Market benchmarks / compute economics / market structure
The price of compute, tracked.
GPU rental markets, market-implied forward curves, chip economics, token prices, and the gigawatt buildout — one independent desk for the machines that make intelligence.
01Forward curves
What the market thinks compute will cost
Instrument-specific near anchors, monthly-average tenors through Aug ’27, and the shape of implied supply pressure ahead.
● Market-implied term structure
Implied forward curve
First point is a weekly exact-time anchor (RTX 5090: front-month monthly average); remaining points are monthly-average contracts — shown in $/GPU·hr or % vs anchor
Data — endpoints, license, and freshness behind Implied forward curve
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.
Licensed CC BY 4.0 — attribution “AI Compute Tracker (aicomputetracker.com), CC BY 4.0”.
Market snapshot
Cite — copy a citation for Implied forward curve
Copy-ready citation
Cite as: AI Compute Tracker, “Implied forward curve”, market snapshot as of 2026-08-28, https://aicomputetracker.com/#curves, AI Compute Tracker (aicomputetracker.com), CC BY 4.0, accessed <date you visit>
Replace “<date you visit>” with your access date.
Embed — copy an iframe snippet for Implied forward curve
Copy-paste iframe
<div style="position:relative;width:100%;aspect-ratio:16/10;">
<iframe src="https://aicomputetracker.com/embed/forward-curve/" title="Implied forward curve — AI Compute Tracker" loading="lazy" style="position:absolute;inset:0;width:100%;height:100%;border:0;"></iframe>
</div>The frame keeps this chart’s own freshness stamp and CC BY 4.0 attribution; it updates when the site redeploys.
Kalshi event ladders referenced to Ornn. Raw implied median = the 50% bid/ask-mid crossing; monthly back points use two [1,2,1] smoothing passes from the third month onward. Hollow points are thin or interpolated.
How to read this curve
The shape is the story. H100 trades backwardated; B200 and H200 sit in contango; RTX 5090 eases lower out the strip; A100 is flat, priced to hold through Aug ’27.
Replay the archived market-implied term structure.
Computed signals
The market in five numbers
Derived from the current datasets on every build — never hand-entered.
- B200 on-demand specialist median
- $6.92/GPU·hr
- +20% y/yBlackwell demand outran supply
- NVIDIA data-center revenue · FY27 Q1
- $75.2B
- +92% y/yNext-quarter guide $91B total
- Tracked AI capex · 2026E
- $780B
- +66% vs ’25MSFT · GOOG · META · AMZN · ORCL
- GPT-4-class token floor
- $0.18/1M
- −99.5% since Mar ’23DeepSeek V4 Flash · blended 3:1
- H100 venue spread
- 4.0×
- hyperscaler floor ÷ marketplace floorAWS (p5) $6.88 ÷ Vast.ai $1.73/GPU·hr — venues sell different products
Today’s read · market snapshot AUG 28 2026 · 22:35 UTC
Compute is becoming a commodity — the fourth factor of production — and for the first time it trades like one.
RTX 5090 repriced +13% since Aug 21. Its curve is easing: the strip drifts lower into the back months. A100 is the anchor of the board: ~$1.14 through Aug ’27 on a flat curve.
The monthly-average anchor is $0.62; the Aug ’27 tenor implies $0.49.
The marketplace floor moved −3.9% over the latest three-quarter window to $1.73/GPU·hr.
The GPT-4-class blended floor is now about 210× cheaper than launch at $0.18 per million tokens.
CME Group is set to list Silicon Data H100 and B200 rental-index futures on NYMEX, pending CFTC review. Venue monitor
Desk log · latest entries
Full log →- Anchor moveRTX 5090 anchor repriced +6.9% between sessions
- Curve sessions1 forward-curve session archived · week of Aug 24, 2026
- Anchor moveRTX 5090 anchor repriced +5.5% between sessions
- Curve sessions2 forward-curve sessions archived · week of Aug 17, 2026
Derived from the immutable archives on every build — no stored editorial event file.
02The Board
Every tracked chip, one index grid
Series-labeled reference list prices for all 16 tracked chips — with venue-class spreads, forward anchors where a market lists one, and archive deltas that self-activate as the daily record grows.
The Board · one reference price per chip
Every chip in the survey, led by its citable reference series — the specialist on-demand median, falling back (always labeled) to the hyperscaler on-demand floor, then the marketplace-access floor. Deltas compare the two latest archive dates; all figures are posted list/ask prices, not transactions. Manual survey 2026-07-06; automated rows through 2026-07-23; forward anchors from the AUG 28 2026 · 22:35 UTC market snapshot.
| Chip | Reference list price | Δ vs prior aggregate | Class spread | Forward anchor | Archive trend |
|---|---|---|---|---|---|
| B300 | $7.62/GPU·hrspecialist on-demand median | — first obs. | 0.9×spec median ÷ mkt floor specialist on-demand median $7.62/GPU·hr, marketplace-access floor $8.34/GPU·hr | no listed forward | building history |
| B200 | $6.92/GPU·hrspecialist on-demand median | — first obs. | 2.2×hyp floor ÷ mkt floor marketplace-access floor $6.36/GPU·hr, specialist on-demand median $6.92/GPU·hr, hyperscaler on-demand floor $14.24/GPU·hr | $6.03Sep 4 weekly | building history |
| GB200 NVL72 | $10.50/GPU·hrspecialist on-demand median | — first obs. | single class — no spread to measure | no listed forward | building history |
| H200 | $4.39/GPU·hrspecialist on-demand median | — first obs. | 2.0×hyp floor ÷ mkt floor marketplace-access floor $3.93/GPU·hr, specialist on-demand median $4.39/GPU·hr, hyperscaler on-demand floor $7.91/GPU·hr | $4.59Sep 4 weekly | building history |
| H100 | $3.88/GPU·hrspecialist on-demand median | — first obs. | 4.0×hyp floor ÷ mkt floor marketplace-access floor $1.73/GPU·hr, specialist on-demand median $3.88/GPU·hr, hyperscaler on-demand floor $6.88/GPU·hr | $3.00Sep 4 weekly | building history |
| A100 80GB | $2.30/GPU·hrspecialist on-demand median | — first obs. | single class — no spread to measure | $1.14Sep 4 weekly | building history |
| MI355X | no comparable series3 quotes outside comparison policies | — | single class — no spread to measure | no listed forward | building history |
| MI300X | $3.45/GPU·hrspecialist on-demand median | — first obs. | single class — no spread to measure | no listed forward | building history |
| L40S | $1.50/GPU·hrspecialist on-demand median | — first obs. | single class — no spread to measure | no listed forward | building history |
| RTX 5090 | $0.99/GPU·hrspecialist on-demand median | — first obs. | 3.2×spec median ÷ mkt floor marketplace-access floor $0.31/GPU·hr, specialist on-demand median $0.99/GPU·hr | $0.62Aug ’26 monthly avg | building history |
| RTX 4090 | $0.69/GPU·hrspecialist on-demand median | — first obs. | 2.7×spec median ÷ mkt floor marketplace-access floor $0.26/GPU·hr, specialist on-demand median $0.69/GPU·hr | no listed forward | building history |
| V100 | no comparable series1 quote outside comparison policies | — | single class — no spread to measure | no listed forward | building history |
| TPU v6e | $2.70/chip·hrhyperscaler on-demand floor | — first obs. | single class — no spread to measure | no listed forward | building history |
| TPU v5e | $1.20/chip·hrhyperscaler on-demand floor | — first obs. | single class — no spread to measure | no listed forward | building history |
| TPU v5p | no comparable series1 quote outside comparison policies | — | single class — no spread to measure | no listed forward | building history |
| Trainium2 | no comparable series1 quote outside comparison policies | — | single class — no spread to measure | no listed forward | building history |
03GPU cloud
Renting an H100, 2023 → today
US$ per GPU-hour. From $12 at peak scarcity to a stable ~$4; primary on-demand now starts at $1.99, with separately labeled marketplace access near $1.73.
H100 SXM · provider and marketplace price
Quarterly midpoints by provider class
Data — endpoints, license, and freshness behind H100 SXM · provider and marketplace price
The complete GPU price-history ledger: every series and dated observation, with billing, comparison, quality, and raw-evidence fields.
curl -s https://aicomputetracker.com/price-history.jsonPer-accelerator shard of the price-history ledger — the same series and point shapes filtered to one chip.
curl -s https://aicomputetracker.com/price-history/h100.jsonNo API key, CORS open (Access-Control-Allow-Origin: *) — fetch it from a browser, a notebook, or a cron job.
Licensed CC BY 4.0 — attribution “AI Compute Tracker (aicomputetracker.com), CC BY 4.0”.
Live quotes
Cite — copy a citation for H100 SXM · provider and marketplace price
Copy-ready citation
Cite as: AI Compute Tracker, “H100 SXM · provider and marketplace price”, live quotes as of 2026-07-23, https://aicomputetracker.com/#gpu-cloud, AI Compute Tracker (aicomputetracker.com), CC BY 4.0, accessed <date you visit>
Replace “<date you visit>” with your access date.
Provider list prices, launch announcements, and archived snapshots; legacy quarterly midpoints ±15%. Current 2026-07-06 quote-table aggregates are provisional unless an evidence record pins an exact archived Git object.
04The macro picture
Compute demand is compounding
The frontier training-compute trend runs at 4.4× a year; tracked 2026E capex is +66% versus 2025.
Training compute of notable models
Total FLOP, log scale — estimates marked
Epoch AI estimates and lab disclosures; estimated points carry wide error bars.
Big-tech capital expenditure
US$B per calendar year (2026 = guidance)
Company earnings releases and SEC filings; Microsoft includes finance leases; Oracle fiscal year mapped.
05Inference
Intelligence, repriced monthly
GPT-4-level capability became roughly 210× cheaper since launch. Then 2026 flagships reintroduced a frontier premium.
Cheapest GPT-4-class · blended
$0.18/1M
−99.5% since Mar ’23 · DeepSeek V4 Flash
OpenAI flagship · GPT-5.5
$11.25/1M
+227% vs GPT-5 · Flagship list price rose in Apr 2026
Batch & cached reality
−50 / −90%
Batch APIs ~50% off list; cached input ~90% off — production pays below list
06Quick answers
The methodology, without the footnotes
The three definitions readers need before comparing the curves, provider classes, or token-price series.
How are the GPU forward curves derived?
From Kalshi event-contract ladders on GPU rental prices referenced to Ornn's hourly index. The first point is a future exact-time weekly implied median for B200, H200, H100, and A100; RTX 5090 uses its front-month monthly-average implied median. Each raw implied level is the ladder's 50% crossing. The monthly strip receives two [1,2,1] smoothing passes from the third monthly tenor onward. Thin and interpolated points are labeled; none are current spot or executable quotes.
What does the blended token price mean?
Blended price = (3 × input + output) ÷ 4 per million tokens, reflecting a typical 3:1 read/write mix at standard list rates.
What are the provider classes?
Under primary on-demand policy primary-on-demand-v1, hyperscaler is the undiscounted big-cloud provider-list floor and specialist is the AI-native provider-list median. Marketplace access is the separately labeled open-market floor on community hardware.
07Go deeper
Explore the full desk
01
GPU Cloud
Primary on-demand rates for 16 chips across 19+ providers, with separately labeled marketplace access and history.
H100 on-demand from $1.99/hr
02
Hardware
Specs and economics for every accelerator that matters, V100 → Rubin.
Dense BF16 $/PFLOP·hr
03
Inference
Token prices across OpenAI, Anthropic, Google, xAI, and DeepSeek.
Floor $0.18/1M
04
Buildout
Capex, NVIDIA’s run-rate, and every gigawatt-class cluster.
~36 GW frontier pipeline