Published 07:13 AEST · Before US Open · Wednesday, 9 September 2026Pipeline 04:30 AESTVol. I  No. 121
The Grid
Independent Price Reporting for the Compute Market
thegridco.ai
EST. 2026 · SYDNEY
Specification · Class 1 · assessment — a price for a defined thing

GAP-H100-R1Y · Grid Assessed Price — H100, 1-year reserved

The provider-pooled median of the freshest 1-year reserved / committed H100 SXM effective rate per provider (quote age ≤ 3 days), expressed per GPU-hour. Reinstatement requires a new methodology version with its own specification: a reserved panel that actually moves and a staleness rule that withholds rather than freezes (notice 2026-08-14 §7).
series id gap_h100_reserved_1y · methodology v1.0 · unit usd_per_gpu_hr

· Class and status

Part B class · print status · latest print of record
Class 1 · assessmentSuspended from 2026-08-22
Class — and why

Registered as an assessed reserved leg (decision of record 2026-07-24, v1.0). SUSPENDED by methodology change notice 2026-08-14 §1: all 20 prints since inception carry the identical value (population σ = 0). The class label is historical — from the effective date the series receives no new values.

Suspension. Effective 2026-08-22; final prints dated 2026-08-21per the notice. Published history is not removed, revised or retroactively annotated — each value stands as published on the information set of its day. From the effective date the series appears on /api/print with status “suspended” and no value. Notice of record: /standards/notices/2026-08-14-instrument-actions.

Latest print of record · read from storage, not recomputed
Series idRegistry labelMethodologyLatest printValueNote
gap_h100_reserved_1yGAP-H100-R1Y (1-yr reserved)v1.02026-08-21$6.4645/GPU-hrsuspended from 2026-08-22

· Specification

A4 field template · honest current state · n/a fields shown, not hidden
CodeGAP-H100-R1Yseries id gap_h100_reserved_1y
Accelerator classNVIDIA H100 SXM-class, 80GBAzure leg: ND96 H100 v5 (8× SXM); Lambda leg: 1-Click Cluster committed tier (H100 SXM)
Interconnect / topologyNot normalised8-GPU node (Azure) and cluster tiers (Lambda) enter at their per-GPU effective rate
Contract basisReserved / committed-term
Term1 year (standard term for this code)
CommitmentNot normalisedthe venue's committed-tier minimum applies; no Grid quantity or spend basis
RegionNot normalisedAzure leg priced in eastus; Lambda leg unspecified
InclusionsThe committed compute rate only
Currency / unitUS$ per GPU-hour, printed to four decimals
Assessment close18:30 UTC — the scheduled daily cycle anchorThe print date is the cycle date, derived from the scheduled anchor and never from wall-clock time (incident of record 2026-08-06). Observations are those captured in that cycle's collection run.
Submission cut-offNone — no submission channel is open. Participation v0 (submission definition, mailbox, receipt discipline) is at /standards/participate; no submission has been received.
Publication timeWith the daily print (masthead: 04:30 AEST) — on the briefing, /instruments, the daily report, /api/print/<date> (permanent) and MCP.
Data hierarchy appliedTier 2 / tier 3 — public retail reservation prices and published committed-cluster pricing. Transactions: 0. Judgment: none (arithmetic).The reserved panel is near-static by construction: this is the defect the suspension names.

· Contributing panel (v1.0)

inputs of record
ProviderSourceBasis
AzureRetail Prices API (public)cheapest ND96 H100 v5 (8× SXM) 1-Year reservation, eastus; effective $/GPU-hr = term price ÷ 8,760 h ÷ 8
Lambdapublished committed-cluster pricingmedian 1-Click Cluster committed tier (H100 SXM)

· Normalisation and construction rules

the named adjustments, in order
  • Effective per-GPU-hour rate — Azure term price ÷ 8,760 hours ÷ 8 GPUs; Lambda committed tier taken as published.
  • Freshest quote per provider within 3 days; provider-pooled median across providers.
  • Sanity bounds — US$0.20–30.00 per GPU-hour.

· Minimum observations and withholding

below which the print is flagged · below which it is not published

None was specified for this code — the omission is the defect: presence-tested freshness let a frozen panel print for 20 cycles. Any reinstated reserved leg carries a withholding rule (notice 2026-08-14 §7).

· Disclosures

verbatim from the derived-series registry · the same string every surface carries
GAP-H100-R1YDefect disclosed 2026-08-14: every print since inception (2026-07-24) is the identical value — zero variance. The reserved panel (near-static Azure retail plus a Lambda fallback) does not move, so the series has carried no market information. Suspended from 2026-08-22 (methodology change notice 2026-08-14); published history stands as-published.

· Revision history

dated changelog for this code · never rewritten
DateVersionChangeReference
2026-07-24v1.0Reserved 1-year leg introduced with GAP methodology v3 (decision of record 2026-07-24).docs/gap-methodology-v3.md
2026-08-14Defect disclosed (zero variance across all 20 prints since inception). Suspended effective 2026-08-22; final prints dated 2026-08-21; five-business-day notice period 2026-08-17 → 2026-08-21 with the deviation from the three-month cessation guidance stated (§4). Published history stands as published.docs/methodology-notices/2026-08-14-instrument-actions.md §1, §4

· Downstream, surfaces, notices, sources

where it flows · where it shows · what governs it
Downstream series
  • GTS-1Y — the reserved leg (suspended together).
  • RWI self-host (reserved) columns for RW-S and RW-L cease printing from the effective date (notice §6); RWI-S / RWI-L cheapest-path indices are unaffected.
Surfaces
  • /instruments §Ⅰ (chart stops advancing at the final print)
  • /api/print (status: suspended from the effective date)
  • /api/mcp get_term_structure
Notices that govern this code
Sources of record
scripts/compute_gap_print.py (reserved_1y_level, RESERVED_MAX_AGE_DAYS = 3)scripts/derived_series.py ↔ dashboard/lib/derived-series.ts (suspended gate)docs/gap-methodology-v3.md
This page renders the method of record for GAP-H100-R1Y; it does not define it. Printed values are never rewritten — a change in method is announced by notice before it affects a print (publish-before-effect); a correction restates only what the day’s own information set supported. Permanent print URLs: /api/print/<date>.