Skip to content

$COMPUTE IS COMING TO ROBINHOOD CHAIN

01Compute market

THE COMPUTEMARKET.

Capacity for intelligence is bought and sold by the hour. This is the hardware that sets the price.
Reference dataDemo

Unit · USD per GPU-hour. Five accelerators. Updated by the configured provider.

02 · The compute market

DEMO DATA · ILLUSTRATIVE
  • H100 SXM$2.107/GPU·HR−3.13%DEMO
  • H200$3.178/GPU·HR+5.48%DEMO
  • B200$4.951/GPU·HR−4.05%DEMO
  • A100 SXM4$1.205/GPU·HR−1.71%DEMO
  • RTX 5090$0.626/GPU·HR−2.80%DEMO

02Compute board

DEMO DATA · ILLUSTRATIVE

THE BOARD.

Accelerator capacity is rented by the hour and priced like a commodity: by generation, by memory, by availability. The board tracks the reference rate of the hardware that runs modern AI.

01H100 SXM

DATACENTER · Hopper

SXM5 · 80 GB HBM3

$2.107
24H −3.13%
Demo
02H200

DATACENTER · Hopper

SXM5 · 141 GB HBM3e

$3.178
24H +5.48%
Demo
03B200

DATACENTER · Blackwell

SXM6 · HBM3e

$4.951
24H −4.05%
Demo
04A100 SXM4

DATACENTER · Ampere

SXM4 · 80 GB HBM2e

$1.205
24H −1.71%
Demo
05RTX 5090

CONSUMER · Blackwell

PCIe · 32 GB GDDR7

$0.626
24H −2.80%
Demo

Reference = USD per GPU-hour · Source: Demo data · 2026-09-30 00:27 UTC

Demo data. Illustrative values for interface preview only — not market prices.

03How compute is priced

PRICED LIKEA COMMODITY.

The unit

The GPU-hour.

Accelerator capacity is rented in hours of one GPU. The rate reflects the chip, its memory, the interconnect around it and the data center that powers and cools it.

The drivers

Generation, memory, availability.

Newer architectures command a premium until supply catches up. High-bandwidth memory capacity decides which models fit. Contract length and location move the price.

The market

Spot, reserved, forward.

Capacity trades on demand, on multi-year reservations and increasingly on forward terms — the vocabulary of a commodity market, applied to computation.

04Hardware reference

Schematics · illustrative, not to scale

FIVE ACCELERATORS.ONE MARKET.

01

H100 SXM

Segment
DATACENTER
Architecture
Hopper
Form factor
SXM5
Memory
80 GB HBM3

02

H200

Segment
DATACENTER
Architecture
Hopper
Form factor
SXM5
Memory
141 GB HBM3e

03

B200

Segment
DATACENTER
Architecture
Blackwell
Form factor
SXM6
Memory
HBM3e

04

A100 SXM4

Segment
DATACENTER
Architecture
Ampere
Form factor
SXM4
Memory
80 GB HBM2e

05

RTX 5090

Segment
CONSUMER
Architecture
Blackwell
Form factor
PCIe
Memory
32 GB GDDR7

05Methodology & sources

WHAT THESENUMBERS ARE.

DEMO DATA · ILLUSTRATIVE

The board is currently in demo mode. Values are generated by an illustrative model so the interface can be reviewed. They are not quotes, not market prices, and not derived from any index or vendor.

Reference = the rental rate for one GPU for one hour, in US dollars. Change = relative move over 24 hours. Proprietary compute-index data is never scraped or embedded without a written licence.

$COMPUTE does not track, peg to, or settle against any of these references. $COMPUTE is an experimental crypto asset. It does not represent ownership of, entitlement to, or redemption for physical compute capacity and does not track any external compute benchmark.