# H100 forward curve

> The Silicon Data H100 forward curve: the market’s expected path for the H100 rental index from spot out to three years, anchored on the settlement index, shown for the latest seven days.

Updated 2026-09-06. Canonical: https://gpuquant.com/docs/market-data/h100-forward-curve

## What it measures

What the market expects the Silicon Data H100 rental index to be at every point from today out to three years, as Silicon Data publish it: 145 points from tenor zero to tenor 36 in quarter-month steps, in US dollars per GPU-hour. Tenor zero is the same instrument as that day’s neo-cloud index to two more decimal places, so the curve is the market’s expected path for the settlement reference of the CME H100 contract, not a separate quantity.

Each day publishes a whole curve, so the interesting reading is across days: fix a month and watch how the market’s view of it was revised. That is the direct analogue of a futures contract’s price history, and it is what the compare control on the chart is for.

The curve has been wrong in a telling direction. Seen in January 2025, it priced an H100-hour for October 2026 at $1.17; seen in late August 2026, two months out, it priced the same month at $2.50, a revision of more than 110 percent. The steep decay it expected did not arrive.

Silicon Data’s history of the curve starts on 2 January 2025 and the early curves are shorter, six months at quarter steps, so a curve’s length is a fact about the day it was published.

## How it is built

By Silicon Data, who publish a curve every day: the expected value of the H100 rental index at each tenor from zero to 36 months in quarter-month steps. Two rates are published for each tenor, the term rate, which is the average expected rate out to that tenor, and the forward rate, which is the implied rate for that single period; they coincide at tenor zero and diverge after it, and the chart draws the forward rate.

A tenor is relative, so tenor six today and tenor six tomorrow are different calendar months. GPUQuant stores every point against the month it lands in as well as its tenor, which is what lets one future month be followed across the days the market revised its view of it. Where the curve comes from and how it anchors on the index is on [Silicon Data indices](https://gpuquant.com/docs/silicon-data#forward-curve).

## Publication

**Source**

Silicon Data

**Unit**

USD / GPU-hour

**Frequency**

Daily, a whole curve each day, with the index

## What GPUQuant shows

The newest **7 days**, and nothing older, under a written display licence from Silicon Data that covers display on this site and excludes redistribution and the accumulation of history. The figures are read from Silicon Data's public chart, which serves exactly that window, and the explorer opens the row on its seven-day view. The day picker on the chart compares any two days inside the window; a day outside it is answered exactly as a day never captured.

On [Market Data](https://gpuquant.com/market-data) as Forward curve · Silicon Data under H100.

## Reading it beside the others

-   [H100 neocloud index](https://gpuquant.com/docs/market-data/h100-neocloud-index): The list price of the same hour on the specialist clouds, roughly a third of the hyperscaler figure.
-   [Silicon Data H100 index](https://gpuquant.com/docs/market-data/silicon-data-h100-index): Silicon Data's modelled figure, which counts negotiated and committed-use deals and sits below any list price.
-   [H100 residual value](https://gpuquant.com/docs/market-data/h100-residual-value): What one H100 is still worth, from that curve.
-   [Silicon Data H100 Rental Index Futures](https://gpuquant.com/docs/instruments/h100-futures): The contract this curve is the expectation of.

## Sources

-   [Silicon Data GPU Forward Curve](https://www.silicondata.com/products/gpu-forward-curve), the publisher's page.
-   [Silicon Data indices](https://gpuquant.com/docs/silicon-data#forward-curve), how the curve is read and how it anchors on the index.