No positions in the markets we measure BY 07:30 ET
CCIR Compute Credit
Index Research
Research · 2026-07-08 · window Oct 2017 – Apr 2026

GPU Rental Rates by Generation: Nine Years of Posted Prices

Each generation's own posted rental rate through time, rebuilt from 1,869 dated captures of providers' pricing pages, October 2017 to April 2026. Rates generally fall as a generation ages. Twice, demand for compute outran the supply of chips and pushed rates back up, most recently from late 2025.

2021–23 firming late-2025 → $0.5 $1 $2 $4 $8 201820192020202120222023202420252026 V100 $0.82 A100 $2.70 H100 $3.64 fit -15.3%/yrfit -15.3%/yr
V100 A100 H100 dashed = month has n<3 providers (shown, never fitted) straight dashes = the fitted declines below (R² 0.5 or higher), over each fit window solid = 3-month rolling median · faint steps = the raw monthly repricing path
Basis Posted panel-median guaranteed on-demand rate, $/GPU-hr, monthly, log scale, Oct 2017 – Apr 2026 (the archive's edge — one uniform right edge for every generation; the live CCIR panel continues from there). Solid lines are a centered 3-month rolling median (display smoothing only — fits, labels, and the table below are computed on the raw monthly medians, drawn as the faint steps). Step-function repair: posted prices carry forward between captures. Shaded bands are the two demand episodes described in §02.

01 The Paths

Fitted annual rates of change on the panel median, months with n≥3 only (the endpoint column is a compound annual growth rate, CAGR). A fitted rate is shown when the fit's R² is 0.5 or higher. Below that, the row reads "no clear trend":

GenerationWindow (n≥3)Log-linear fitLess CPIR²Endpoint CAGRMedian, first → last
V100Jun 2019 – Apr 2026 (6.8y) −15.3%/yr−19.2%/yr 0.82 −15.9%/yr$2.69 → $0.82
A100Sep 2021 – Apr 2026 (4.6y) no clear trendno clear trend 0.05 +4.6%/yr$2.20 → $2.70
H100Sep 2023 – Apr 2026 (2.6y) −15.3%/yr−18.4%/yr 0.57 +1.5%/yr$3.50 → $3.64

V100 and H100 declined about 15% a year. A100 shows no clear trend. Twice in nine years, demand for compute outran the supply of chips, and posted rates rose on generations that were already several years old (§02). V100, the one generation observable from mid-life to end-of-life, declined as fast as H100 and most regularly. A100's posted median was $2.20 in September 2021, peaked at $3.05 in 2022, fell to $1.70 in late 2024, and ended at $2.70 in April 2026, +4.6% per year endpoint to endpoint. H100's posted median also ends the window higher than it began it.

All prices are nominal, as posted. The "less CPI" column subtracts each window's consumer price index (CPI) inflation rate, the CPI for urban consumers (CPI-U) series (3–4%/yr over these years), from the fit. This matters most for A100: a path that rose 4.6% per year in posted dollars is roughly flat in real terms, endpoint to endpoint.

02 Two Demand Episodes

The decline has been interrupted twice, visibly, in nine years.

2021–2023. Posted rates on then-current and prior generations firmed through the GPU shortage and the first phase of the AI buildout. V100, four to six years old, rose 48% at DataCrunch and was repriced upward at OVHcloud. A100 rates at the repricers rose through 2023 into early 2024, and Lambda raised its A100 list prices outright in January 2024. The episode ended in mid-2024: A100 and V100 posted rates stepped down 30–56% from their peaks over the following year.

Late 2025 – the window's edge. The declines bottomed between late 2024 and August 2025. From roughly December 2025, posted rates turned up across every generation still listed, at every provider that reprices, driven by demand from the AI buildout: H100 +29% at Verda by December, H200 +70% by March. In spring 2026 Lambda raised A100 by 56% in seven weeks and repriced V100, silicon approaching its ninth year, from a list unchanged since June 2022. The sticky tier, as in the first episode, has not moved. This is the sharpest broad repricing of prior-generation silicon in the record.

03 Cohort and Cross-Section

The ≈−21%/yr figure in the age-curve note and the slower rates in §01 measure different things, and the difference is in the data rather than between two opinions. The cross-section compares differently-aged chips at one date: how the market prices age, today. A cohort path is one chip through time, and includes everything that happened to the whole market level along the way, including the two demand episodes above. A generation that ages through a firming market declines more slowly than the cross-sectional age gradient. V100, most of whose life predates the boom, and H100 come closest to it. Both series are published. Which one a reader needs depends on the question being asked of it.

04 Limitations

Posted asks, not transactions. Panel composition changes as providers enter, exit, redesign pages, or stop rendering prices server-side. The carry-forward repair and per-point counts are disclosed, and thin months are excluded from fitted figures. A100 pools memory variants. Archive coverage is not uniform: providers are captured at different intervals, and each posted price carries forward to the next capture. Step dates are bounded by capture dates, not observed exactly: a raise "in spring 2026" occurred between two dated captures. Methodology per the published documents. The underlying observation file, one archive URL per row, is available on request via /contact.

Methodology & data — sources, filters, repair, basis

Sources. Archived captures of provider pricing pages from the Internet Archive's Wayback Machine and from Common Crawl web-archive (WARC) records. Providers: Amazon Web Services (AWS), CoreWeave, DataCrunch (Verda from December 2025), Genesis Cloud, Lambda, OCI, OVHcloud, Paperspace, RunPod, with Common Crawl recoveries of Crusoe, Hyperstack, Together, and TensorDock. Every archived observation is a price the provider itself posted, on a dated page, with the capture URL recorded. From 2023, OCI's prices also come from the price-list data file its pricing page loads. From mid-2025, RunPod's come from the price fields in its archived pricing pages. The reconstruction ends at April 2026: one uniform right edge for every generation, all from the same measurement system. CCIR's live panel, which tracks the same providers daily, is the continuation from there. It is deliberately not spliced into this dated note.

Filters. Guaranteed on-demand rates only: reserved, interruptible (hyperscaler "Spot" products and community tiers alike), and "starting-at" teaser rows are excluded. USD only. Per-instance prices are divided to per-GPU with the derivation recorded per row. Known page bugs are excluded.

Repair. Posted prices are step functions: a page that read $1.79 in January and $1.79 in June did not move in between. A provider's posted price carries forward between captures until its next capture. A provider leaves the panel after its last capture. Change dates are bounded by capture dates. The monthly panel value is the median across providers present that month, with the count disclosed per point. Months with fewer than three providers are drawn dashed and excluded from every fitted figure. Trailing thin months at the end of a series are truncated rather than drawn.

Basis notes. CoreWeave's componentized per-GPU rate card (constant from 2021) was withdrawn in November 2025. Its instance-derived prices are a different billing basis and are treated as a separate series. One operator is one series: DataCrunch and Verda, and Paperspace and its parent DigitalOcean. A100 pools 40GB and 80GB memory variants at chip grain, matching the published series convention.