Chip Economics
Earning power per spec-sheet unit.
Analytical decomposition of published reference rates — not a reference rate.
01 Four Lenses on One Rate
what the chip can hold — the size of model and context it can host
how fast it moves data — the dimension inference workloads buy
how fast it does arithmetic — training speed
what it costs to keep lit — rent per watt against facility cost per watt
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.
| Generation | Age | $/GPU-hr | n | $/GB-hr | $/TB/s-hr | $/PFLOP-hr | $/kW-hr |
|---|---|---|---|---|---|---|---|
| B300 | ~0.8y | $7.85 | 5 | 0.0273 | 0.98 | ~3.5 | ~6.0 |
| B200 | ~1.4y | $6.92 | 8 | 0.0384 | 0.87 | 3.1 | 6.92 |
| H200 | ~2.1y | $4.18 | 14 | 0.0296 | 0.87 | 4.2 | 5.97 |
| H100 | ~3.7y | $3.22 | 20 | 0.0403 | 0.96 | 3.3 | 4.60 |
| A100 80GB | ~5.7y | $1.80 | 13 | 0.0225 | 0.90 | 5.8 | 4.50 |
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.
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.
table view — watt-rent panel
| Generation | TDP | $/kW-hr | Shell class |
|---|---|---|---|
| B200 | 1.0 kW | 6.92 | liquid |
| B300 | ~1.3 kW | ~6.0 | liquid |
| H200 | 0.7 kW | 5.97 | dense air / liquid |
| L4 (Mkt) | 72 W | 5.83 | any air / edge |
| L40S | 350 W | 4.66 | air |
| H100 | 0.7 kW | 4.60 | air |
| A100 80GB | 400 W | 4.50 | air |
| A40 (Mkt) | 300 W | ~1.3 | air |
| 5090 / 4090 / 3090 (Mkt) | 350–575 W | 1.0–1.2 | consumer |
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.
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.
| Lens | Denominator | A100 80GB | H100 SXM | H200 | B200 | B300 | Spec source |
|---|---|---|---|---|---|---|---|
| Capacity · $/GB-hrwhat the chip can hold — the size of model and context it can host | HBM capacity (GB) | 80 | 80 | 141 | 180 | 288 | vendor datasheet · spec_source pending registry ext. |
| Bandwidth · $/TB/s-hrhow fast it moves data — the dimension inference workloads buy | memory bandwidth (TB/s) | 2.0 | 3.35 | 4.8 | 8.0 | ~8.0 | vendor datasheet · spec_source pending registry ext. |
| Compute · $/PFLOP-hrhow fast it does arithmetic — training speed | dense BF16 (PFLOPS) | 0.31 | 0.99 | 0.99 | 2.25 | ~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) | 0.4 | 0.7 | 0.7 | 1.0 | ~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.