Skip to content
NeoShunt
Create an organization

For capacity providers

List your GPUs where the buying decision is actually made.

Colocation and MSP operators have capacity and no distribution. NeoShunt has customers running measured bake-offs and asking which platform to standardise on. Certify a node, get listed under Private Cloud, and be paid by revenue share computed from the same ledger that bills the customer.

Start certificationRead the onboarding doc
Private Cloudcertified

RTX PRO 6000 · Llama 3.1 70B FP8

receipt · 2026-08-27

Published certification receipt for one node
MeasuredResult
Throughput @ N=8150 tok/s
Honest context ceiling49,152
Limits waivednone
Front-needle at depthexact @ 71,751

Every certification is published as a receipt, including the failures. A node that near-misses a needle test says so.

How onboarding works

  1. 01

    Certify the node

    We run the certification script against your hardware and publish the receipt: throughput at stated batch sizes, the honest context ceiling, and whether any limit was waived.

  2. 02

    Listed under Private Cloud

    Certified partner capacity is named, single-operator hardware — it sits alongside the GPUs we own, not in a spot market.

  3. 03

    Ranked on measurement

    Your node competes in customer bake-offs on the same rubric as everything else. Customers see which provider served each answer.

  4. 04

    Paid from the same ledger

    Revenue-share statements are computed from the identical ledger that bills the customer. Same numbers, both directions.

Recertification is enforced, and the policy is public

The rules you are held to live in the provider onboarding document, not in a private agreement — because you have to be able to read them. Capacity that drifts from its certified numbers is re-measured, and the receipt is republished.

Revenue share, from one ledger

Statements are generated from the same credits ledger that charges the customer — not from a separate provider-side count that can quietly disagree with it. Split percentages are an operator setting entered per agreement; nothing is offered until it has been priced.

Customers see the provider that served each answer, because the same model served at different numeric precisions by different providers is not the same product.

What we need from a node

  • Named hardware and a single operator — no resold spot capacity
  • An OpenAI-compatible endpoint we can join over a tagged, expiring key
  • A stated context ceiling you will stand behind under load
  • Willingness to have the numbers published, including the ones that miss

Have capacity sitting idle? Certify one node and find out what it earns.

Start certification