CCIR Compute Credit Index Research
Market Intelligence

Chip Economics

Earning power per spec-sheet unit.

Analytical decomposition of published reference rates — not a reference rate.

01 Four Lenses on One Rate

Capacity $ / GB-hr spread 56%

what the chip can hold — the size of model and context it can host

.02 .04 0y 3y 6y B300 B200 H200 H100 A100
Bandwidth $ / TB/s-hr spread 12%

how fast it moves data — the dimension inference workloads buy

.50 1.00 0y 3y 6y
flattest cross-section — see §02
Compute $ / PFLOP-hr spread 68%

how fast it does arithmetic — training speed

$2 $4 $6 0y 3y 6y
Power $ / kW-hr spread 43%

what it costs to keep lit — rent per watt against facility cost per watt

$2 $4 $6 0y 3y 6y
Blackwell Hopper Ampere same horizontal position = same generation in every panel · hover or focus any dot for all four readings

Neocloud (T2) guaranteed on-demand, US, generations on age (x, years) against each lens (y, from zero). Spread = (max − min) ÷ mean across the five generations. Example prints · 2026-07-07 · rates from the 2026-07-06 snapshot. Headline stat per published rule (mean where n ≥ 10, else median). The compute lens is denominated in petaFLOPs (PFLOP, 10^15 floating-point operations per second), dense, in the BF16 number format.

One observed number underlies every panel: the Neocloud guaranteed on-demand rate, in dollars per GPU-hour. Each panel divides that rate by one spec-sheet denominator — and where a denominator makes the cross-section flat, that denominator is the dimension the market prices.

GenerationAge$/GPU-hrn$/GB-hr$/TB/s-hr$/PFLOP-hr$/kW-hr
B300~0.8y$7.8550.02730.98~3.5~6.0
B200~1.4y$6.9280.03840.873.16.92
H200~2.1y$4.18140.02960.874.25.97
H100~3.7y$3.22200.04030.963.34.60
A100 80GB~5.7y$1.80130.02250.905.84.50

CRI-T2-{chip}-{form}-GTD-OD-US-{GB|BW|PF|KW} · derived Compute Rate Index (CRI) series carry the underlying citable cell in the first seven segments; strip the lens suffix to reach the parent. $/GB-hr computed from registry high-bandwidth memory (HBM) capacity (80 / 80 / 141 / 180 / 288 GB).

02 The Bandwidth Band

Re-denominated per unit of memory bandwidth, datacenter silicon from eight months to 5.7 years of age clears inside $0.87–0.98 per TB/s-hour — ±6% around a $0.92 mean. The band is the test: a spread that holds tight means the age discount remains priced away; a widening spread is the falsifiable marker firing. Members are disclosed on every exhibit.

The flatness is a like-for-like property. It holds across cells that share an operator segment, interruptibility grade, term, and region — the series named under each exhibit. Pooled cuts, marketplace minimums, or mixed interruptibility grades will not reproduce it: grade and tenor are different markets per unit of bandwidth, as everywhere else on this site.

0.80 0.85 0.90 0.95 1.00 0.98 — max 0.92 — mean ±6% 0.87 — min 0y 2y 4y 6y age axis zoomed: $0.80–1.02, not to zero B300 0.98 B200 0.87 H200 0.87 H100 SXM 0.96 A100 0.90
$/TB/s-hr against generation age; band = min–max across qualifying members, mean line at $0.92. Denominator: vendor nameplate memory bandwidth. Example prints · 2026-07-07 · rates from the 2026-07-06 snapshot.
Band history 2026-07-07 · first print · band mean $0.92 · spread 12% series begins +30d +60d +90d history accrues from launch — nothing is plotted here because nothing exists yet

CRB-T2-BW-ALL-OD-US · band row fields: band_min · band_max · band_mean · band_spread_pct · n_chips · member_chips. Emission gate n_chips ≥ 3, all members publication-qualified. Research-grade at launch (Shadow → Provisional).

03 Watt-Rent by Shell Class

Per unit of nameplate power, the panel sorts by shell class rather than by age — here the dispersion is the finding. Consumer silicon with no datacenter bundle rents near $1 per kW-hr; air-cooled datacenter silicon is flat at $4.50–4.66 across five years of age (A100 = H100 = L40S per watt); liquid-class silicon rents at $6–7. Watt-rent prices the shell a chip can occupy.

$0 $1 $2 $3 $4 $5 $6 $7 $/kW-hr LIQUID-CLASS $6–7 AIR-COOLED DC $4.50–4.66 MKT / CONSUMER ~$1.0–1.3 B200 6.92 B300 ~6.0 H200 5.97 † A100 4.50 · 5.7y H100 4.60 · 3.7y L40S 4.66 · 2.7y L4 5.83 ‡ 5090 / 4090 / 3090 · $1.0–1.2 (Mkt) A40 ~1.3 (Mkt)
Revenue per kW of nameplate thermal design power (TDP), by shell class — gross rent per unit of power envelope, not margin. † H200 hosts in dense-air or liquid shells. ‡ L4 is an edge part — any air shell — and sits outside the datacenter air band. Marketplace rows (Mkt) are Marketplace-segment prints, indicative. Example prints · 2026-07-07 · rates from the 2026-07-06 snapshot.
table view — watt-rent panel
GenerationTDP$/kW-hrShell class
B2001.0 kW6.92liquid
B300~1.3 kW~6.0liquid
H2000.7 kW5.97dense air / liquid
L4 (Mkt)72 W5.83any air / edge
L40S350 W4.66air
H1000.7 kW4.60air
A100 80GB400 W4.50air
A40 (Mkt)300 W~1.3air
5090 / 4090 / 3090 (Mkt)350–575 W1.0–1.2consumer

CRB-T2-KW-AIR-OD-US · CRB-T2-KW-LIQ-OD-US · cohort = cooling_class (air | liquid); consumer prints are Marketplace-segment context, not band members.

04 Economic Life — the Breakeven Altimeter

Watt-rent and cash operating cost are the same unit. A chip exits economic life when its watt-rent decays to the cash boundary beneath it — so one log-scale strip reads as an altimeter: today's rent bands above, the cash operating boundary below, and the gap between them is distance to shutdown. Air-cooled silicon today covers its cash operating cost ≈5–7×; the consumer band near $1 is the visible preview of late life, close above the boundary.

Cost-side basis. Chip TDP understates system draw: an 8×H100 server on NVIDIA's HGX server platform draws ≈10.2 kW against 5.6 kW of chip TDP — a ≈1.8× system factor (CPUs, NVSwitch, fans). Every cost rung below is restated per chip-TDP-kW through that factor, so both sides of the altimeter share the watt-rent lens's denominator. The factor is a disclosed election.
$ / TDP-kW-hr · log scale $8 $5 $2 $1 $0.50 $0.20 $0.10 rent decays toward cost ≈5–7× cash coverage LIQUID-CLASS RENT $6–7 B200 6.92 · B300 ~6.0 · H200 5.97 AIR-COOLED DC RENT $4.50–4.66 A100 4.50 · H100 4.60 · L40S 4.66 — flat across 5 yrs CONSUMER RENT ~$1.0–1.2 no datacenter bundle — the preview of late life SHUTDOWN BOUNDARY — CASH COST ≈$0.64–0.87 power + colo · sunk capex excluded · ≈$0.45–0.61 per H100-hr before staffing / ops elections COMPONENT — WHOLESALE COLO ASKING ≈$0.42–0.66 CBRE $165–265/kW-mo × 1.8 system draw COMPONENT — POWER-ONLY FLOOR ≈$0.22 EIA 8.66¢/kWh × PUE 1.4 × 1.8 system draw
Rent bands are CCIR prints (filled, member ticks at left edge); the cash boundary is a derivation from disclosed elections (tinted), its components external reference data (outlined), cited inline. Whether power is metered inside or beside the colo rate varies by contract — the power-only floor and the colo asking band are therefore shown separately. All cost rungs per chip-TDP-kW via the 1.8× system factor. Component sources: CBRE (the commercial-real-estate services firm) for the wholesale colo asking band; the U.S. Energy Information Administration (EIA) for the power price, with a power usage effectiveness (PUE) of 1.4 elected. Example prints · 2026-07-07 · rates from the 2026-07-06 snapshot; cost derivations verified 2026-07-07.
The crossing identity. rent per watt = band level × (chip TB/s ÷ chip kW). Watt-rent is the chip's breakeven cash operating cost: economic life ends when watt-rent decays to cash cost per TDP-kW — sunk capex excluded, which is why this boundary governs whether an installed chip keeps running. Both sides are $/TDP-kW-hr. CCIR publishes the left side daily; the cost side — colo terms, power price, staffing, the system factor, any decay election — is the reader's.

Worked illustration — hypothetical; every input elected by the reader

Elect a cash operating boundary c = $0.75/TDP-kW-hr (inside the combined cash band above) and a decay election g = 26%/yr (the cross-generation pace — see the caveat below). Air-band watt-rent today ≈ $4.58/TDP-kW-hr.

years to crossing = ln(watt-rent ÷ c) ÷ g = ln(4.58 ÷ 0.75) ÷ 0.26 ≈ 7

This is arithmetic on today's band against two elected inputs — a band-implied rent set against an elected cash cost. It is not a projection of any rate, and CCIR publishes neither election.

The honest gap. The −26% per year slope is a cross-generation reading — the pace of the bandwidth frontier across five generations on one date — not the band's own decay through time, which is the parameter the crossing actually needs. No one has that parameter, because the band has no published history. That is the argument for starting the series now: every month of publication shrinks the one unmeasured input. The band is also an observed equilibrium, not a law — in the 2023 shortage, frontier chips cleared far above any spec-sheet line. The flatness carries a date, which is the second argument for watching it.

05 Method & Caveats

Normalized values equal the published parent series value divided by the disclosed constant. The parent cell's pooling, stat, and gate decisions are inherited unchanged; denominators are vendor nameplate constants from the silicon registry and are never restated — a correction is a changelog entry, not a series break. Normalized series adopt the trimmed-mean stat (T2, shadow phase) from first print.

LensDenominatorA100 80GBH100 SXMH200B200B300Spec source
Capacity · $/GB-hrwhat the chip can hold — the size of model and context it can host HBM capacity (GB) 8080141180288 vendor datasheet · spec_source pending registry ext.
Bandwidth · $/TB/s-hrhow fast it moves data — the dimension inference workloads buy memory bandwidth (TB/s) 2.03.354.88.0~8.0 vendor datasheet · spec_source pending registry ext.
Compute · $/PFLOP-hrhow fast it does arithmetic — training speed dense BF16 (PFLOPS) 0.310.990.992.25~2.25 vendor datasheet · dense, not sparse · pending
Power · $/kW-hrwhat it costs to keep lit — rent per watt against facility cost per watt TDP (kW) 0.40.70.71.0~1.3 vendor datasheet · spec_source pending registry ext.

Denominators are variant-level: H100 SXM 3.35 TB/s and H100 PCIe 2.0 TB/s are different denominators. A chip without a disclosed denominator emits no row for that lens — no imputation.

Lens admission rule

A lens is admitted when its denominator is a disclosed vendor nameplate constant, it answers a question a credit reader has, and it tells a distinct story. Considered and excluded on record: token throughput (measured, not nameplate), interconnect / fabric (a cluster property with no per-chip denominator — the tier axis prices the fabric), PUE-adjusted power (PUE is the reader's election), and FP8 / FP4 precision variants (a precision election — disclosure, not columns).

Blind spots.

  • Denominators are nameplate constants — they capture none of a deployment's realized capacity, interconnect domain, or measured throughput.
  • System draw exceeds chip TDP (≈1.8× for HGX-class servers); TDP is the disclosed lens constant — delivered-power cost comparisons apply the system factor.
  • PUE, facility cost, and any decay election are the reader's; CCIR publishes prices, not the elections.
  • The cross-section is one date across generations, not a cohort through time; §04 states the consequence.
  • All numerators are posted list asks, not transactions; thin cells (n < 3) are indicative.

Full construction: /documents/methodology · underlying cells: /explorer · the dated study these lenses generalize: /research/gpu-age-curve.