Silicon Data · Apr 27, 2026
Carmen Li argues in Silicon Data that H100 Hyperscaler pricing is stable between events
Markdown · JSON · Original source
On April 27, 2026, Carmen Li argues in Silicon Data that Hyperscaler on-demand rates stayed within a narrow range from March 1 to April 20, while remaining vulnerable to event-driven jumps. She attributes the quiet to contracted supply, switching-cost-constrained demand and the chip’s mature commercial lifecycle. The contrast with more volatile B200 and H100 pricing, she argues, marks a distinct market tier, not evidence that all rental prices behave alike. Li says the index can help enterprise teams reservation pricing and credit teams stress-test existing models. But it is a price ceiling above reserved capacity, not a cleared spot price, and the observed period is too short to establish the longer history needed for long-term credit decisions or cleared derivatives. A March 23 jump during GTC shows why products using the index need to account for event risk. Li concludes that the index offers institutional readers a useful second reference point, while its track record is still being built. How long the index must be observed, and through what market conditions, before it could support long-dated products remains unknown.
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: Apr 27, 2026 (2026-04-27).
Date basis: Date stated on the publisher’s page.
Added to Escape Meridian: 2026-11-07T05:43:23.663Z. First observed: 2026-10-25T02:29:43.001Z. Retrieved: 2026-10-25T02:31:37.157Z.
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 25, 2026; originally published April 27, 2026.
Open question
How long, and through what market conditions, would the index need to be observed before it could support long-dated products?
Source-stated
Carmen Li reports that H100 Hyperscaler on-demand rates stayed within a narrow range over the period she examined, with very low measured variation.
Between March 1 and April 20, 2026, Silicon Data's H100 Hyperscaler On-Demand Index traded in a band of $7.40 to $7.52. The index was unchanged on 39 of 50 daily transitions.
Source-stated
Li attributes the index’s relative stability to contracted supply, switching-cost-constrained demand and the chip’s mature commercial lifecycle.
Put together: a mature chip, contracted supply, workload-locked demand.
Source-stated
Li says the index is a price ceiling rather than a cleared spot-market price, and cautions that the observed period is too short to establish a long-term volatility regime.
Fifty trading days is too short a window to settle that question — credit models span 12 to 24 months, and seven weeks of low realized volatility does not establish a regime that long.
Source-stated
Li identifies reservation-pricing benchmarks and stress-testing existing credit models as uses the data may support, while cautioning against treating it as a standalone basis for long-term underwriting.
Credit teams can layer observed H100 Hyperscaler volatility on top of their existing base-case revenue assumptions to tighten or widen stress bands — not to replace the base case.
Source-stated
Li argues that event-driven price shocks remain a risk despite the index’s low baseline volatility.
The correct inference is that this tier is stable between events and shock-sensitive at them — narrower than "behaves like a money-market rate."
Source-stated
Li concludes that the index offers a distinct reference point for institutional readers, while its track record is still accumulating.
The Hyperscaler index adds a second reference point — one with a different volatility profile, sitting at a different tier of the market, covering a different population of counterparties — and with its own specific event-risk signature that product design has to account for.
Source-stated
The source was written by Carmen Li and published on April 27, 2026.
Written by Carmen Li Founder at Silicon Data # Industry Apr 27, 2026 10 Mins Read
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.
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.
Neocloud
A new breed of cloud compute provider focused on renting out GPU computing power: a pure-play GPU cloud whose economics are still evolving.
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
NVIDIA develops processors, networking and software used for AI, graphics and scientific computing.
Editorial provenance
Version 1. Publication ID: 5d4f4edd-0b3a-9711-0dd6-f234f4e3d060.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-07T05:38:39.631Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_admitted_by_model.
Observation: d00ac48abde8897e1be4d8eccd0f8a28c39ab1272f4cf601645a22b77418dfda. Source text hash: d24877c881d89a5d2344e19f34718e3a710b579224882a3e305efa0e217bbfbd.
Revision and source-check history
Version 1 · 2026-11-07T05:43:23.663Z
Carmen Li argues in Silicon Data that H100 Hyperscaler pricing is stable between events
On April 27, 2026, Carmen Li argues in Silicon Data that Hyperscaler on-demand rates stayed within a narrow range from March 1 to April 20, while remaining vulnerable to event-driven jumps. She attributes the quiet to contracted supply, switching-cost-constrained demand and the chip’s mature commercial lifecycle. The contrast with more volatile B200 and H100 pricing, she argues, marks a distinct market tier, not evidence that all rental prices behave alike. Li says the index can help enterprise teams reservation pricing and credit teams stress-test existing models. But it is a price ceiling above reserved capacity, not a cleared spot price, and the observed period is too short to establish the longer history needed for long-term credit decisions or cleared derivatives. A March 23 jump during GTC shows why products using the index need to account for event risk. Li concludes that the index offers institutional readers a useful second reference point, while its track record is still being built. How long the index must be observed, and through what market conditions, before it could support long-dated products remains unknown.
Original source
Silicon Data · Article
Source details
Original source date: Apr 27, 2026 (2026-04-27).
Date basis: Date stated on the publisher’s page.
Added to Escape Meridian: 2026-11-07T05:43:23.663Z. First observed: 2026-10-25T02:29:43.001Z. Retrieved: 2026-10-25T02:31:37.157Z.
Limits
Text as retrieved on October 25, 2026; originally published April 27, 2026.
Open question
How long, and through what market conditions, would the index need to be observed before it could support long-dated products?
Source-stated
Carmen Li reports that H100 Hyperscaler on-demand rates stayed within a narrow range over the period she examined, with very low measured variation.
Between March 1 and April 20, 2026, Silicon Data's H100 Hyperscaler On-Demand Index traded in a band of $7.40 to $7.52. The index was unchanged on 39 of 50 daily transitions.
Source-stated
Li attributes the index’s relative stability to contracted supply, switching-cost-constrained demand and the chip’s mature commercial lifecycle.
Put together: a mature chip, contracted supply, workload-locked demand.
Source-stated
Li says the index is a price ceiling rather than a cleared spot-market price, and cautions that the observed period is too short to establish a long-term volatility regime.
Fifty trading days is too short a window to settle that question — credit models span 12 to 24 months, and seven weeks of low realized volatility does not establish a regime that long.
Source-stated
Li identifies reservation-pricing benchmarks and stress-testing existing credit models as uses the data may support, while cautioning against treating it as a standalone basis for long-term underwriting.
Credit teams can layer observed H100 Hyperscaler volatility on top of their existing base-case revenue assumptions to tighten or widen stress bands — not to replace the base case.
Source-stated
Li argues that event-driven price shocks remain a risk despite the index’s low baseline volatility.
The correct inference is that this tier is stable between events and shock-sensitive at them — narrower than "behaves like a money-market rate."
Source-stated
Li concludes that the index offers a distinct reference point for institutional readers, while its track record is still accumulating.
The Hyperscaler index adds a second reference point — one with a different volatility profile, sitting at a different tier of the market, covering a different population of counterparties — and with its own specific event-risk signature that product design has to account for.
Source-stated
The source was written by Carmen Li and published on April 27, 2026.
Written by Carmen Li Founder at Silicon Data # Industry Apr 27, 2026 10 Mins Read
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.
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.
Neocloud
A new breed of cloud compute provider focused on renting out GPU computing power: a pure-play GPU cloud whose economics are still evolving.
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
NVIDIA develops processors, networking and software used for AI, graphics and scientific computing.
Editorial provenance
Version 1. Publication ID: 5d4f4edd-0b3a-9711-0dd6-f234f4e3d060.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-07T05:38:39.631Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_admitted_by_model.
Observation: d00ac48abde8897e1be4d8eccd0f8a28c39ab1272f4cf601645a22b77418dfda. Source text hash: d24877c881d89a5d2344e19f34718e3a710b579224882a3e305efa0e217bbfbd.
uncertain · 2026-11-07T05:38:39.631Z · anthropic/claude-opus-5.5
Publication: 5d4f4edd-0b3a-9711-0dd6-f234f4e3d060
attribution — pass: The reading attributes the analysis to Carmen Li, Founder at Silicon Data, writing on Silicon Data's blog. This matches the byline. The page is a single argued piece, not a listing.
Written by Carmen Li Founder at Silicon Data
summary — uncertain: The summary faithfully follows Li's argument and caveats. Several central terms are neither explained in the text nor covered by the supplied definitions: 'Hyperscaler' (the big cloud providers' enterprise tier), 'B200', 'cleared spot price', 'cleared derivatives' and 'NVIDIA GTC'. The reading also never says plainly that the index is Silicon Data's daily benchmark of hourly H100 rental rates. A general reader may not grasp what the index measures or what the tier distinction means. Smallest repair: add one plain-language clause each for the index, Hyperscaler and cleared spot price.
It is Silicon Data's daily benchmark tracking on-demand H100 rental rates on the enterprise-SLA hyperscaler tier.
date — pass: April 27, 2026 matches the source dateline and originalPublished at day precision. The kind is publisher-page, so no archive disclosure is needed. The retrieval date is noted. There is no hindsight framing, and the reading does not present the forecasted regime-breaking events as outcomes.
# Industry Apr 27, 2026 10 Mins Read
support — pass: Each claim is source-stated and backed by its excerpt: the narrow band, the structural causes, the ceiling-versus-spot distinction, the short-window caveat, the supported uses, the event risk and the second-reference-point conclusion. The closing unknown about required track record follows from Li's own statement that the record is still accumulating.
That track record is still being built. The seven weeks of data covered here is a useful chapter, not the whole book.
media — pass: No media was selected, and no verified media candidates were supplied. This is appropriate.
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 graphics processing unit (GPU) built on the company's Hopper architecture and designed to accelerate AI workloads, including large language models.
Definition referenceA new breed of cloud compute provider focused on renting out GPU computing power: a pure-play GPU cloud whose economics are still evolving.
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 referenceNVIDIA develops processors, networking and software used for AI, graphics and scientific computing.
Explore NVIDIAAn NVIDIA graphics processing unit (GPU) built on the company's Hopper architecture and designed to accelerate AI workloads, including large language models.
Definition referenceA new breed of cloud compute provider focused on renting out GPU computing power: a pure-play GPU cloud whose economics are still evolving.
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 referenceNVIDIA develops processors, networking and software used for AI, graphics and scientific computing.
Explore NVIDIA