Silicon Data · Oct 21, 2026
Steve Hou argues in Silicon Data that dynamic GPU spot prices reveal market conditions
Markdown · JSON · Original source
Steve Hou published this analysis in Silicon Data on October 21, 2026. He argues that Nebius’s dynamic spot prices make changes in capacity and demand more visible than posted prices, which often stay fixed. Silicon Data reports that more than half of the spot price series it tracked did not change during the nine-day window. Nebius’s , B200 and B300 prices moved on most days, while its price stayed at its floor for most of the period. Hou says this pattern is consistent with older being in ample supply and the newest in strong demand, but nine days from one dynamic seller cannot measure volatility or show how prices respond to the wider market. Dynamic pricing also has costs: buyers may lose budget certainty as rates change and workloads stop when prices exceed a set maximum; sellers face less predictable revenue and added pricing infrastructure. Hou’s conclusion is that dynamic, transparent spot pricing is a key ingredient in price discovery for compute, even as most capacity remains on term contracts and spot is the residual balancing layer. He says broader adoption would add to price discovery and give , forward curves and futures more information. The analysis uses offered prices, not realized rentals. Nebius supplied its quote data and did not review or approve the analysis.
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: Oct 21, 2026 (2026-10-21).
Date basis: Date stated on the publisher’s page.
Added to Escape Meridian: 2026-11-04T20:42:02.988Z. First observed: 2026-10-28T01:48:13.500Z. Retrieved: 2026-10-28T01:49:51.721Z.
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 28, 2026; originally published October 21, 2026. The comparison covers nine days and one dynamic seller; the source says it is not enough to measure volatility or how prices respond to the rest of the market. The figures are offered prices, not realized rentals. Nebius supplied the quote data, and the source says Nebius did not review or approve the analysis. The source notes that market coverage is smallest for B300, making its market statistics more sensitive to any one seller.
Open question
How would prices from more providers, observed over a longer period and compared with realized rentals, change the picture of spot-market conditions?
Source-stated
Steve Hou argues that dynamic prices can reveal supply and demand conditions that posted prices conceal.
A price that resets with capacity reveals conditions a posted price hides.
Source-stated
Silicon Data reports that more than half of the tracked spot price series did not change during the nine-day window.
More than half of the spot price series we track did not change once in nine days
Source-stated
Hou says Nebius’s H200, B200 and B300 prices moved on most days, while H100 stayed at its floor for most of the window.
Nebius's dynamic price changed on most days for H200, B200 and B300. Its H100 price sat at $0.79, the floor of Nebius's H100 spot range, for most of the window
Source-stated
Hou says the pattern is consistent with ample supply of older GPUs and strong demand for the newest, but the data cannot measure volatility or market response.
it is consistent with older GPUs being in ample supply and the newest in strong demand. Nine days from one dynamic seller is enough to describe price levels and how often prices change. It is not enough to measure volatility
Source-stated
Hou identifies costs for buyers and sellers: buyers give up budget certainty and sellers face less predictable revenue and added pricing infrastructure.
Buyers give up budget certainty: the hourly rate can change several times a day
Source-stated
Hou concludes that dynamic, transparent spot pricing is a key ingredient in compute price discovery, while most capacity remains on term contracts.
Dynamic, transparent spot pricing is a key ingredient in price discovery for compute, even as most capacity continues to be sold on term contracts
Source-stated
Silicon Data says Nebius supplied the quote data and did not review or approve the analysis.
Nebius provided the quote data used in this post. Silicon Data and Nebius are in discussions about broader data sharing. Nebius did not review or approve the analysis.
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.
NVIDIA H200
An NVIDIA GPU built on the Hopper architecture with 141 gigabytes of memory at 4.8 terabytes per second: nearly double the H100's memory capacity, with 1.4 times its memory bandwidth.
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.
Benchmark
A defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Editorial provenance
Version 1. Publication ID: fe01ef90-c8e1-04ef-b37f-477ee5a3cd31.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-04T20:38:59.494Z. 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: 42f764aab7e9d5c2d1355fc66696a6a0cc5081af45ad160a5fd8acf059e98f80. Source text hash: 3e0ace1068fa7599972fa1b23f0289b0620ed3e3e3860e80c4a0ef4a759a03a2.
Revision and source-check history
Version 1 · 2026-11-04T20:42:02.988Z
Steve Hou argues in Silicon Data that dynamic GPU spot prices reveal market conditions
Steve Hou published this analysis in Silicon Data on October 21, 2026. He argues that Nebius’s dynamic spot prices make changes in capacity and demand more visible than posted prices, which often stay fixed. Silicon Data reports that more than half of the spot price series it tracked did not change during the nine-day window. Nebius’s , B200 and B300 prices moved on most days, while its price stayed at its floor for most of the period. Hou says this pattern is consistent with older being in ample supply and the newest in strong demand, but nine days from one dynamic seller cannot measure volatility or show how prices respond to the wider market. Dynamic pricing also has costs: buyers may lose budget certainty as rates change and workloads stop when prices exceed a set maximum; sellers face less predictable revenue and added pricing infrastructure. Hou’s conclusion is that dynamic, transparent spot pricing is a key ingredient in price discovery for compute, even as most capacity remains on term contracts and spot is the residual balancing layer. He says broader adoption would add to price discovery and give , forward curves and futures more information. The analysis uses offered prices, not realized rentals. Nebius supplied its quote data and did not review or approve the analysis.
Original source
Silicon Data · Article
Source details
Original source date: Oct 21, 2026 (2026-10-21).
Date basis: Date stated on the publisher’s page.
Added to Escape Meridian: 2026-11-04T20:42:02.988Z. First observed: 2026-10-28T01:48:13.500Z. Retrieved: 2026-10-28T01:49:51.721Z.
Limits
Text as retrieved on October 28, 2026; originally published October 21, 2026. The comparison covers nine days and one dynamic seller; the source says it is not enough to measure volatility or how prices respond to the rest of the market. The figures are offered prices, not realized rentals. Nebius supplied the quote data, and the source says Nebius did not review or approve the analysis. The source notes that market coverage is smallest for B300, making its market statistics more sensitive to any one seller.
Open question
How would prices from more providers, observed over a longer period and compared with realized rentals, change the picture of spot-market conditions?
Source-stated
Steve Hou argues that dynamic prices can reveal supply and demand conditions that posted prices conceal.
A price that resets with capacity reveals conditions a posted price hides.
Source-stated
Silicon Data reports that more than half of the tracked spot price series did not change during the nine-day window.
More than half of the spot price series we track did not change once in nine days
Source-stated
Hou says Nebius’s H200, B200 and B300 prices moved on most days, while H100 stayed at its floor for most of the window.
Nebius's dynamic price changed on most days for H200, B200 and B300. Its H100 price sat at $0.79, the floor of Nebius's H100 spot range, for most of the window
Source-stated
Hou says the pattern is consistent with ample supply of older GPUs and strong demand for the newest, but the data cannot measure volatility or market response.
it is consistent with older GPUs being in ample supply and the newest in strong demand. Nine days from one dynamic seller is enough to describe price levels and how often prices change. It is not enough to measure volatility
Source-stated
Hou identifies costs for buyers and sellers: buyers give up budget certainty and sellers face less predictable revenue and added pricing infrastructure.
Buyers give up budget certainty: the hourly rate can change several times a day
Source-stated
Hou concludes that dynamic, transparent spot pricing is a key ingredient in compute price discovery, while most capacity remains on term contracts.
Dynamic, transparent spot pricing is a key ingredient in price discovery for compute, even as most capacity continues to be sold on term contracts
Source-stated
Silicon Data says Nebius supplied the quote data and did not review or approve the analysis.
Nebius provided the quote data used in this post. Silicon Data and Nebius are in discussions about broader data sharing. Nebius did not review or approve the analysis.
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.
NVIDIA H200
An NVIDIA GPU built on the Hopper architecture with 141 gigabytes of memory at 4.8 terabytes per second: nearly double the H100's memory capacity, with 1.4 times its memory bandwidth.
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.
Benchmark
A defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Editorial provenance
Version 1. Publication ID: fe01ef90-c8e1-04ef-b37f-477ee5a3cd31.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-04T20:38:59.494Z. 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: 42f764aab7e9d5c2d1355fc66696a6a0cc5081af45ad160a5fd8acf059e98f80. Source text hash: 3e0ace1068fa7599972fa1b23f0289b0620ed3e3e3860e80c4a0ef4a759a03a2.
uncertain · 2026-11-04T20:38:59.494Z · anthropic/claude-opus-5.5
Publication: fe01ef90-c8e1-04ef-b37f-477ee5a3cd31
attribution — pass: The reading correctly attributes the analysis to Steve Hou, Head of Research at Silicon Data. It frames the data findings as Silicon Data's and the disclosure as the source's statement.
Written by Steve Hou Head of Research at Silicon Data
summary — uncertain: The summary is faithful, but central terms are left unexplained for a general reader. It never says what spot or preemptible capacity is: discounted capacity the provider can interrupt. It does not identify Nebius as a neocloud GPU cloud provider. It does not explain price discovery, forward curves, futures, B200 or B300. The page tooltip for benchmark describes AI test suites, not the price benchmarks meant here. Smallest repair: add short plain-language glosses for spot, Nebius and price discovery, drawn from the source's own terms list, and suggest library entries for spot price, B200/B300 and price benchmark.
Spot (or preemptible): discounted capacity the provider can interrupt and reclaim at short notice.
date — pass: The October 21, 2026 publication date matches the source dateline at day precision, and the October 28 retrieval is disclosed. Data-window events are not presented with hindsight. The source is a publisher page, so no archive disclosure is needed.
# Industry Oct 21, 2026 9 Mins Read
support — pass: Each claim and summary sentence is supported by the source: the unchanged series, Nebius price movement, the H100 floor, the consistency inference with its stated limits, the trade-offs, the conclusion, offered-not-realized prices and the disclosure. The question follows from limitations the source itself states and adds no new claims.
Nine days from one dynamic seller is enough to describe price levels and how often prices change. It is not enough to measure volatility or how these prices respond to the rest of the market
media — pass: No media was selected and no verified media candidates were supplied, so there is nothing to misidentify.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
An NVIDIA GPU built on the Hopper architecture with 141 gigabytes of memory at 4.8 terabytes per second: nearly double the H100's memory capacity, with 1.4 times its memory bandwidth.
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 referenceA chip that performs many calculations at once; the main hardware used to train and run AI models.
Definition referenceAn NVIDIA GPU built on the Hopper architecture with 141 gigabytes of memory at 4.8 terabytes per second: nearly double the H100's memory capacity, with 1.4 times its memory bandwidth.
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 referenceA chip that performs many calculations at once; the main hardware used to train and run AI models.
Definition reference