What is the ACI Global Compute Index?

ACI (AI Compute Index) is a market thermometer for enterprise-grade GPU compute (B300/H100/H200 etc.): it blends listing, liquidity and yield pillars to reflect real-time market heat. Base 2025-09-01 = 100; live value updates every minute.

How is ACI calculated?

ACI = 0.4×Listing score + 0.3×Liquidity score + 0.3×Yield score. Listing score is driven by hall spot prices via the median single-rack price (0.8) plus listing base (0.2); liquidity uses 7-day rolling sell-through and buy heat plus ecosystem; yield blends rental level and payback months.

How is ACI different from the score column (73.8/A+)?

ACI is a market-wide thermometer for bulk compute (updated every minute); the score column is a per-card investment scorecard (payback 40% + appreciation 30% + sell-through 30%, updated daily) — two independent formulas. Consumer SKUs have a separate ACI-C index.

How often is ACI updated?

The live value updates every minute (same source across ticker, homepage and detail pages); the daily history line is anchored to 2025-09-01 = 100 for continuity.

中文 English

← Back to market home

ACI Global Compute Index · Definition & Methodology

v1.0 finalized · adopted 2026-10-28 · live value updates every minute · this page = methodology anchor (see the market home ticker for the live value)

What is ACI (AI Compute Index)? ACI is a market thermometer for enterprise-grade GPU compute (B300/H100/H200 etc.): it blends three pillars — listing, liquidity and yield — to reflect real-time heat in the enterprise compute market. Base 2025-09-01 = 100; the live value updates every minute and is served from the same source across the ticker, homepage and detail pages.

Calculation formula

Master formula: ACI = 0.4×P(Listing) + 0.3×L(Liquidity) + 0.3×Y(Yield)

Pillar	Composition	Notes
P Listing (40%)	`0.8×listed_s + 0.2×100`	listed_s = 403.3 × (today's NVL16 median per rack ÷ 1812.3 baseline median) — driven by hall spot prices
L Liquidity (30%)	`0.5×sold7 + 0.3×buy7 + 0.2×eco`	sold7/buy7 = 100×(1+log10(7-day rolling ÷ baseline)), capped at 250 — sell-through and buy heat
Y Yield (30%)	`0.6×rent + 0.4×payback_s`	payback_s = clamp((60 − payback months) ÷ 48 × 100, 0, 100)

Base period: 2025-09-01 = 100; the minute-level live value joins the daily line seamlessly (bootstrapped baseline).

Components & samples

Listing sample: the NVL16 per-rack median price serves as the market-wide representative sample (driven by hall spot prices).
Liquidity sample: rolling 7-day sell-through and buy data; the ecosystem and rental pillars are snapshot constants at filing time (honestly disclosed, pending live feeds).
Weight/constant changes go through the v1.1/v2.0 process: update the charter → pipeline applies → all windows notified; ad-hoc script-side changes are prohibited.

ACI vs Scorecard vs ACI-C

Name	What it is	Cadence
ACI (the big number)	Bulk thermometer: market-wide index for datacenter cards (B300/H100/H200 etc.)	Every minute
Scorecard (score column 73.8/A+)	Per-card investment score: payback 40 / appreciation 30 / sell-through 30	Daily
ACI-C	Consumer index: RTX 5090/4090/memory/SSD/Mac	Daily · base 100

FAQ

What is ACI? A market-wide thermometer for bulk GPU compute (see the answer block above).
How is it calculated? Three weighted pillars: Listing 40% + Liquidity 30% + Yield 30% (formula above).
Why is it different from the score column? The score column is a daily per-card investment scorecard; ACI is a minute-level market thermometer — two independent formulas.
How often is it updated? Live value every minute; the daily history is anchored to 2025-09-01 = 100.

Methodology source of truth: docs/ACI_INDEX_V1.md (v1.0 charter). Live index value on the market home.