Turn idle GPU capacity into inference revenue — without building an AI platform.
Across roughly 23,000 production Kubernetes clusters, average GPU utilisation runs at 5% (Cast AI, 2026). Only 7% of surveyed enterprises exceed 85% at peak (ClearML, 2024). Graphene sits on top of the GPUs you already own, fills idle hours with routed, metered, billable inference, and splits the upside with you. Your hardware stays yours. No exclusivity. No upfront fee.
From audit to renewal
Three steps take you from a no-cost utilisation audit to a renewed commercial agreement — see Docs for the detail behind each stage.
Utilisation audit
No cost, about 5 business days. We review fleet composition, telemetry surface and existing demand, then hand you a written read-out of your real utilisation rate and quantified pilot upside.Pilot LOI and telemetry
A 60-day pilot on 50-100 GPUs. We measure your baseline for 10-14 days, then expose capacity through the Graphene gateway and begin routing demand.Commercial proof and renewal
Between day 60 and 90 we prove the incremental revenue against your baseline and move to renewal — with 30-90 day off-ramp terms already in the agreement.
Why list on Graphene
The economics in plain terms — grounded in your own utilisation, not comparative headlines against unnamed competitors.
Fill idle hours with real demand
- Demand arrives as routed, metered, billable inference on the GPUs you already own — not a standalone GPU classifieds listing.
- Your existing raw-GPU customer relationships stay yours; we only route on top of the capacity you choose to expose.
More billable throughput on the same hardware
- Expect a projected 3-6x lift in billable token throughput on the same hardware (design-basis: margin routing, continuous batching, prefix caching, quantisation).
- Enterprise and sovereign demand stays agreement-backed; listing never bypasses procurement.
Transparent economics, no lock-in
- Revenue share starts at 85% at pilot volumes and never falls below 50%, on a published sliding scale.
- No exclusivity — run other buyers in parallel, and walk if the pilot lift does not materialise.
Transparent economics, not opaque auctions
We only earn on incremental tokens above your measured baseline, after a transparent cost-recovery floor. Your share follows a published sliding scale that starts at 85% at pilot volumes and never falls below 50%.
A live partner dashboard shows the same capacity, workload and revenue numbers we do — no black box, and no surprise reconciliation at the end of the month.
Payouts settle in fiat through our payments provider once your pilot is generating billable demand. Before then, start a conversation and we will size the opportunity with you.
Ready to size it up? Start a conversation through intake.
Frequently asked questions
Trust and data-handling basics live in Privacy; dedicated security collateral lands when the programme publishes it.
How does the engagement start?
With a no-cost utilisation audit (about 5 business days). We review fleet composition, telemetry surface and existing demand, and give you a written read-out of your real utilisation rate and quantified pilot upside. If we cannot lift you, we say so.
What does a pilot look like?
A 60-day pilot on 50-100 GPUs. We measure your baseline for 10-14 days first, then expose capacity through the Graphene gateway and begin routing demand. Off-ramp terms of 30-90 days are in the agreement.
How does the revenue share work?
We only earn on incremental tokens above your measured baseline, after a transparent cost-recovery floor. Your share starts at 85% at pilot volumes and never falls below 50%, on a published sliding scale that follows who is generating the demand.
Do I lose control of my hardware or customers?
No. The hardware stays yours, your existing raw-GPU customer relationships stay yours, and a live partner dashboard shows the same capacity, workload and revenue numbers we see.
Is there exclusivity or lock-in?
No. If a competitor offers a better split, run both. If the lift does not materialise in the pilot, you walk.
List your compute
High-intent providers start here—ops will route you into the right agreement and onboarding when supply listing opens for your region.
Set NEXT_PUBLIC_SUPPLY_INTAKE_URL or NEXT_PUBLIC_SUPPLY_INQUIRY_EMAIL for your deployment — see .env.example.