Silicon Data · Sep 24, 2026
Silicon Data reports a quiet August for B200 rental prices
Markdown · JSON · Original source
Silicon Data’s update, published September 24, 2026, reports that its B200 rental index entered September at 5.62 dollars per -hour, up 27.6% year to date, after August’s quietest month in the series. Silicon Data describes the index as a built from standardized provider quotes, not a traded price; it says the measure can be compared with an invoice and is used to settle B200 futures, pending regulatory review. The update says the term structure sat below spot at every tenor on September 7. For buyers with known multi-month workloads, Silicon Data says those prices quote a discount for commitment, while emphasizing that the curve describes what commitment costs now, not future rates. It also argues that B200’s roughly twofold price relative to through the summer is consistent with a structural generational spread, while noting that price spacing is not a performance ratio and depends on workload. Silicon Data’s conclusion is that the daily series will record whether August’s calm was a resting level or a pause. It says capacity arrives in lumps, quotes can move together, and three quiet months are not a trend. The announced futures are designed to transfer the difference between the index and a level agreed in advance; the update is not a recommendation to trade.
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
Silicon Data · Article
Source details
Original source date: Sep 24, 2026 (2026-09-24).
Date basis: Date stated on the publisher’s page.
Added to Escape Meridian: 2026-11-05T14:48:23.182Z. First observed: 2026-10-24T21:09:50.539Z. Retrieved: 2026-10-24T21:11:38.254Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
Text as retrieved on October 24, 2026; originally published September 24, 2026. The update says the futures listing is pending CFTC review; the evidence does not establish the outcome of that review.
Open question
As B200 capacity arrives in lumps and provider quotes move, will the index record August’s calm as a resting level or a pause?
Source-stated
Silicon Data reports that its B200 rental index entered September at $5.62 per GPU-hour, up 27.6% year to date, and describes August as unusually quiet.
The SDB200RT index, our daily benchmark of B200 rental rates quoted by neocloud providers, entered September at 5.62 dollars per GPU-hour, up 27.6% year to date. August was the quietest month in the series so far
Source-stated
Silicon Data says the index is a benchmark derived from standardized provider quotes, not a traded price.
It is a benchmark, not a traded price: the number to compare an invoice against, and from October the number that B200 futures settle against.
Source-stated
Silicon Data reports that the term structure was below spot at every tenor on September 7, and says it describes current commitment costs rather than a forecast of future prices.
on September 7 it sat below spot at every tenor: 5.73 dollars per hour for a 3-month term, 5.64 for 6 months, and 5.39 for 12 months
Source-stated
Silicon Data says the B200-to-H100 price relationship is consistent with a structural generational spread, while noting that price spacing is not a performance ratio and depends on workload.
A newer generation renting at about twice the prior one, and holding that ratio through the summer, is consistent with the market pricing the generational gap as a structural spread rather than a launch premium that decays. The H200's smaller premium over the H100, reviewed in our H200 vs H100
Source-stated
Silicon Data says the quiet data does not establish whether August’s calm is a resting level or a pause, and that three quiet months are not a trend.
What it does not establish is whether August's calm is a resting level or a pause. B200 is a young index in a market where capacity arrives in lumps and quotes can move together when it does, and three quiet months are not a trend.
Source-stated
Silicon Data says the announced futures are designed to transfer the difference between the index and a level agreed in advance, and states that the update is not a recommendation to trade.
From October, that is what the contract is designed to carry: the difference between the index and a level agreed in advance, independent of any one provider's terms. The practitioner's guide to the contracts walks through that arithmetic step by step.
Source and contributor context
Silicon Data · Publisher
The publisher is identified here. No individual author has been inferred.
Source media
Media retrieval status: unchecked.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Graphics processing unit
A chip that performs many calculations at once; the main hardware used to train and run AI models.
Benchmark
A defined set of tests used to compare systems. Results depend on the tasks and measures selected.
NVIDIA H100
An NVIDIA graphics processing unit (GPU) built on the company's Hopper architecture and designed to accelerate AI workloads, including large language models.
Editorial provenance
Version 1. Publication ID: f3198921-defb-84cd-7191-8c9246d41bd5.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-05T14:43:32.574Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_admitted_by_model, second_opinion_categorical.
Observation: dc9b91b7c02780565e8f348542ab6327d9f3d78ca9ae728639585807502bcb3f. Source text hash: 4300f7a1f7a67ebb1fab415fde32b1839413fc36ab1a2c04dd0a15475d2e1c04.
Revision and source-check history
Version 1 · 2026-11-05T14:48:23.182Z
Silicon Data reports a quiet August for B200 rental prices
Silicon Data’s update, published September 24, 2026, reports that its B200 rental index entered September at 5.62 dollars per -hour, up 27.6% year to date, after August’s quietest month in the series. Silicon Data describes the index as a built from standardized provider quotes, not a traded price; it says the measure can be compared with an invoice and is used to settle B200 futures, pending regulatory review. The update says the term structure sat below spot at every tenor on September 7. For buyers with known multi-month workloads, Silicon Data says those prices quote a discount for commitment, while emphasizing that the curve describes what commitment costs now, not future rates. It also argues that B200’s roughly twofold price relative to through the summer is consistent with a structural generational spread, while noting that price spacing is not a performance ratio and depends on workload. Silicon Data’s conclusion is that the daily series will record whether August’s calm was a resting level or a pause. It says capacity arrives in lumps, quotes can move together, and three quiet months are not a trend. The announced futures are designed to transfer the difference between the index and a level agreed in advance; the update is not a recommendation to trade.
Original source
Silicon Data · Article
Source details
Original source date: Sep 24, 2026 (2026-09-24).
Date basis: Date stated on the publisher’s page.
Added to Escape Meridian: 2026-11-05T14:48:23.182Z. First observed: 2026-10-24T21:09:50.539Z. Retrieved: 2026-10-24T21:11:38.254Z.
Limits
Text as retrieved on October 24, 2026; originally published September 24, 2026. The update says the futures listing is pending CFTC review; the evidence does not establish the outcome of that review.
Open question
As B200 capacity arrives in lumps and provider quotes move, will the index record August’s calm as a resting level or a pause?
Source-stated
Silicon Data reports that its B200 rental index entered September at $5.62 per GPU-hour, up 27.6% year to date, and describes August as unusually quiet.
The SDB200RT index, our daily benchmark of B200 rental rates quoted by neocloud providers, entered September at 5.62 dollars per GPU-hour, up 27.6% year to date. August was the quietest month in the series so far
Source-stated
Silicon Data says the index is a benchmark derived from standardized provider quotes, not a traded price.
It is a benchmark, not a traded price: the number to compare an invoice against, and from October the number that B200 futures settle against.
Source-stated
Silicon Data reports that the term structure was below spot at every tenor on September 7, and says it describes current commitment costs rather than a forecast of future prices.
on September 7 it sat below spot at every tenor: 5.73 dollars per hour for a 3-month term, 5.64 for 6 months, and 5.39 for 12 months
Source-stated
Silicon Data says the B200-to-H100 price relationship is consistent with a structural generational spread, while noting that price spacing is not a performance ratio and depends on workload.
A newer generation renting at about twice the prior one, and holding that ratio through the summer, is consistent with the market pricing the generational gap as a structural spread rather than a launch premium that decays. The H200's smaller premium over the H100, reviewed in our H200 vs H100
Source-stated
Silicon Data says the quiet data does not establish whether August’s calm is a resting level or a pause, and that three quiet months are not a trend.
What it does not establish is whether August's calm is a resting level or a pause. B200 is a young index in a market where capacity arrives in lumps and quotes can move together when it does, and three quiet months are not a trend.
Source-stated
Silicon Data says the announced futures are designed to transfer the difference between the index and a level agreed in advance, and states that the update is not a recommendation to trade.
From October, that is what the contract is designed to carry: the difference between the index and a level agreed in advance, independent of any one provider's terms. The practitioner's guide to the contracts walks through that arithmetic step by step.
Source and contributor context
Silicon Data · Publisher
The publisher is identified here. No individual author has been inferred.
Source media
Media retrieval status: unchecked.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Graphics processing unit
A chip that performs many calculations at once; the main hardware used to train and run AI models.
Benchmark
A defined set of tests used to compare systems. Results depend on the tasks and measures selected.
NVIDIA H100
An NVIDIA graphics processing unit (GPU) built on the company's Hopper architecture and designed to accelerate AI workloads, including large language models.
Editorial provenance
Version 1. Publication ID: f3198921-defb-84cd-7191-8c9246d41bd5.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-05T14:43:32.574Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_admitted_by_model, second_opinion_categorical.
Observation: dc9b91b7c02780565e8f348542ab6327d9f3d78ca9ae728639585807502bcb3f. Source text hash: 4300f7a1f7a67ebb1fab415fde32b1839413fc36ab1a2c04dd0a15475d2e1c04.
uncertain · 2026-11-05T14:43:32.574Z · anthropic/claude-opus-5.5
Publication: f3198921-defb-84cd-7191-8c9246d41bd5
attribution — pass: This is a single Silicon Data blog update with its own analysis. The reading consistently attributes the index values, term-structure interpretation, generational-spread argument and caveats to Silicon Data rather than presenting them as established fact.
SDB200RT is Silicon Data's daily benchmark for renting NVIDIA's Blackwell-generation B200 on the neocloud tier
summary — uncertain: The reading faithfully keeps the source's argument and caveats. Several central terms are unexplained and are not in the definitions list: B200 (NVIDIA's Blackwell-generation GPU), spot, term structure, tenor and backwardation, which the reading implies without naming. A general reader cannot easily follow 'the term structure sat below spot at every tenor.' Smallest repair: add tooltips for these terms, or a plain gloss such as: committing for 3, 6 or 12 months was priced at or below today's hourly rate.
The spot index is the price of a B200-hour today. The term structure is the price of committing to one for longer
date — uncertain: The reading's September 24, 2026 date matches originalPublished and the byline. The page's 'Updated September 3' note predates that byline, which is the source's own inconsistency. However, the reading says the index 'is used to settle B200 futures.' In the source, those futures list only from October 5, after publication, and are pending CFTC review. The hedge partly attributes this, but the present tense reads as an outcome. Smallest repair: 'is set to settle B200 futures from October, pending regulatory review.'
As currently announced, and pending CFTC review, NYMEX lists B200 rental index futures for trading from October 5 under the code GPU2
support — pass: Every claim matches the source text. This covers the 5.62 level and the 27.6% year-to-date rise, the quietest month, the benchmark rather than traded-price status, and the curve below spot on September 7 as a cost of commitment rather than a forecast. It also covers the roughly 2x spread over the H100 as consistent with a structural spread, the not-a-performance-ratio caveat, the resting-level-or-pause uncertainty, and the no-trade-recommendation disclaimer. The question follows from the source's own open issue.
Nor does a backwardated curve establish that rates will fall; it establishes what commitment costs today.
media — pass: No media was selected, and no verified media candidates were supplied, so nothing is misidentified.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
A chip that performs many calculations at once; the main hardware used to train and run AI models.
Definition referenceA defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition referenceAn NVIDIA graphics processing unit (GPU) built on the company's Hopper architecture and designed to accelerate AI workloads, including large language models.
Definition referenceA chip that performs many calculations at once; the main hardware used to train and run AI models.
Definition referenceA defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition referenceAn NVIDIA graphics processing unit (GPU) built on the company's Hopper architecture and designed to accelerate AI workloads, including large language models.
Definition reference