Model-level audit trail

H200 GPU cloud price history

Follow timestamped H200 cloud price observations and actual normalized-rate changes. The history keeps repeated market evidence separate from price moves so a refreshed source is not presented as a new price.

  • 362 timestamped observations in this model slice
  • 14 tracked offer identities across 7 providers
  • Historical price ranges are calculated from retained observations, not reconstructed marketing claims
Historical observations362
Actual rate moves0
Rate decreases0
Rate increases0

Historical snapshots

H200 observed market range by date

Each row summarizes the retained H200 observations for that calendar date. The low, median, and high are descriptive market observations across the tracked offers present in the ledger; they are not forecasts or purchasing recommendations.

Observed range chart

H200 low, median, and high price observations

MedianObserved range
H200 GPU cloud observed price range over timeStatic chart of low, median, and high USD per GPU-hour observations for H200 across 11 retained dates from 2026-08-02 to 2026-08-13.$4.66$4.07$3.47$2.88$2.292026-08-02: median $4.4300, range $4.3900–$4.4700 / GPU hr2026-08-04: median $4.3900, range $4.0000–$4.4700 / GPU hr2026-08-05: median $4.1450, range $2.4500–$4.5000 / GPU hr2026-08-06: median $4.1450, range $2.4500–$4.5000 / GPU hr2026-08-07: median $4.1450, range $2.4500–$4.5000 / GPU hr2026-08-08: median $4.1450, range $2.4500–$4.5000 / GPU hr2026-08-09: median $4.1450, range $2.4500–$4.5000 / GPU hr2026-08-10: median $4.1450, range $2.4500–$4.5000 / GPU hr2026-08-11: median $4.1450, range $2.4500–$4.5000 / GPU hr2026-08-12: median $4.1450, range $2.4500–$4.5000 / GPU hr2026-08-13: median $4.0000, range $2.4500–$4.5000 / GPU hr2026-08-022026-08-082026-08-13

Observed median movement: -9.7%. $4.4300 → $4.0000 / GPU hr from 2026-08-02 to 2026-08-13. At 730 GPU-hours, that observed median-rate change corresponds to $313.90 lower compute spend. Descriptive snapshot comparison; provider mix can differ, so this is not a matched-cohort index or a savings forecast.

Observed price spread widened by $1.9700/GPU hr. Range width $0.0800 → $2.0500 / GPU hr from 2026-08-02 to 2026-08-13. Relative to each snapshot median, spread width moved from 1.8% to 51.2%, a +49.4 percentage-point change. This compares retained snapshot dispersion; changing provider mix can also change the width.

The shaded band is the observed low-to-high range across retained H200 offers for each date; the solid line is the median. Provider mix can change between dates, so this is descriptive history rather than a matched-cohort index.
Observed dateLow / GPU hrMedian / GPU hrHigh / GPU hrTracked offersProviders
$2.4500$4.0000$4.5000136
$2.4500$4.1450$4.5000147
$2.4500$4.1450$4.5000147
$2.4500$4.1450$4.5000147
$2.4500$4.1450$4.5000147
$2.4500$4.1450$4.5000147
$2.4500$4.1450$4.5000147
$2.4500$4.1450$4.5000147
$2.4500$4.1450$4.5000147
$4.0000$4.3900$4.470033
$4.3900$4.4300$4.470022

Range movement

How the latest range compares with the first retained snapshot

This comparison describes the coverage currently retained in the ledger. A changing provider mix can move the observed range, so the range itself should not be interpreted as a matched-cohort market index. Use the GPU Cloud Price Index for the provider-balanced cohort view.

First retained snapshot

2026-08-02

$4.39–$4.47 / GPU hr, median $4.43.

2 tracked offers across 2 providers on that date.

Latest retained snapshot

2026-08-13

$2.45–$4.50 / GPU hr, median $4.00.

13 tracked offers across 6 providers on that date.

Evidence-only updates

164

Distinct source or evidence changes where the normalized H200 rate did not change. These remain separate from true price moves.

Verified price moves

Latest actual H200 rate changes

These cards include only transitions where the normalized USD-per-GPU-hour rate changed. Source refreshes at the same rate are excluded from this section. The 730 GPU-hour impact multiplies that exact rate delta by 730 and covers compute only.

No normalized H200 rate moves have been recorded yet. The observation history remains active.

Interpretation

Use this history for timing and auditability, not false certainty

Observed market slice

Snapshot ranges depend on the offers retained for H200 on that date. A wider or narrower provider set can move the range even when no matched offer changed price.

Availability is separate

A historical or current price does not prove capacity. Check the current H200 decision page and availability evidence before treating a low historical rate as deployable now.

No forecast

Historical movement is not a prediction of future GPU prices. It is an evidence trail for what the approved sources returned and when those observations changed.