Data downloads & API

Download a spreadsheet or use a public data endpoint: free, with no API key or sign-up. Each dataset keeps its units, source terms and original dates.

28 endpoints · Contract v1.27.0CC BY 4.0 for the site’s compilationsOpenAPI specificationStatus JSON

Download a dataset

CSV opens in a spreadsheet; JSON adds structured metadata. Each card shows when its own data was last observed or checked, from the file itself. Files are rebuilt with the site; sizes are uncompressed.

Current GPU rental prices

The provider quotes the chip pages compare, with billing terms, observation dates and sources.

Latest observation · 62 provider quotes

Preview the first rows of quotes.csv
gpuproviderbilling_modelusd_per_unit_hourobserved_at
B300ModalServerless / per-second7.09922026-10-20T00:00:00Z
B300HyperstackOn-demand7.42026-11-20T04:29:50.412Z
B300NebiusOn-demand9.52026-11-20T04:29:50.658Z

GPU rental price history

Dated observations for analysis. Compare like-for-like series and billing bases.

History through · 2,924 rows

Preview the first rows of price-history.csv
dategpuprovider_classbilling_modelusd_per_unit_hour
2022-03-31A100 80GBhyperscaleron-demand4.1
2022-09-30A100 80GBhyperscaleron-demand4.1
2023-03-31A100 80GBhyperscaleron-demand4.1

Model API prices

Recorded input and output token rates, including cache and batch rates where available.

Prices checked · 39 models

Preview the first rows of token-prices.csv
modelproviderinput_usd_per_mtokoutput_usd_per_mtok
GPT-5.4 ProOpenAI30180
GPT-5.5 ProOpenAI30180
Claude Fable 5Anthropic1050

Accelerator specifications

Memory, bandwidth and compute specifications with estimate flags and cited documents.

Specs checked · 29 accelerators

Preview the first rows of accelerators.csv
vendornamememorymemory_unitmemory_bandwidth_tb_per_s
AMDInstinct MI455X432GB23.3
NVIDIARubin GPU288GB19.2
MicrosoftMaia 200216GB7

Market-implied forward prices

Forward estimates from prediction-market ladders, with quality evidence. Not posted rental quotes.

Market snapshot · 5 chips

Preview the first rows of curves.csv
tenorb200_price_usd_per_gpu_hourb200_qualityb200_is_estimate
Anchor9.31thintrue
Nov '269.31thintrue
Dec9.63quality-passedfalse

Start with a task

Copy a browser-console example, or open the same data in the comparison tools. Keep the source terms and dates when reusing results.

Compare provider quotes

List H100 on-demand GPU-hour quotes from provider sources. Region, configuration and date stay visible; these rows are not interchangeable instances or a guarantee of availability.

Compare current offers →
JavaScript example
const data = await fetch("https://aicomputetracker.com/quotes.json").then(r => r.json());
const quotes = data.quotes.filter(q =>
  q.gpu === "H100" && q.billingModel === "On-demand" &&
  q.rateUnit === "GPU" && q.evidenceKind === "provider-primary" && !q.est
);
console.table(quotes.map(q => ({
  provider: q.provider, usdPerGpuHour: q.usdPerUnitHour,
  region: q.region, configuration: q.configuration,
  observedAt: q.observedAt, source: q.sourceUrl
})));
console.log(data.license);

Retrieve recorded history

Fetch one chip’s series and dated observations. Join by series ID and keep each configuration separate; missing dates remain gaps.

Explore H100 history →
JavaScript example
const data = await fetch("https://aicomputetracker.com/price-history/h100.json").then(r => r.json());
const series = new Map(data.series.map(s => [s.id, s]));
console.table(data.points.map(p => ({
  seriesId: p.seriesId, ...series.get(p.seriesId),
  date: p.date, price: p.usdPerUnitHour,
  quality: p.qualityStatus, observedAt: p.observedAt
})));
console.log(data.license);

Embed a sourced chart

Paste a responsive H100 history chart into your page. Keep the visible source link, license wording and the chart’s observation dates.

Choose a chart →
HTML embed example
<div style="position:relative;width:100%;aspect-ratio:4/3;min-height:420px;">
  <iframe src="https://aicomputetracker.com/embed/price-history/h100/" title="H100 · comparable price history — AI Compute Tracker" loading="lazy" style="position:absolute;inset:0;width:100%;height:100%;border:0;"></iframe>
</div>
<p style="margin:6px 0 0;font-size:12px;">Source: <a href="https://aicomputetracker.com/gpu-rental-prices/h100/#price-history-H100">AI Compute Tracker</a> · compilation 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</p>

Browse endpoints

Find a dataset, open its file or copy its URL. A JSON payload and its CSV twin share a card; expand it for code examples, source notes and schemas.

GPU rental prices

Current GPU quotes

Inspect the current provider quote table with billing terms, observation dates and evidence.

/quotes.json/quotes.csvLive quotes ·

Schema, examples and source notes
JSON · /quotes.json

Full URL: https://aicomputetracker.com/quotes.json

The current GPU rental quote table the chip pages compare: one row per provider and chip with its billing model, price, observation time, evidence fields and source, plus each citable class aggregate's latest dated point from the price-history ledger.

curl -s https://aicomputetracker.com/quotes.json
Fields (97) of the /quotes.json payload

The 97 fields are also listed in this JSON file and in openapi.json.

  • Values are posted list or ask prices, never transactions; capacity blocks, serverless, marketplace and secondary rows stay separate from on-demand rows.
  • A provider whose page is collected daily shows its newest collection (collectionMethod automated); any other keeps its reviewed observation. These are the rows the chip pages compare.
  • Each row keeps its own observation time: liveQuotesAsOf stamps the latest automated collection and pricesAsOf the manual survey, and neither re-dates a retained row.
  • A class aggregate is its citable series' latest daily point from the ledger, dated by that point; the quote rows can be older (retained) or newer than it, so a rule over the rows can differ from it.
  • The full dated history of every row and aggregate is /price-history.json.
  • Third-party sources: Amazon Web Services, Microsoft Azure, Google Cloud, Oracle Cloud Infrastructure, Vultr, GPU cloud pricing pages, Beam. Each payload's license.sources lists the terms of the sources it contains.
CSV · /quotes.csv

Full URL: https://aicomputetracker.com/quotes.csv

Spreadsheet twin of /quotes.json's quote rows: one row per provider and chip with raw prices, billing model, evidence fields and source.

curl -s https://aicomputetracker.com/quotes.csv
  • UTF-8 BOM and formula-neutralized cells via the site-wide CSV policy.
  • The columns are the quote table's own export columns plus collection_method, source_ids (the row's registry source id) and attribution (the attribution line for that row's data), so the terms travel with any row copied out of the file, then spot_status (/quotes.json's spotStatus: why a spot_usd_per_unit_hour cell is blank). Columns are only ever appended.
  • The trailing spot_pairing, spot_region, spot_configuration, spot_observed_at and spot_sku columns retain spot identity independently; unverified legacy rates cannot establish a discount. The citable class aggregates are in /quotes.json.
  • Third-party sources: Amazon Web Services, Microsoft Azure, Google Cloud, Oracle Cloud Infrastructure, GPU cloud pricing pages, Beam. Each source's own terms govern its data (see /openapi.json for each source's terms).

Reference prices for citing sites

Use the site's labeled reference prices, provider coverage and observation dates in a citation.

/reference-prices.jsonLive quotes ·

Open JSON: Reference prices for citing sites
Schema, examples and source notes

Full URL: https://aicomputetracker.com/reference-prices.json

What each chip page and provider page headlines, in one file for sites that quote the tracker: each chip's on-demand range and cloud count, its reference price with the page's own label, coverage, observation dates and citation, its reviewed spec strip, and each provider page's cheapest on-demand price per chip.

curl -s https://aicomputetracker.com/reference-prices.json
Fields (60) of the /reference-prices.json payload

The 60 fields are also listed in this JSON file and in openapi.json.

  • Every figure is composed by the same functions as the page that headlines it, so a citing site shows the tracker's own label, coverage and dates.
  • Posted list and ask prices, not transactions; billing models other than on-demand are on the chip and provider pages.
  • Third-party sources: Amazon Web Services, Microsoft Azure, Google Cloud, Oracle Cloud Infrastructure, GPU cloud pricing pages, Beam, Chip vendor documentation. Each payload's license.sources lists the terms of the sources it contains.

GPU price history

Analyze the complete published GPU rental observation ledger, keeping each series and billing basis separate.

/price-history.json/price-history.csvLive quotes ·

Schema, examples and source notes
JSON · /price-history.json

Full URL: https://aicomputetracker.com/price-history.json

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.json
Fields (54) of the /price-history.json payload

The 54 fields are also listed in this JSON file and in openapi.json.

  • Incomparable series (units, billing models, methodologies) are never joined; missing capture dates are real gaps.
  • A retired provider-series ID resolves through providerSeriesReplacements to its configured children; retain their separate series identities. This changes no citable class-aggregate ID or historical aggregate price.
  • Third-party sources: Amazon Web Services, Microsoft Azure, Google Cloud, Oracle Cloud Infrastructure, Vultr, GPU cloud pricing pages, Beam. Each payload's license.sources lists the terms of the sources it contains.
CSV · /price-history.csv

Full URL: https://aicomputetracker.com/price-history.csv

Long-form spreadsheet twin of /price-history.json: one row per observation with its series context denormalized in.

curl -s https://aicomputetracker.com/price-history.csv
  • UTF-8 BOM and formula-neutralized cells via the site-wide CSV policy.
  • Columns are append-only: citable_id and citable_id_qualifier were appended so existing positions stay stable, then source_ids (the row's series sources, space-separated registry ids) and attribution (that series' attribution line), the same fields each series carries in /price-history.json, then venue_count (a class-aggregate point's venueCount; empty when not recorded).
  • Third-party sources: Amazon Web Services, Microsoft Azure, Google Cloud, Oracle Cloud Infrastructure, Vultr, GPU cloud pricing pages, Beam. Each source's own terms govern its data (see /openapi.json for each source's terms).

Per-chip price history

Fetch the published rental history for one tracked chip.

/price-history/{chip}.jsonLive quotes ·

Open JSON: Per-chip price history
Schema, examples and source notes

Full URL: https://aicomputetracker.com/price-history/b300.json

Per-accelerator shard of the price-history ledger — the same series and point shapes filtered to one chip.

curl -s https://aicomputetracker.com/price-history/b300.json

{chip} — Chip permalink slug (the same slug as /gpu-rental-prices/<chip>/). One of: b300, b200, gb200-nvl72, h200, h100, a100-80gb, mi355x, mi300x, l40s, rtx-5090, rtx-4090, v100, tpu-v6e, tpu-v5e, tpu-v5p, trainium2.

Fields (52) of the /price-history/{chip}.json payload

The 52 fields are also listed in this JSON file and in openapi.json.

  • Shard series do not carry citableId; resolve citable ids from /price-history.json.
  • providerSeriesReplacements is scoped to this chip and resolves retired provider IDs without joining their different configurations.
  • Third-party sources: Amazon Web Services, Microsoft Azure, Google Cloud, Oracle Cloud Infrastructure, Vultr, GPU cloud pricing pages, Beam. Each payload's license.sources lists the terms of the sources it contains.

Archived quote aggregates

Retrieve the quote aggregates recorded for a particular published date.

/archive/quotes/{date}.jsonLive quotes ·

Open JSON: Archived quote aggregates
Schema, examples and source notes

Full URL: https://aicomputetracker.com/archive/quotes/2026-07-23.json

Dated permalink for one day of quote-table class aggregates: primary on-demand hyperscaler floor and specialist median plus the separately labeled marketplace-access floor, per GPU, as archived, without the venues of any source withdrawn since.

curl -s https://aicomputetracker.com/archive/quotes/2026-07-23.json

{date} — Archived UTC capture date (YYYY-MM-DD); /archive/ lists every date. One of: 2026-07-23, 2026-10-11, 2026-10-13, 2026-10-14, 2026-10-19, 2026-10-20, 2026-10-25, 2026-10-26, 2026-11-02, 2026-11-09, 2026-11-16.

Fields (24) of the /archive/quotes/{date}.json payload

The 24 fields are also listed in this JSON file and in openapi.json.

  • Dated permalink: an archived day's aggregates never change, except that a withdrawn source's rows leave every day: each aggregate is then the one its recorded venue set gives without them. So the permalink revalidates (an hour's shared cache) instead of being cached as immutable. New days land as new dated URLs.
  • Vast.ai marketplace prices were withdrawn on 2026-10-20: Vast's terms bar their use in price comparisons. Each archived day is served without the Vast.ai rows (/corrections/#vast-marketplace-prices-removed).
  • CoreWeave prices were withdrawn on 2026-10-20: CoreWeave's terms bar displaying its site's content with services that are not its own. Each archived day is served without the CoreWeave rows (/corrections/#restricted-source-prices-removed).
  • Crusoe prices were withdrawn on 2026-10-20: Crusoe's terms bar copying, publishing or displaying its site's content without written permission. Each archived day is served without the Crusoe rows (/corrections/#restricted-source-prices-removed).
  • Salad prices were withdrawn on 2026-10-20: Salad's terms bar using or displaying any part of its site without written consent. Each archived day is served without the Salad rows (/corrections/#restricted-source-prices-removed).
  • RunPod prices were withdrawn on 2026-10-20: RunPod's terms bar compiling its data into a collection or database without written permission. Each archived day is served without the RunPod rows (/corrections/#restricted-source-prices-removed).
  • Voltage Park prices were withdrawn on 2026-10-20: Voltage Park's terms bar collecting data from its platform without written permission. Each archived day is served without the Voltage Park rows (/corrections/#restricted-source-prices-removed).
  • GetDeploying estimates were withdrawn on 2026-10-20: the quote tables list providers' own prices only. Each archived day is served without the GetDeploying rows (/corrections/#restricted-source-prices-removed).
  • AIMultiple's market-average estimate was withdrawn on 2026-10-20: the quote tables list providers' own prices only. Each archived day is served without the AIMultiple rows (/corrections/#restricted-source-prices-removed).
  • A null is a real gap — aggregates are never carried forward between observation dates.
  • Third-party sources: Amazon Web Services, Microsoft Azure, Google Cloud, Oracle Cloud Infrastructure, Vultr, GPU cloud pricing pages. Each payload's license.sources lists the terms of the sources it contains.

Model API prices

Current token prices

Compare recorded model input, output, cache and batch prices with their billing conditions.

/token-prices.json/token-prices.csvCore data ·

Schema, examples and source notes
JSON · /token-prices.json

Full URL: https://aicomputetracker.com/token-prices.json

The current LLM API price table in USD per million tokens: input, output, cache-hit and batch prices where recorded, the long-context tier, and the blended 3:1 price per model, with each row's review date, terms and source — plus the dated GPT-4-class price-floor history.

curl -s https://aicomputetracker.com/token-prices.json
Fields (73) of the /token-prices.json payload

The 73 fields are also listed in this JSON file and in openapi.json.

  • Prices are standard (non-batch) list prices at the short-context or peak tier where a provider tiers them; longContext gives a recorded long-context tier, and each row's notes give the other terms and any promotional end date.
  • Batch and cache-hit prices are per model and recorded only where a review found them; a null is not recorded, never zero and never an assumed discount.
  • The blended price is a single-number price comparison, not a quality ranking.
  • gpt4ClassFloor is archived history with its own dates (observedThrough), separate from the current table's review.
  • Third-party sources: Model API pricing pages. Each payload's license.sources lists the terms of the sources it contains.
CSV · /token-prices.csv

Full URL: https://aicomputetracker.com/token-prices.csv

Spreadsheet twin of /token-prices.json's model rows: one row per model with raw prices in USD per million tokens (batch and long-context columns empty where not recorded), review date, terms and source.

curl -s https://aicomputetracker.com/token-prices.csv
  • UTF-8 BOM and formula-neutralized cells via the site-wide CSV policy.
  • The columns are the AI Model API Prices table's own export columns plus source_ids (the row's registry source id) and attribution (the attribution line for that row's data). Columns are only ever appended.
  • The GPT-4-class price-floor history is in /token-prices.json.
  • Third-party sources: Model API pricing pages. Each source's own terms govern its data (see /openapi.json for each source's terms).

Measured serving reference workloads

Inspect versioned vendor-measured workloads before using throughput in an inference-cost scenario.

/serving-benchmarks.jsonCore data ·

Open JSON: Measured serving reference workloads
Schema, examples and source notes

Full URL: https://aicomputetracker.com/serving-benchmarks.json

Historical vendor-reported synthetic throughput with exact model, system, precision and request lengths; unknown latency and measurement dates stay null.

curl -s https://aicomputetracker.com/serving-benchmarks.json
Fields (39) of the /serving-benchmarks.json payload

The 39 fields are also listed in this JSON file and in openapi.json.

  • Historical vendor measurements are not current software or interactive-service rankings. Source version and source update date travel with every workload; third-party terms apply.
  • Third-party sources: Vendor-published inference measurements. Each payload's license.sources lists the terms of the sources it contains.

Hardware

Accelerator specs

Compare accelerator specifications, individual estimate flags and cited vendor documents.

/accelerators.json/accelerators.csvCore data ·

Schema, examples and source notes
JSON · /accelerators.json

Full URL: https://aicomputetracker.com/accelerators.json

The accelerator spec table: dense tensor TFLOPS by precision, memory, bandwidth, power and interconnect per chip, with est flags and, for each reviewed row, its review date, reviewed fields and the vendor documents it cites.

curl -s https://aicomputetracker.com/accelerators.json
Fields (46) of the /accelerators.json payload

The 46 fields are also listed in this JSON file and in openapi.json.

  • FLOPS are dense tensor throughput with FP32 accumulate — never vendor sparsity figures — and FP16/BF16, FP8 and FP4 are never mixed in one comparison.
  • Only a row's specReview.fields were re-read on its review date; launch months and other fields keep earlier dates.
  • Third-party sources: Chip vendor documentation. Each payload's license.sources lists the terms of the sources it contains.
CSV · /accelerators.csv

Full URL: https://aicomputetracker.com/accelerators.csv

Spreadsheet twin of /accelerators.json: one row per chip with raw spec figures, est flag, review date, reviewed fields and cited document URLs.

curl -s https://aicomputetracker.com/accelerators.csv
  • UTF-8 BOM and formula-neutralized cells via the site-wide CSV policy.
  • The columns are the /hardware table's own export columns plus source_ids (the row's registry source ids, space-separated) and attribution (the attribution line for that row's data). Columns are only ever appended.
  • Third-party sources: Chip vendor documentation. Each source's own terms govern its data (see /openapi.json for each source's terms).

Shared compute identities

Resolve chip families and exact systems to reviewed Tracker destinations and Wiki audiences.

/compute-entities.jsonCore data ·

Open JSON: Shared compute identities
Schema, examples and source notes

Full URL: https://aicomputetracker.com/compute-entities.json

Versioned subject and system identities connecting Tracker destinations to reviewed Wiki audiences, with reference memory kept distinct from provider offers.

curl -s https://aicomputetracker.com/compute-entities.json
Fields (27) of the /compute-entities.json payload

The 27 fields are also listed in this JSON file and in openapi.json.

  • Provider-advertised memory is retained separately; an ambiguous capacity never silently replaces a reference specification.
  • Third-party sources: Chip vendor documentation. Each payload's license.sources lists the terms of the sources it contains.

Infrastructure

Frontier buildout ledger

Explore disclosed compute campuses and their separately labeled operating and planned capacity.

/buildout.jsonCore data ·

Open JSON: Frontier buildout ledger
Schema, examples and source notes

Full URL: https://aicomputetracker.com/buildout.json

Provenanced ledger of frontier-scale AI datacenter sites: operator, location, accelerators, current and planned capacity, status, key dates — every figure traceable to a dated public source via per-site claims.

curl -s https://aicomputetracker.com/buildout.json
Fields (46) of the /buildout.json payload

The 46 fields are also listed in this JSON file and in openapi.json.

  • Best public estimates, not an audited census: capacity figures mix IT-load and campus-power conventions per underlying source — each claim names its basis.
  • A null capacity is a real gap (no public figure), never zero.
  • Third-party sources: Epoch AI, Public disclosures and reporting. Each payload's license.sources lists the terms of the sources it contains.

Compute deals ledger

Inspect announced compute contracts with each value's basis, status and dated public sources.

/deals.jsonCore data ·

Open JSON: Compute deals ledger
Schema, examples and source notes

Full URL: https://aicomputetracker.com/deals.json

Contracts for AI computing capacity since 2023 — cloud capacity, chip supply, data-center leases and builds, the financing tied to them, and power — each with the basis of its stated value, its capacity and status, and its dated public sources.

curl -s https://aicomputetracker.com/deals.json
Fields (85) of the /deals.json payload

The 85 fields are also listed in this JSON file and in openapi.json.

  • A ledger of public disclosures, not an audited census: deals whose terms were never disclosed are absent, and a reported deal is marked as such.
  • Stated values are not comparable across bases — rank by value only within one basis.
  • Third-party sources: Public disclosures and reporting. Each payload's license.sources lists the terms of the sources it contains.

Buildout financials

Analyze reported capex and NVIDIA data-center revenue with guidance and estimate flags.

/financials.jsonCore data ·

Open JSON: Buildout financials
Schema, examples and source notes

Full URL: https://aicomputetracker.com/financials.json

The buildout's financial tables: big-tech capital expenditure by calendar year with each company's basis note and guidance source, and NVIDIA data-center revenue by fiscal quarter with the release behind it.

curl -s https://aicomputetracker.com/financials.json
Fields (82) of the /financials.json payload

The 82 fields are also listed in this JSON file and in openapi.json.

  • Companywide capital expenditure includes spending beyond AI, and company definitions differ — read each basisNote before comparing companies.
  • A guidance value is the company's guidance (the midpoint where a range was given), not an actual; secondarySource marks a value taken from a secondary report, and both carry est: true.
  • Only each company's latest guidance, NVIDIA's latest release and the quarterly disclosures carry a dated source link in this payload; the other figures carry no link of their own.
  • Third-party sources: Public disclosures and reporting. Each payload's license.sources lists the terms of the sources it contains.

Forward markets

Forward curves

Inspect market-implied forward estimates with bid/ask brackets, identification verdicts and evidence.

/forward-curves.json/curves.csvMarket snapshot ·

Schema, examples and source notes
JSON · /forward-curves.json

Full URL: https://aicomputetracker.com/forward-curves.json

Full forward-curve snapshot: instrument-specific anchors, monthly tenors, methodology identity, and per-point ladder evidence.

curl -s https://aicomputetracker.com/forward-curves.json
Fields (124) of the /forward-curves.json payload

The 124 fields are also listed in this JSON file and in openapi.json.

  • Values are reconstructed market-implied levels, never current spot or executable quotes.
  • Check per-point quality: thin or interpolated points are gaps on the site's chart and carry estimated: true here.
  • Fields marked v2 (contract 1.13.0) come from estimator v2: a bid/ask bracket per point, a PAVA fit of tight quotes and fresh trades, no smoothing. v1 read every strike's midpoint or last trade, repaired with a running minimum and smoothed. The payload note states the snapshot's own estimator and counts.
  • A v2 value whose valueSource is interpolated, or bracket with an open side, is a placeholder that keeps curve rectangular, not a median: the site prints its bracket instead and never plots it.
  • A chip whose series falls below the series coverage minimum is still published, with every point marked estimated and the failed checks listed in provenance.coverageShortfalls; a refresh in which most chips fall below it is not published.
  • Curve values are derived from Kalshi market data — show “Data: Kalshi (KalshiEX LLC)” with them; that data remains subject to Kalshi’s terms.
  • Third-party sources: Kalshi (KalshiEX LLC). Each payload's license.sources lists the terms of the sources it contains.
CSV · /curves.csv

Full URL: https://aicomputetracker.com/curves.csv

Spreadsheet 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.csv
  • UTF-8 BOM and formula-neutralized cells via the site-wide CSV policy.
  • The first row is a `# as_of` stamp; the header row follows it.
  • `<chip>_market_updated_at` is the ladder's latestMarketAt: when Kalshi last edited one of its market records, not a trade or quote time.
  • The attribution column repeats on every row the attribution line /forward-curves.json carries in license.attribution, so the terms travel with any row copied out of the file.
  • Columns are only appended: from 1.13.0 seven bracket, quartile and `_identified` columns per chip follow attribution, empty where a point has none. A price with an open bracket side or no ladder is a placeholder.
  • Curve values are derived from Kalshi market data — show “Data: Kalshi (KalshiEX LLC)” with them; that data remains subject to Kalshi’s terms.
  • Third-party sources: Kalshi (KalshiEX LLC). Each source's own terms govern its data (see /openapi.json for each source's terms).

Archived forward-curve session

Retrieve one dated forward-curve session with its own estimator and pricing disclosure.

/archive/curves/{date}.jsonMarket snapshot ·

Open JSON: Archived forward-curve session
Schema, examples and source notes

Full URL: https://aicomputetracker.com/archive/curves/2026-07-02.json

Dated permalink for one archived forward-curve session: monthly values keyed to the archive's union tenor window, anchor identity, spots, and series-level ingestion evidence, exactly as archived.

curl -s https://aicomputetracker.com/archive/curves/2026-07-02.json

{date} — Archived UTC session date (YYYY-MM-DD); /archive/ lists every date. One of: 2026-11-11, 2026-11-12, 2026-11-13, 2026-11-14, 2026-11-15, 2026-11-16, 2026-11-17, 2026-11-18, 2026-11-19, 2026-11-20, 2026-11-21, 2026-11-22 and 57 earlier ones; the archive lists every one.

Fields (94) of the /archive/curves/{date}.json payload

The 94 fields are also listed in this JSON file and in openapi.json.

  • New sessions land as new dated URLs, and each refresh replaces the current UTC day's session until the day closes.
  • A closed session changes only when an estimator version bump re-derives it from its retained raw ladders: `rederived` records the run and keeps the values first published in `rederived.superseded`, so a cited value still resolves here.
  • Values are reconstructed market-implied levels under the stated methodology, never current spot or executable quotes.
  • Third-party sources: Kalshi (KalshiEX LLC). Each payload's license.sources lists the terms of the sources it contains.

Evidence

Evidence digest manifest

Verify published evidence digests by UTC day and inspect timestamp proofs where available.

/evidence/digests.jsonLive quotes ·

Open JSON: Evidence digest manifest
Schema, examples and source notes

Full URL: https://aicomputetracker.com/evidence/digests.json

Every SHA-256 evidence digest the published rows cite, by UTC day, with its reference, the series it backs and a Merkle root per day: digests only, never a payload. /evidence/digests/ shows how to recompute both.

curl -s https://aicomputetracker.com/evidence/digests.json
Fields (35) of the /evidence/digests.json payload

The 35 fields are also listed in this JSON file and in openapi.json.

  • The site's own computation over its published rows: it holds no third-party data.
  • Contains no third-party source data.

Raw-evidence references (CSV)

Audit the published ledger's raw-evidence references and payload digests; raw private payloads are not included.

/evidence/refs.csvLive quotes ·

Download CSV: Raw-evidence references (CSV)
Schema, examples and source notes

Full URL: https://aicomputetracker.com/evidence/refs.csv

Every distinct raw-evidence reference the price-history ledger carries, newest coverage first: the reference, the payload digests it pins, how many observations cite it and on which dates, and the sources behind them. /evidence shows the newest few.

curl -s https://aicomputetracker.com/evidence/refs.csv
Columns (11) of the /evidence/refs.csv file

The 11 columns are also listed in this JSON file and in openapi.json.

  • UTF-8 BOM and formula-neutralized cells via the site-wide CSV policy.
  • References identify payloads in the site's private raw archive: they are identifiers, not links. Payloads are available on request; /evidence describes how a payload is checked against its digest.
  • Rebuilt from the committed price-history ledger on every build; a reference never changes once an observation carries it.
  • Third-party sources: Amazon Web Services, Microsoft Azure, Google Cloud, Oracle Cloud Infrastructure, Vultr, GPU cloud pricing pages. Each source's own terms govern its data (see /openapi.json for each source's terms).

Reviewed research evidence

Review primary-source changes with stable subjects and separate publication, observation, effective and review dates.

/research-evidence.jsonCore data ·

Open JSON: Reviewed research evidence
Schema, examples and source notes

Full URL: https://aicomputetracker.com/research-evidence.json

Primary-source review events with stable entity identities, exact Wiki audiences and separate source, observation, effective and review dates.

curl -s https://aicomputetracker.com/research-evidence.json
Fields (27) of the /research-evidence.json payload

The 27 fields are also listed in this JSON file and in openapi.json.

  • Static reviewed events. Publication, observation, effective and review dates retain different meanings; unknown dates are absent.
  • Third-party sources: Public disclosures and reporting. Each payload's license.sources lists the terms of the sources it contains.

Site and feeds

Freshness status

Check every freshness clock separately; build time is not an observation time.

/status.jsonMarket snapshot · Live quotes · Core data ·

Open JSON: Freshness status
Schema, examples and source notes

Full URL: https://aicomputetracker.com/status.json

The three freshness clocks side by side — market snapshot, live quotes and core data, each with its own stamp, cadence and build-time age — plus every core dataset's own review, the quote snapshot's stamps and each automated row's own observation time.

curl -s https://aicomputetracker.com/status.json
Fields (69) of the /status.json payload

The 69 fields are also listed in this JSON file and in openapi.json.

  • A static export: ageHours and isStale were computed once, at buildGeneratedAt. Re-derive live staleness from each clock's asOf plus staleAfterHours.
  • The clocks are independent and never merged into one “last updated”; the production check (check:live) reads this file.
  • Contains no third-party source data.

Market & quote events (RSS)

Subscribe to market, quote, correction and reviewed research events with RSS.

/feed.xmlMarket snapshot ·

Open XML: Market & quote events (RSS)
Examples and source notes

Full URL: https://aicomputetracker.com/feed.xml

RSS 2.0. Data-written market and quote events: the desk note for the current forward snapshot, the newest market and quote entries from the desk log (quote moves and venue changes between quote snapshots, marketplace-floor lows, forward anchor moves and market-structure milestones), the newest corrections to the site's own figures and claims, and dated research publications, each pointing to its full record. Research publications are separate from price moves. Archive housekeeping (curve-session roll-ups, tenor-window rolls and methodology changes) stays on the desk log.

curl -s https://aicomputetracker.com/feed.xml
  • The channel's copyright element carries the feed's attribution line, derived from the source-rights registry for the sources listed below.
  • Correction items carry only their headline, the surfaces affected and how the error was found, and link their desk-log entry, which links the full record on /corrections: no figure a correction quotes enters the feed.
  • The three newest reviewed research publications have a separate bounded window. They retain their own publication dates and never advance the market-observation clock or displace price events.
  • An item computed from a withdrawn source's prices keeps its guid, link and date, with figure-free text that says it was withdrawn and links the withdrawal's record.
  • Third-party sources: Kalshi (KalshiEX LLC), Amazon Web Services, Microsoft Azure, Google Cloud, Oracle Cloud Infrastructure, Vultr, GPU cloud pricing pages, Public disclosures and reporting. Each source's own terms govern its data (see /openapi.json for each source's terms).

Per-chip market & quote events (RSS)

Subscribe to published market and quote events for one chip with RSS.

/feed/{chip}.xmlLive quotes ·

Open XML: Per-chip market & quote events (RSS)
Examples and source notes

Full URL: https://aicomputetracker.com/feed/b300.xml

RSS 2.0, one per chip with market or quote events: the newest 20 of that chip's quote and anchor events and the milestones whose source names it, with the site feed's item markup and guids. Each chip page links its feed.

curl -s https://aicomputetracker.com/feed/b300.xml

{chip} — Chip slug, as in /gpu-rental-prices/<chip>/. One of: b300, b200, h200, h100, a100-80gb, mi300x, rtx-5090, rtx-4090.

  • Each feed's copyright line covers its own items' sources; the sources below cover every chip's feed.
  • An item computed from a withdrawn source's prices keeps its guid, link and date, with figure-free text that says it was withdrawn and links the withdrawal's record.
  • Third-party sources: Kalshi (KalshiEX LLC), Vultr, GPU cloud pricing pages. Each source's own terms govern its data (see /openapi.json for each source's terms).

llms.txt

Find the site's machine-readable data and documentation links.

/llms.txtCore data ·

Open TXT: llms.txt
Examples and source notes

Full URL: https://aicomputetracker.com/llms.txt

Curated plain-text site map with dataset conventions for agents and LLM crawlers (llmstxt.org).

curl -s https://aicomputetracker.com/llms.txt
  • Contains no third-party source data.

llms-full.txt

Read a plain-text site briefing with labeled prices, coverage and source terms.

/llms-full.txtCore data ·

Open TXT: llms-full.txt
Examples and source notes

Full URL: https://aicomputetracker.com/llms-full.txt

Full plain-text dataset for answer engines: current GPU quotes, the forward board, token prices, accelerator specs, citable class aggregates, and the buildout (capex, NVIDIA data-center revenue, the largest campuses) inlined as tables, each section stamped by its owning clock (market snapshot, live quotes, or core data).

curl -s https://aicomputetracker.com/llms-full.txt
  • Third-party sources: Kalshi (KalshiEX LLC), Amazon Web Services, Microsoft Azure, Google Cloud, Oracle Cloud Infrastructure, Vultr, GPU cloud pricing pages, Beam, Epoch AI, Public disclosures and reporting, Model API pricing pages, Chip vendor documentation. Each source's own terms govern its data (see /openapi.json for each source's terms).

Withdrawn

Kept at the same address so old links resolve; it carries no prices.

Marketplace ask tape (withdrawn)

Compatibility only: this withdrawn endpoint contains no price points.

/marketplace-tape.jsonLive quotes ·

Open JSON: Marketplace ask tape (withdrawn)
Schema, examples and source notes

Full URL: https://aicomputetracker.com/marketplace-tape.json

Withdrawn on 2026-10-20: the intraday ask tape carried only Vast.ai asks. It keeps its shape with no points (days is empty, collectedThrough null) and is removed in the next major contract version.

curl -s https://aicomputetracker.com/marketplace-tape.json
Fields (37) of the /marketplace-tape.json payload

The 37 fields are also listed in this JSON file and in openapi.json.

  • Vast.ai marketplace prices were withdrawn on 2026-10-20: Vast's terms bar their use in price comparisons.
  • The tape is served empty, in its 1.x shape, until the next major contract version removes it; /corrections/#vast-marketplace-prices-removed records the withdrawal.
  • Contains no third-party source data.

Embeds and badges

Embed a chart

35 embeddable charts

Each chart’s iframe twin keeps its freshness stamp and credits; its snippet adds a source line that links back to the chart and states the licence of the data it shows.

Badges

14 live badges

SVG price stamps for a README or a sidebar, rebuilt with the site, with Markdown and HTML snippets that link each badge to the page it stamps.

Reuse, citing & versioning

The site’s own text and data compilations are licensed CC BY 4.0. Third-party source data — such as Kalshi market data and providers’ list prices — remains subject to its sources’ terms.

Attribution: AI Compute Tracker (aicomputetracker.com), CC BY 4.0; includes data from Epoch AI (epoch.ai), CC BY 4.0.

Freshness clocks

The dates on each dataset above are that dataset’s own. The site also keeps three independent clocks; none of them is the age of every row, and a rebuild never makes an observation newer.

Market snapshot

Forward curves and the terminal

The forward-market pipeline (Kalshi ladder snapshot)

Live quotes

Automated quote rows and the daily provider capture, dated by whichever collected last

The weekly quote refresh plus the scheduled price-capture workflow

Core data

Hand-reviewed core datasets, dated per domain (the oldest domain's latest complete review)

A complete review of that domain; partial reviews move no date

Core data is the last complete survey; selected datasets have newer reviews. Daily history retains each observation’s date and real gaps. Status JSON reports all clocks and scoped reviews.

Access and code examples

Every endpoint is a static file with open CORS (Access-Control-Allow-Origin: *) and an hour-level cache policy. No authentication or SDK is required. Per-card examples use a concrete path, with available parameter values in the details.

The OpenAPI 3.1 specification and this catalog are generated from the same contract. 19 of 28 endpoints document their fields or columns individually.

curl -s https://aicomputetracker.com/openapi.json
License terms per file

Each JSON payload’s root license block carries the attribution for that payload and the terms of its third-party sources. CSV rows carry attribution where documented; do not assume one blanket license covers every source.

Citation and BibTeX

Cite the site directly; reuse of the site’s compilations carries the CC BY 4.0 attribution line “AI Compute Tracker (aicomputetracker.com), CC BY 4.0; includes data from Epoch AI (epoch.ai), CC BY 4.0”.

@misc{aicomputetracker2026,
  author = {{AI Compute Tracker}},
  title  = {{AI Compute Tracker — The price of compute, tracked}},
  year   = {2026},
  url    = {https://aicomputetracker.com},
  note   = {Core data as of 2026-07-23; CC BY 4.0},
}
Versioning and compatibility

DATA CONTRACT v1.27.0

The contract follows semantic versioning. Additive, backwards-compatible changes — a new field, a new endpoint — bump the minor version and never rename, remove, or re-type an existing field. Breaking changes — a renamed or removed field, changed units or semantics — bump the major version and are announced with a deprecation window during which the previous shape keeps being served. Documentation-only corrections bump the patch version.

Every JSON payload carries contractVersion and license at its root so a saved file stays self-describing. /price-history.json additionally keeps its own integer schemaVersion, which versions the ledger artifact itself and is never renamed.

A provider ID corrected into separate recorded configurations resolves through the optional providerSeriesReplacements map in the full ledger and chip shards. Read each child's series and observations separately. Citable class-aggregate IDs keep their identities.