NeoCloud & Exchange · Spot · Forward · Futures
AI Compute Exchange
One venue for GPU compute. Data centers sell it. AI labs buy it. Traders price it.
Bids
Asks
Built to clear capacity across the world’s clouds & neoclouds
The stack
Megawatts at one end, tokens at the other.
Capacity is contracted and metered as megawatts, sold on as GPU-hours, served as tokens, and quoted on a book that prices the two units beneath it. Enter at whichever unit your work is already denominated in; what you stop needing goes back on the same book.
Exchange
A central limit order book over the two units below it, cleared and settled in dollars.
- Who
- Traders
- Why
- To hedge and arbitrage
- What
- Spot · forwards · futures
- TradersBuy / Sell
- AI companiesBuy side
- AI companiesBuy side
- Data centersSell side
GPU owners · Operators
Own the GPUs. We do everything else.
Investors and data center financiers own the hardware; Exascale sources it, finds or builds the data center, installs and operates the fleet in Texas and contracts it to offtakers for one to five years, for a share of the revenue. Already run a fleet? Connect it and list its idle hours.
For investors and data center financiers
Own GPUs, operated by Exascale
- Dallas, Texas
- North Texas
- Houston, Texas
How many GPUs do you want to deploy?
| Measure | 1Y | 2Y | 3Y | 4Y | 5Y |
|---|---|---|---|---|---|
| ROI | 35% | 83% | 131% | 173% | 202% |
| IRR | 39% | 47% | 49% | 50% | 48% |
$19.6M all-in · $9.4M/yr net to you · 48% cash yield a year · payback 2.4 yrs
Illustrative only: a hypothetical projection from the assumptions on the intake page, before income tax and financing, at B300 SXM contract rates of $7.27 per GPU-hour from the Exascale GPU Index, B300 global index, 24 Oct 2026, assuming the GPUs are resold at 100% of their cost. GPU-hour rates, hardware prices and resale values are volatile; actual results may differ materially and you could lose money. Not an offer of securities or a promise of returns.
End to end, for a share of the revenue
- Source GPUs
- Source the data center
- Build the data center
- Inspect colocation
- Inspect GPUs
- Test GPUs
- Transit insurance
- Delivery
- Burn-in
- Installation
- Networking
- Connect to the cloud
- Kubernetes and Slurm
- Offtake contracts
Already operating
Connect your fleet
Connect once and idle hours become listable capacity, attested at connection and continuously after it, metered per second and settled to you on the venue’s cycle.
- Onboards
- Data centers, neoclouds, enterprise fleets
- Attestation
- At connection, then continuously
- Metering
- Per node, per second
Model library · Token factory · Inference · Router
Open models, one endpoint.
513 catalog entries, 52 with verified sources, across text, code, reasoning, vision, image, video and speech, OpenAI-compatible and priced per million tokens.
One base URL.
Point the OpenAI SDK at Exascale. Nothing else in your code changes.
from openai import OpenAI client = OpenAI( base_url="https://api.exascale.dev/v1", api_key=os.environ["EXASCALE_API_KEY"], ) stream = client.chat.completions.create( model="deepseek-v3", messages=[{"role": "user", "content": prompt}], stream=True, )
Available now, and next
Capacity in every class, from Hopper to Vera Rubin.
Rack-scale NVL72 systems, HGX and AMD Instinct baseboards, under contract across operators and regions. Available by the GPU-hour or reserved for up to five years, whichever generation the work is written for. Vera Rubin can be reserved ahead.
Preview · H2 2026NVIDIAVera Rubin NVL144Pricing at launch · joins the Exascale GPU Index on its first public priceReserve ahead
NVIDIAGB300 NVL72View in GPU Cloud
NVIDIAGB200 NVL72View in GPU Cloud
NVIDIAHGX B300View in GPU Cloud
NVIDIAHGX B200View in GPU Cloud
NVIDIAHGX H200View in GPU Cloud
NVIDIAHGX H100View in GPU Cloud
AMDInstinct MI355XView in GPU CloudExchange · Trading terminal
GPU compute, on a trading screen.
Every GPU class and region quotes on one terminal: a candlestick chart, the order book and time and sales beside it, and order entry on the right. Price an hour of H100 the way a desk prices anything else.
- Chart
- Candlesticks from 1m to 1Y, EMA and volume
- Book
- Depth of market, trades and spread in basis points
- Ticket
- Market, limit and pro orders, cross or isolated margin
Request access
Whichever side you are on.
Onboarding for data-center operators, AI enterprises, frontier labs and trading desks, in rolling cohorts.