid-average facility
02 — Capital

Invest in the
infrastructure layer.

DGE&I is the first commercial BTM compute infrastructure deployment built on CEH™ measurement. 300 MW anchor offtake. Structural energy cost advantage. Qualified capital partners only.

300 MW anchor offtake BTM power advantage Energy-rate underwriting Structural cost edge
Generation
300 MW
BTM anchor offtake
Energy Source
Any
Grid, gas, nuclear, solar
CEH™ Benchmark
Grade A+
H200 / B200 class
03 — Research

Co-publish the
research.

CEH™ methodology is open, versioned, and dated. Independent replication, co-publication, and standards body submission — MLCommons, Green Grid, IEEE, and university partners invited.

Open methodology Independent validation Co-publication track Standards body submission
✓
CEH™ Standard v1.0
April 2026 · Complete
▶
Independent Validation
Q2–Q3 2026 · In Progress
▷
CEH™ Index v2.0
2027 · Planned
▷
Industry Reference Standard
2028+ · Vision
The gap in AI infrastructure

Compute has no energy unit.
Until now.

Every metric used to evaluate AI infrastructure today measures only one dimension. GPU-hour pricing bakes in margin and contractual terms. FLOPS have no energy layer. kWh carry no output dimension. CEH™ fills that gap — bridging hardware performance, energy economics, and carbon accounting into a single, auditable number owned by the industry.

Energy-Agnostic Workload-Portable Auditable Reproducible Open Methodology Versioned