GPU rental price indexes compared: what each one counts

Summary

The GPU rental price indexes compared here differ in the prices they count, the basis they adjust to and the rules that govern them, as each publisher's own documents, viewed on 7 November 2026, describe.

• Ornn's Compute Price Indices (OCPI) count only realized prices paid under active on-demand GPU rentals, as a volume-weighted mean with extreme prices capped (winsorized) over a rolling one-hour window, and leave out reserved, long-term and forward contracts and indicative or offered prices [1].
• Silicon Data's H100 and B200 rental indexes, which NYMEX's proposed futures settle to, combine verified executed transactions with verified executable listings from non-hyperscaler (neocloud) providers and GPU marketplaces, adjusted to an on-demand U.S. baseline configuration, and exclude traditional hyperscaler rate cards [8, 10, 24].
• SemiAnalysis builds its one-year H100 contract price index from a monthly survey of more than 100 market participants, as a 25th to 75th percentile range validated against transaction data, and writes that most of the GPU rental market trades on contracts of six months or longer [17].
• AIMultiple and Price of Compute build their indexes from the prices providers post publicly, each taking a median per provider and then a median across providers [19, 20].
• In documents viewed on 7 November 2026, Silicon Data describes a Benchmark Oversight Committee with an independent member, a third-party calculation agent and an independent attestation examination, and Ornn's methodology says Ornn "is establishing" an oversight committee and "is developing formal conflicts procedures" [1, 10].
• Futures contracts are proposed on Silicon Data's and Ornn's indexes. On 21 September 2026 the Commodity Futures Trading Commission (CFTC) extended its review of NYMEX's Silicon Data H100 and B200 rental index futures to the end of 9 November 2026, and on 4 November NYMEX consented, at the CFTC's request, to a further extension to 8 January 2027 [24, 25, 35]. ICE Futures U.S. said on 29 September 2026 that no listing date was set for futures on Ornn's H100 and B200 indexes [7].
• A request for comment from the Commodity Futures Trading Commission, published on 21 August 2026, asks what would prevent manipulation of compute settlement prices built from posted rates or from venues that a small number of participants operate or dominate, and what data thresholds such prices need [29].
• The Rillor Compute Index methodology v1.1 builds its rental series from listings in providers' official pricing APIs and published pricing pages, as a median of provider medians, and states that its series are for research and planning use and are not offered for settlement of any contract [34].

Background

GPU capacity rents on several kinds of terms. Silicon Data lists on-demand and spot rentals and reserved commitments "with terms ranging from 1 month to 60 months", and writes that "a short-duration Spot rental in one region is not economically equivalent to a multi-year Reserved contract in another" [11]. Its October 2026 post, defining reserved capacity, adds: "Most GPU capacity is sold this way" [14].

The CFTC set out its preliminary understanding of the market in August 2026 [29]. It wrote that "compute markets are fragmented and price formation primarily occurs in opaque bilateral transactions, hindering the availability of current and historical price data", and that "pricing can vary dramatically across providers, regions, and contract structures" [29]. It also understood that "non-public, bilateral agreements carry the majority of economic value" [29].

Several publishers built indexes in that market. Silicon Data launched its H100 Rental Index, Bloomberg ticker SDH100RT, on 27 May 2025 [16]. Ornn's index reached "six months as a live index" when it was added to the Bloomberg Terminal on 2 April 2026 [4]. SemiAnalysis launched its one-year H100 contract index on 2 April 2026, and says it has kept rental indexes for clients "Since 2023" [17].

Several of these series reach readers through the Bloomberg Terminal under their publishers' tickers: SDH100RT for Silicon Data [16], ORNNH100 and ORNNB200 for Ornn [3] and SAH100SC for SemiAnalysis [18]. Each ticker belongs to the publisher that calculates the series.

Exchanges then announced futures on some of these indexes: CME Group with Silicon Data on 12 May 2026 [22], ICE with Ornn on 19 May [6] and ICE with NATIVX on 1 July [21]. The CFTC requested comment on 19 August, a step its chairman called "the first step toward establishing clear rules of the road for American compute markets" [28]. In an academic paper, Bandi and Su write that "At least two start-up companies, Silicon Data and Ornn, have compiled and launched public compute price indexes", and that the CFTC request "cites an earlier version of this paper" [33].

Method

What was examined. The published methods and governance documents of GPU rental price indexes available on 7 November 2026: Ornn's OCPI [1-7], Silicon Data's GPU Rental Index series [8-16], SemiAnalysis's contract and composite indexes [17, 18], AIMultiple's Cloud GPU Rental Price Index [19], Price of Compute's index [20] and the Rillor Compute Index methodology v1.1 [34]. NATIVX's COIL index is described from its launch release only [21].

Sources. Publishers' methodology pages, product pages, releases and comment letters [5, 15], exchange notices, filings and CME Group's comment letter [7, 24, 26, 35], the CFTC's release, request, review letter and docket [25, 28-30], IOSCO's Principles for Financial Benchmarks [31], Appendix C to 17 CFR Part 38 [32] and one academic paper [33], each opened on 7 November 2026.

How. Each index is described only from its publisher's own dated documents, in their words where the wording matters. Where two documents state different figures for the same index, both are given with their dates and scopes. IOSCO's principles and Appendix C frame questions only. No index is scored against them and none is ranked. No index values are reported, because the report compares methods.

Disclosure. Rillor Corporation publishes this report and administers the Rillor Compute Index, one of the indexes described. The Rillor Compute Index is described only as its methodology v1.1 states [34].

Findings

1. The indexes and their publishers, as of 7 November 2026

Silicon Data's governance framework calls SDH100RT and SDB200RT its "flagship settlement indices", tracking the "non-hyperscaler (“neoclouds”) ecosystem" [10].

2. What each index counts: rates, executed rentals or surveyed contracts

OCPI. "Ornn records the price of every GPU under active on-demand rental across its contributor sources" [1]. Only "realized rental prices, meaning prices actually being paid under active rentals, are included", and "Indicative or offered prices are excluded" [1]. Eligible trades must pass a "Verified Counterparties" test, under which compute "must have been successfully transferred between two different counterparties", and "No minimum trade-size threshold is applied" [1]. Reserved, long-term and forward contracts are "outside the scope of the indices", and reserved quotes "are published separately as the GPU forward curve" [1]. In ICE's May 2026 release, OCPI "tracks live-traded spot prices", and ICE and Ornn call it "the first compute index to be built only from printed transactions" [6]. Ornn's comment letter adds that "Longer-term or reserved arrangements may trade at different levels" because of duration and commitments [5].

Silicon Data. Its governance framework describes "a broad data foundation comprising both transaction records and executable listing prices" [10]. Two inputs are admissible: verified executed transactions on spot, on-demand and reserved terms, "under binding legal data license agreements", and "Directly Executable Listings" whose availability and rates are verified [10]. Excluded are "Static marketing rate cards, unverified aggregator-relayed feeds, promotional/discounted trial rates, and records lacking verified hardware metadata" [10]. The settlement indexes "explicitly exclude traditional hyperscaler rate cards" [10]. Its October 2026 post says "Silicon Data combines posted prices with transaction data from neocloud data partners" [14].

SemiAnalysis. The one-year index is "constructed from direct survey data across a pool of 100+ market participants including Neocloud providers, buyers and sellers of compute", captured monthly, and validated "with transaction data as well as by arranging a few transactions ourselves" [17]. SemiAnalysis writes that "There is no guarantee that buyers of compute are actually transacting at prices publicly posted by Hyperscalers and Neoclouds" [17]. The composite collects prices "from hyperscale clouds, neoclouds, GPU marketplaces, and rental routers/aggregators", synthesized with contract surveys, which cover "currently only the H100" [18].

AIMultiple and Price of Compute. AIMultiple tracks "posted cloud GPU rental prices for 17 models across 75 providers, covering on-demand, spot, and reserved rates", and "Contact-sales quotes and negotiated enterprise contracts fall outside the index" [19]. Price of Compute's methodology describes observations from 16 providers every hour, taken from official APIs where a provider offers one and otherwise from public pricing pages [20]. It states: "Listed ≠ attainable" [20].

Bandi and Su use the word "surveyed" in a different sense from SemiAnalysis: "The current indexation takes a “surveyed” approach: the indexes are aggregation of observed heterogeneous compute prices" [33].

Silicon Data's documents state the size of its data in different scopes and on different dates.

CME Group's comment letter of 20 October 2026 says the SD-H100 and SD-B200 indexes average "thousands of daily transactions" [26].

3. Normalization and basis adjustment

OCPI. Prices are "denominated in USD-equivalent per GPU-hour", indexes are "localized per GPU type", and six regions are recorded: "North America, South America, Europe, Asia, Africa, and Oceania" [1]. OCPI "applies no smoothing, averaging across windows, or other filtering beyond the winsorization" [1]. Ornn's April 2026 release described pricing "normalized across hardware configuration, provider, and deployment context" [4].

Silicon Data. "Every GPU rental record is first converted to a standardized per-hour, per-GPU rate", and a basis adjustment then standardizes each record "to a benchmark-equivalent contract, accounting for rental type, geography, CPU platform, and GPU variant" [8]. The governance framework names the baseline as "an on-demand, U.S.-baseline reference configuration specific to the GPU model" [10]. "Spot and long-term reserved commitments are normalized to their on-demand equivalent", and "Non-U.S. and regional inputs are adjusted to the U.S. baseline" [10]. The H100 page adds standardization "for rental term length, cluster scale, and interconnect" [12]. The August 2026 methodology overview says "The pricing model is recalibrated weekly" [9], and CME Group's October letter says it "is calibrated using historical data and reviewed quarterly" [26].

SemiAnalysis. The contract index fixes one tenor, a year [17]. The contract table on SemiAnalysis's index page notes "25% prepay for 3m and beyond", and the composite's rule is that "Each provider contributes at most one price per hour to a given index" [18].

AIMultiple and Price of Compute. AIMultiple groups prices by provider, GPU and billing tier, and "Reservation records include different commitment lengths" [19]. Price of Compute states "Pricing types never mix" [20].

COIL. "By normalizing compute prices to a consistent energy unit, the COIL Index strips out the noise introduced by regional power cost disparities" [21].

Bandi and Su propose a "par" underlying with "published quality adjustments", "analogous to adjusting delivery prices for differences in grade in commodities such as wheat and corn" [33].

4. Aggregation, weighting and outlier treatment

Silicon Data filters records for completeness, flags day-over-day shifts for manual review and tests outliers against "dynamic historical price bands" [9]. A new provider needs "at least 90 consecutive days of internal model history" and approval by the Benchmark Oversight Committee (BOC), and weights stay "static between BOC reviews", reviewed at least every 3 months [10]. Expert judgment is confined to three circumstances, one of them a daily move "exceeding the 5% variance trigger", with "four-eyes" sign-off, and is "publicly flagged when a published index value reflects its application" [10].

SemiAnalysis writes that "sources entering or leaving an index's panel should not mechanically move its level" [18]. AIMultiple takes the median of weekly observations for each provider, GPU and billing tier, then the median across providers, and "Each GPU contributes equally to its group median" [19].

5. Windows, publication times and segments

OCPI. Values are published "as promptly as practicable following the close of each one-hour window", to "the nearest tenth of a cent ($0.001)" [1]. Ornn's Bloomberg page gives a daily settlement at 12:00 AM ET for OCPI-H100 [3]. "As a general matter, Ornn does not restate historical index values", with exceptions for data falsification or a material calculation error [1]. Bandi and Su note that "The Ornn data do not separately identify the hyper-scaler and the neo-cloud segments" [33].

Silicon Data. "Every index is processed and published once per business day" [8], by 21:00 UTC [10]. "Once a daily index value is finalized and published at 21:00 UTC, it is immutable", and late contributor feeds do not restate it [10]. Neocloud settlement indexes and hyperscaler readings for the H100 and A100 are published separately [8]. Silicon Data reported that its H100 Hyperscaler On-Demand Index was unchanged on 39 of 50 daily transitions between 1 March and 20 April 2026 [13].

SemiAnalysis. "We update the index monthly" for the one-year contract index [17]. The composite is "computed and published hourly" on its page [18].

Bandi and Su compared the series' correlations: "The Silicon Data indexes (blue) tend to be separated from the Ornn indexes (orange). This separation reflects either differences in index construction or differences in the market segments covered" [33].

6. Governance: administration, conflicts, changes and complaints

The table quotes Ornn's and Silicon Data's own governance texts.

Silicon Data has appointed Syntax Data as "the independent third-party calculation agent" and engages an independent accounting firm for an attestation examination "under AICPA AT-C 205 / ISAE 3000 standards" [10]. Ornn's methodology changelog lists only its first publication, on 24 July 2026, and on 7 November 2026 its oversight section read as quoted above [1]. It says the committee's composition is to be published upon its constitution [1]. Silicon Data discloses that its chief executive also serves as chief executive of "an independent spot GPU procurement marketplace", and that she "is excluded from all benchmark production and data validation processes, holds no voting rights on the BOC" [10].

SemiAnalysis states that its summary "describes the methodology at the level SemiAnalysis publicly discloses" [18]. AIMultiple and Price of Compute publish methodology sections with changelogs [19, 20].

7. Settlement use and the CFTC's questions

ICE and Ornn announced their contracts on 19 May 2026 "subject to regulatory approval" [6]. Ornn's page on 7 November 2026 still gave the ICE contracts' status as "Pending CFTC comment period" [3].

CME Group and Silicon Data announced their contracts on 12 May 2026 [22] and on 11 August set a launch on 5 October 2026, "pending regulatory review" [23]. NYMEX filed the same day for product approval under CFTC Regulation 40.3(a) [24]. In its Core Principle 3 statement, NYMEX wrote that the index "is sufficiently broad in definition and scope and supported by a robust level of trading in the underlying cash market" [24]. The CFTC's Division of Market Oversight extended its review on 21 September 2026 "in light of novel or complex issues" [25]. On 4 November 2026 NYMEX consented, "At the Commission’s request", to "a further 60-day extension of the review period until January 8, 2027" [27, 35].

NATIVX and ICE said on 1 July 2026 that COIL contracts were "expected to be launched later this year, subject to completion of relevant regulatory processes" [21].

The CFTC's request for comment, with comments due by 20 October 2026 [29], asks:

• "What proportion of compute transactions occur at publicly disclosed prices, and what proportion does not?" (Question 1(c)) [29].
• Whether a listed futures contract settling to a published compute index would "change provider incentives with respect to the publication of posted rates" (1(f)) [29].
• What would prevent a provider "from manipulating a cash settlement index by adjusting a posted rate", by steering capacity between venues or by trading or declining to trade during the observation window (2(b)) [29].
• "What thresholds would be appropriate for a compute settlement reference price" (2(c)) [29].
• Whether "a cash price series for compute cash markets" exists that is "reliable, acceptable, publicly available, and timely" (2(g)) [29].

Question 2(b) opens with the CFTC's description: "Certain published compute price series are constructed in whole or in part from posted or listed rates that the compute capacity providers themselves administer, with the remainder of transactions executed on venues that a small number of participants operate or dominate" [29]. The docket held 108 comments on 7 November 2026 [30].

Index publishers answered in their own letters. Ornn wrote that administrators "should use a transaction-based methodology and clearly define the economic unit being measured", and that thresholds and contributor limits "should not be prescribed without regard to the structure of the underlying market" [5]. Silicon Data suggested a settlement index "calculated each business day from no fewer than 5 independent providers, with no single provider accounting for more than 20% of index weight" [15]. CME Group wrote that SD-H100 and SD-B200 prices "have been published continuously without interruption since inception", over 729 and 395 days [26].

8. Questions to ask of any GPU index, from IOSCO and Appendix C

IOSCO's principles of July 2013 and Appendix C to Part 38 of the CFTC's rules yield neutral questions for any GPU rental price index [31, 32]. They frame a reading. This report scores no index against them.

IOSCO asks administrators to disclose "the extent of their compliance with the Principles annually" [31]. The publishers' statements are their own: Silicon Data's "in alignment with" [10], Ornn's "designed with reference to" [1], and Silicon Data's letter describing its indexes as "designed to satisfy the Appendix C criteria" [15].

Bandi and Su ask a settlement index for transparency, representativeness and resistance to manipulation, and add: "These are requirements for assessing the suitability of an index for futures settlement, rather than an endorsement of the existing indexes used in our empirical analysis" [33].

9. What the Rillor Compute Index methodology states on the same questions

The list restates methodology v1.1 of 7 November 2026 [34]. It makes no claim about IOSCO's principles or Appendix C.

• Unit and eligible prices. Four series: RCI Rental in USD per GPU-hour, RCI Availability in percent, RCI Hardware in USD per system and RCI Rent-to-Value as a ratio. A source qualifies "when each of its prices can be checked against the source at its capture time and its terms of use allow the use the index makes of them" [34].
• Evidence. "Prices from providers' official pricing APIs and published pricing pages are listings, and RCI Rental and RCI Availability are built from them." "The classes stay apart in calculation, so a listing never counts as a sale" [34].
• Basis. Prices per GPU-hour in USD, a node's price divided by its GPU count, per-second or per-minute billing restated per hour, and currency converted at a published rate recorded with the observation. Each rental observation is mapped to "GPU class and variant, interconnect, node size, region, term and pricing type" [34].
• Calculation. Each provider's median, then the median of provider medians, with each provider at equal weight "whatever the number of offers it lists". Hyperscaler and neocloud segments are calculated and shown separately [34].
• Minimum data and concentration. At least 90 days of captured history before a first value, at least 8 independent provider groups in each neocloud print and 3 in each hyperscaler print, no provider group above 25% of a neocloud print's weight, and a recorded terms-of-use posture for every source [34].
• Missing data. A print that misses a data rule is withheld, "and no earlier value is carried forward in its place". A series that stops meeting the rules is suspended, and the suspension is noted [34].
• Conflicts. "Rillor Corporation administers the index." "Rillor also supplies GPU systems and arranges forward agreements. Its own quotes and agreements are excluded from every published series" [34].
• Judgment, corrections, changes and complaints. Judgment "never sets a published value directly". Corrections are marked and the original kept. Changes carry a new version number and effective date. Complaints to contact@rillor.com are "logged and answered in writing" [34].
• Use. "A posted price is an offer", and "RCI Rental records the posted price." "Rillor Compute Index series are for research and planning use. They are not offered for settlement of any contract" [34].

Implications

For buyers and planners reading a GPU price. Match the index to the contract in view. An on-demand transaction index, a posted-rate index and a one-year contract survey record different prices [1, 17, 19]. The basis matters as much: Silicon Data converts spot and reserved records to an on-demand U.S. baseline [10], while SemiAnalysis holds a one-year tenor [17].

For analysts comparing two series. A gap between two indexes can come from construction or from the segment covered, as Bandi and Su note [33]. Record counts differ by document, date and scope, so a count is read with its scope (Finding 2).

For readers of settlement-linked indexes. On 7 November 2026 both proposed contract families awaited regulatory steps [3, 25]. Both average daily prices over a month [7, 24], so the daily method sets the settlement.

For anyone relying on an index. The questions in Finding 8 apply to every publisher. Ornn, Silicon Data and Rillor each publish conflict disclosures for their indexes [1, 10, 34].

Limits of this analysis

• Methods are compared as publicly disclosed. Silicon Data's methodology overview carries a customer-use notice, so only brief facts are taken from it [9]. SemiAnalysis's full term structure is available to institutional subscribers [17]. Ornn's winsorization level and verification thresholds are kept for licensed parties, venues, regulators and auditors [1].
• No index values were collected, and no index's accuracy was tested.
• Regulatory status lines are dated 7 November 2026 or earlier, and status in this area changes quickly [3, 25, 27, 35].
• AIMultiple's page gives 75 providers in its summary and 86 for its curated history [19]. Price of Compute's methodology describes 16 providers, while its header showed 12 on 7 November 2026 [20].
• Rillor administers one of the indexes compared, so each index is described only from its publisher's documents.

Sources

1. Ornn Compute Price Indices (OCPI): Methodology. Ornn Data LLC, first published 24 July 2026, viewed 7 November 2026. https://data.ornn.com/methodology
2. GPU Market Prices: the Ornn Compute Price Index. Ornn Data, viewed 7 November 2026. https://data.ornn.com/markets
3. Bloomberg Tickers for Ornn GPU and Token Price Indices. Ornn Data, viewed 7 November 2026. https://data.ornn.com/markets/bloomberg
4. Ornn Compute Price Index Added to Bloomberg Terminal. Ornn AI Inc. via PR Newswire, copy on Yahoo Finance, 2 April 2026, viewed 7 November 2026. https://finance.yahoo.com/sectors/technology/articles/ornn-compute-price-index-added-120500818.html
5. Bavaria and Nelms. Comment letter CFTC-2026-1850-0058 on RIN 3038-AF77. Ornn AI, Inc., 19 October 2026, viewed 7 November 2026. https://downloads.regulations.gov/CFTC-2026-1850-0058/attachment_1.pdf
6. ICE and Ornn to Launch GPU Compute Futures Contracts. Intercontinental Exchange, Inc., 19 May 2026, viewed 7 November 2026. https://s2.q4cdn.com/154085107/files/doc_news/ICE-and-Ornn-to-Launch-GPU-Compute-Futures-Contracts-2026.pdf
7. New Compute Index Reference Data to Be Made Available on Monday, October 5, 2026. Exchange notice. ICE Futures U.S., 29 September 2026, viewed 7 November 2026. https://www.ice.com/publicdocs/futures_us/exchange_notices/ICE_Futures_US_Exchange_Notice_ORNN_Compute_4Q2026_20260929.pdf
8. GPU Rental Price Indices. Silicon Data, viewed 7 November 2026. https://www.silicondata.com/products/silicon-index
9. Silicon Data GPU Rental Index Methodology Overview. Silicon Data, updated 27 August 2026, viewed 7 November 2026. https://www.silicondata.com/documents/silicon-data-index-methodology.pdf
10. Silicon Data GPU Rental Index Series: Governance, Methodology & Control Framework. Silicon Data, undated, viewed 7 November 2026. https://www.silicondata.com/documents/silicon-data-index-governance.pdf
11. Li. Building a Robust GPU Index. Silicon Data, 12 March 2026, viewed 7 November 2026. https://www.silicondata.com/blog/building-a-robust-gpu-index
12. H100 Rental Price Index: SDH100RT. Silicon Data, viewed 7 November 2026. https://www.silicondata.com/products/silicon-index/h100
13. Li. H100 Hyperscaler Index, April 2026: A Benchmark in Flat Mode. Silicon Data, 27 April 2026, viewed 7 November 2026. https://www.silicondata.com/blog/h100-hyperscaler-index-april-2026
14. Hou. Dynamic Pricing Brings Price Discovery to GPU Spot Rental Markets. Silicon Data, 21 October 2026, viewed 7 November 2026. https://www.silicondata.com/blog/gpu-spot-prices-dynamic-pricing-price-discovery
15. Li. Comment letter CFTC-2026-1850-0064 on RIN 3038-AF77. Silicon Data, 20 October 2026, viewed 7 November 2026. https://downloads.regulations.gov/CFTC-2026-1850-0064/attachment_1.pdf
16. Silicon Data Launches World's First Daily GPU Rental Index. Silicon Data via EIN Presswire, 27 May 2025, viewed 7 November 2026. https://www.techdogs.com/tech-news/ein-presswire/silicon-data-launches-worlds-first-daily-gpu-rental-index-to-bring-transparency-to-the-ai-infrastructure-economy
17. Nishball, Nanos, Cheang et al. The Great GPU Shortage: Rental Capacity, Launching our H100 1 Year Rental Price Index. SemiAnalysis, 2 April 2026, viewed 7 November 2026. https://newsletter.semianalysis.com/p/the-great-gpu-shortage-rental-capacity
18. H100 Daily & Contract Pricing Index. SemiAnalysis, viewed 7 November 2026. https://gpu-index.semianalysis.com/
19. Sarı. Cloud GPU Rental Price Index. AIMultiple, updated 16 September 2026 (weekly history through 2 November 2026), viewed 7 November 2026. https://aimultiple.com/gpu-index
20. Flanary. Price of Compute: Methodology & About. Price of Compute, initial methodology 9 August 2026, viewed 7 November 2026. https://www.priceofcompute.com/methodology
21. ICE and NATIVX to Launch Energy-Normalized Compute Futures Contracts. Intercontinental Exchange and NATIVX via Business Wire, copy on Yahoo Finance, 1 July 2026, viewed 7 November 2026. https://finance.yahoo.com/energy/articles/ice-nativx-launch-energy-normalized-123000323.html
22. CME Group and Silicon Data Partner to Launch First Compute Futures. CME Group and Silicon Data via PR Newswire, copy on StockTitan, 12 May 2026, viewed 7 November 2026. https://www.stocktitan.net/news/CME/cme-group-and-silicon-data-partner-to-launch-first-compute-xsea0cl8o8z2.html
23. CME Group and Silicon Data to Launch Compute Futures on October 5. CME Group and Silicon Data via PR Newswire, copy on StockTitan, 11 August 2026, viewed 7 November 2026. https://www.stocktitan.net/news/CME/cme-group-and-silicon-data-to-launch-compute-futures-on-october-5-to-wgqge1aqpsum.html
24. Submission No. 26-370: Voluntary Submission for Product Approval. New York Mercantile Exchange, Inc., 11 August 2026, viewed 7 November 2026. https://www.cftc.gov/filings/ptc/ptc08112615405.pdf
25. Letter extending the review of NYMEX Submission No. 26-370. CFTC Division of Market Oversight, 21 September 2026, viewed 7 November 2026. https://www.cftc.gov/filings/documents/2026/orgdcmnymexcompcontr260921.pdf
26. Marchese. Comment letter CFTC-2026-1850-0080 on RIN 3038-AF77. CME Group Inc., 20 October 2026, viewed 7 November 2026. https://downloads.regulations.gov/CFTC-2026-1850-0080/attachment_1.pdf
27. Cuofano. CME Delays Nvidia GPU Rental Futures Again, to January. FourWeekMBA, 6 November 2026, viewed 7 November 2026. https://fourweekmba.com/ai-cme-delays-nvidia-gpu-rental-futures-again-to-january/
28. CFTC Requests Comment on the Listing of Compute Derivatives Contracts. Release 9286-26. Commodity Futures Trading Commission, 19 August 2026, viewed 7 November 2026. https://www.cftc.gov/PressRoom/PressReleases/9286-26
29. Request for Comment on the Listing of Compute Derivatives Contracts, 91 FR 54259 (RIN 3038-AF77). Commodity Futures Trading Commission, Federal Register, 21 August 2026, viewed 7 November 2026. https://www.govinfo.gov/content/pkg/FR-2026-08-21/pdf/2026-17163.pdf
30. Comments on docket CFTC-2026-1850. Regulations.gov, viewed 7 November 2026. https://www.regulations.gov/docket/CFTC-2026-1850/comments
31. Principles for Financial Benchmarks, Final Report (FR07/13). The Board of the International Organization of Securities Commissions, July 2013, viewed 7 November 2026. https://www.iosco.org/library/pubdocs/pdf/IOSCOPD415.pdf
32. Appendix C to Part 38: Demonstration of Compliance That a Contract Is Not Readily Susceptible to Manipulation. 17 CFR Part 38. eCFR, up to date as of 5 November 2026, viewed 7 November 2026. https://www.ecfr.gov/current/title-17/chapter-I/part-38/appendix-Appendix%20C%20to%20Part%2038
33. Bandi and Su. (Early) AI Compute Asset Pricing. arXiv 2607.12156 v4. arXiv, 21 October 2026, viewed 7 November 2026. https://arxiv.org/abs/2607.12156
34. Rillor Compute Index, Methodology version 1.1. Rillor Corporation, 7 November 2026, viewed 7 November 2026. https://rillor.com/markets/compute-index
35. Marchese. Voluntary Extension of Review Period, NYMEX Submission No. 26-370. CME Group Inc., 4 November 2026, viewed 7 November 2026. https://www.cftc.gov/filings/documents/2026/orgdcmnymexvolexten261105.pdf