Research
Explore market-implied estimates, dated rental history and the infrastructure behind AI compute. Every chart keeps its own source, observation period and downloadable table.
Thin markets this session — no reliable move. No chip has a quality-passed anchor at both ends of its comparison window. No curve shape is read this session: each curve has a thin or interpolated point among its anchor, Oct ’26 and Aug ’27 ladders. 2 of 60 points on the board pass the ladder-quality gate.
Market snapshot · Inspect the estimates and quality limits
Data corrections ·
The linked record explains what was corrected, where it appeared and the supporting evidence.
Dated comparison and sources →Market structure ·
The comment period on the CFTC’s request for comment closes. The CFTC’s NYMEX review extension and ICE’s OCPI listing notice both point to this request. Status after: comment closed. Source: CFTC letter to CME Group (comment period closing Oct 20, 2026).
Dated comparison and sources →Market estimates ·
$5.49 (2026-10-13) → $5.81 (2026-10-14) on the same monthly-average instrument (KXH200MS · KXH200MS-26OCT) under one methodology (threshold ±5%).
Dated comparison and sources →These are the latest recorded changes by category, with separate observation periods. They are not live daily returns or a forecast of provider availability.
01Market-implied prices
Instrument-specific near anchors and monthly-average tenors through Aug ’27. Quality-passed points form the curve; thin observations retain their ranges and gaps.
Market-implied estimates, not rental offers. A thin market has insufficient quoted depth for a reliable trend. Thin points show ranges and gaps; moves require quality-passed endpoints.
Each estimate settles on Ornn’s executed-trade index — not comparable with posted list prices. Data: Kalshi (KalshiEX LLC)
● Market-implied term structure
Kalshi-implied: the first point uses a weekly exact-time or labeled front-month monthly-average anchor; remaining points are monthly-average contracts — shown in $/GPU·hr or % vs anchor
Full forward-curve snapshot: instrument-specific anchors, monthly tenors, methodology identity, and per-point ladder evidence.
curl -s https://aicomputetracker.com/forward-curves.jsonSpreadsheet twin of /forward-curves.json: one row per tenor with unit-encoded per-chip columns, quality, and ladder evidence.
curl -s https://aicomputetracker.com/curves.csvNo API key, CORS open (Access-Control-Allow-Origin: *) — fetch it from a browser, a notebook, or a cron job.
The site’s compilation is licensed CC BY 4.0 — attribution for this chart’s data “AI Compute Tracker (aicomputetracker.com), CC BY 4.0; third-party terms apply to data from Kalshi (KalshiEX LLC)”. Third-party source data — such as Kalshi market data and providers’ list prices — remains subject to its sources’ terms.
Market snapshot
Copy-ready citation
Cite as: AI Compute Tracker, “Implied forward curve”, market snapshot as of 2026-10-25, https://aicomputetracker.com/research/#curves, Data: Kalshi (KalshiEX LLC), AI Compute Tracker (aicomputetracker.com), CC BY 4.0; third-party terms apply to data from Kalshi (KalshiEX LLC), accessed <date you visit>
Replace “<date you visit>” with your access date.
Copy-paste iframe
The frame keeps this chart’s own freshness stamp, any third-party source credit, and the attribution line; the source line under it links back here with the data’s licence terms. It updates when the site redeploys.
Market depth
Preserved event ladders for 5 visible chips
Tap point to pin← → tenor · ↑ ↓ chip · Home / End · Enter pin · Esc close
Data: Kalshi (KalshiEX LLC) — event ladders referenced to Ornn. Implied median = the 50% crossing of an isotonic fit of usable quotes (bid/ask midpoints no wider than 20 points and trades with 24-hour volume, bracketed by the resting bids and asks), or where too few rest a closed bracket’s midpoint (an open bracket gives no level) — 3 of 55 ladders read from quotes this snapshot; nothing is smoothed: each monthly value is its own ladder’s estimate. The lines and bid/ask bands join only quality-passed points; gaps are hollow at their level, unplotted without one.
Latest read · market snapshot OCT 25 2026 · 06:37 UTC
Thin markets this session — no reliable move. No chip has a quality-passed anchor at both ends of its comparison window. No curve shape is read this session: each curve has a thin or interpolated point among its anchor, Oct ’26 and Aug ’27 ladders. 2 of 60 points on the board pass the ladder-quality gate.
The typical price (median of 4 specialist clouds) moved +1.8% to $3.99/GPU·hr between the Jul 23 and Oct 20, 2026 quote snapshots, inside the ±5% stability band. The cheapest on-demand price is $3.20/GPU·hr (Hyperstack, last observed Oct 14, 2026). Reviewed quote snapshot 2026-10-20, separate from the current offer comparison.
The archived GPT-4-class floor through 2026-04 was about 210× cheaper than launch at $0.18 per million tokens. Current list prices are reviewed separately. API review 2026-10-24; archived price history through 2026-04.
The September 28 CME notice postponed Silicon Data H100/B200 rental-index futures pending regulatory review. The original October 4 launch (October 5 trade date) is superseded; no replacement date is announced in that notice. Venue monitor
02GPU rentals
US$ per GPU-hour. The big-cloud list price peaked at $12 in 2023; the typical price is now a stable ~$4; the cheapest on-demand price is $3.20 (Hyperstack, last observed Oct 14, 2026).
Quarterly big-cloud list price; this quarter by venue class
The complete GPU price-history ledger: every series and dated observation, with billing, comparison, quality, and raw-evidence fields.
curl -s https://aicomputetracker.com/price-history.jsonPer-accelerator shard of the price-history ledger — the same series and point shapes filtered to one chip.
curl -s https://aicomputetracker.com/price-history/h100.jsonNo API key, CORS open (Access-Control-Allow-Origin: *) — fetch it from a browser, a notebook, or a cron job.
The site’s compilation is licensed CC BY 4.0 — attribution for this chart’s data “AI Compute Tracker (aicomputetracker.com), CC BY 4.0; third-party terms apply to data from Amazon Web Services, Microsoft Azure, Google Cloud, Oracle Cloud Infrastructure, Vultr, GPU cloud pricing pages”. Third-party source data — such as Kalshi market data and providers’ list prices — remains subject to its sources’ terms.
Live quotes
Copy-ready citation
Cite as: AI Compute Tracker, “H100 SXM · rental price by venue class”, live quotes as of 2026-10-25, https://aicomputetracker.com/research/#gpu-cloud, AI Compute Tracker (aicomputetracker.com), CC BY 4.0; third-party terms apply to data from Amazon Web Services, Microsoft Azure, Google Cloud, Oracle Cloud Infrastructure, Vultr, GPU cloud pricing pages, accessed <date you visit>
Replace “<date you visit>” with your access date.
Big-cloud list prices read from archived price lists (legacy quarterly values, indicative ±15%). This quarter's points are the 2026-10-20 quote table's aggregates, provisional unless an evidence record pins an immutable archived payload.
03The macro picture
The frontier training-compute trend runs at 4.4× a year. Companywide capex guidance uses different lease, fiscal-year and cash definitions; the total is indicative.
Total FLOP, log scale — estimates marked
Epoch AI estimates and lab disclosures; estimated points carry wide error bars.
US$B per calendar year (2026 = guidance)
Company disclosures. 2026 guidance basis: net cash for Amazon and Oracle; incl. finance leases for Microsoft and Meta. Basis and dated evidence on the Buildout page.