Roadmaps shift.
A training program is paused, re-scoped or moved to a different architecture. The reservation stays on the invoice for the rest of the term.
Capacity acquisition · GPU & AI
We buy the GPU and AI capacity you’re paying for but not using.
Reserved cloud GPUs that outlived their workload. Clusters that sit quiet between runs. AI commitments sized for a forecast that didn’t land. Tell us what you hold and we’ll make a written offer — for the capacity, not your accounts.
- NDA before details
- Written, non-binding offers
- Capacity, not accounts
- Scope
- Cloud GPU commitments, on-prem and colocation clusters, AI model capacity.
- Model
- We buy capacity. Your accounts and credentials stay yours.
- Terms
- Every structure is checked against the provider’s terms.
- Process
- NDA , then review , then written offer.
01Stranded spend
Capacity is bought against a forecast. Forecasts change.
When plans move, committed GPU and AI capacity keeps costing money whether it’s used or not. Most teams absorb it or let it lapse at the end of the term. Nivergrid gives the unused remainder a buyer.
Adoption lags the commitment.
AI model commitments are sized ahead of rollout, and real usage doesn’t always catch up before the term ends.
Clusters sit between runs.
On-prem and colocation GPUs bought for peak demand can spend long stretches well below it — still drawing on space, power and budget.
The idle months behind you are sunk. The ones ahead can still have a buyer.
02What we buy
Capacity that’s committed, provisioned or racked — and idle.
If you pay for it whether it runs or not, it’s worth a conversation. Tell us what you hold and we’ll tell you plainly whether we can make an offer.
Cloud
Cloud GPU commitments and reserved capacity
Reserved instances, capacity reservations and committed-use agreements on AWS, Microsoft Azure, Google Cloud or another cloud that your workloads no longer fill — after a migration, a cancelled project or a forecast that ran high.
On-prem & colo
On-prem and colocation GPU clusters
Owned or leased GPU servers in your own data center or a colocation facility that sit idle between training runs, or were sized for a roadmap that changed. Access, scheduling and isolation are agreed with your team case by case.
AI models
AI model capacity and commitments
Unused committed spend, provisioned throughput or prepaid capacity with AI model providers such as OpenAI, Anthropic or Google (Gemini). Some agreements restrict sharing or reassignment, so we review the terms with you before making any offer.
Other
Something else
Specialist GPU clouds, accelerator contracts, mixed hardware generations or partial terms that don’t fit a neat category. If it’s capacity you’re paying for and not using, describe it and we’ll tell you whether we can work with it.
03How it works
A process your finance and legal teams will recognize.
Four steps, documented at each stage. There’s no platform to onboard to and nothing to install, and you control what’s shared and when.
-
Confidential introduction
Share the basics through the form or by email: provider, capacity type, rough size and term remaining. No contracts or usage data yet — a mutual NDA comes first if you want one.
-
Review under NDA
We go through the commitment with you — agreement type, remaining term, utilization and the provider terms that apply — to establish what can be acquired, and how.
-
Written offer, or a plain no
If there’s a fit, you receive a written offer setting out structure, price and timing that finance, legal and procurement can review together. If there isn’t, we say so.
-
Agreement and settlement
If you accept, we sign a definitive agreement, complete any steps the provider’s terms require, and settle on the schedule the agreement sets out.
What we’ll typically ask for during review
- Provider and agreement type
- Capacity details: GPU model or instance type, region or location, quantity
- Term start and end dates
- Recent utilization
- The relevant terms of your agreement, under NDA
What we’ll never ask for
- Passwords, API keys or login credentials
- Control or transfer of your provider accounts
- Confidential contract terms before an NDA is in place
04Why Nivergrid
Built for the people who sign off on it.
Selling unused capacity is a finance, legal and infrastructure decision at once. The process is designed so each of them can review it on their own terms.
Recover stranded spend
Turn capacity you’ve already committed to into recovered value, instead of carrying it to the end of the term.
Confidential by default
A mutual NDA is available before contracts or usage data change hands. We use what you share only to evaluate and complete a transaction.
Inside the provider’s terms
Whether and how capacity can be sold depends on the provider and the contract. We check before we offer, and tell you early if something can’t be done.
Everything in writing
Offers set out structure, price and timing in one document, ready for finance, legal and procurement to review together.
Speaks infrastructure
We talk in instance types, regions, term dates and throughput units, so the first conversation with your platform team is a useful one.
No obligation
Requesting an offer commits you to nothing. Nothing is binding until both sides sign a written agreement.
05Who it’s for
For whoever owns the underused line item.
You’re likely a fit if you have:
- A GPU reservation or commitment with term left that you won’t fully use
- GPU servers or clusters that sit idle for long stretches
- An AI model commitment that’s running ahead of your usage
-
Finance & FP&A
Commitments made against a forecast that changed.
CFOs and controllers looking at spend that won’t be consumed before the term ends — and want an option they can explain.
-
Infrastructure & ML platform
You can see which reservations sit idle.
Heads of infra and platform with reserved capacity or clusters running well below plan. Selling the unused part can be cleaner than finding work to fill it.
-
Procurement & vendor management
A documented route for unused capacity.
Teams managing cloud and AI agreements who want a structured, reviewable option instead of letting capacity lapse.
-
Founders & operators
Capacity bought ahead of a roadmap that moved.
Turn compute you locked in back into runway, without spending engineering time finding work to fill it.
Looking for GPU or AI capacity rather than selling it? Our focus is buying, but tell us what you need. Email us
Every idle month is term you can’t get back.
A commitment’s remaining value shrinks as its term runs down. The sooner we review it, the more there is to work with.
Do you take over our cloud or AI provider accounts?
No. We buy capacity, not accounts. You keep your accounts, credentials and provider relationship. How a transaction is structured depends on the capacity type and the provider’s terms, and you’ll see the proposed structure in writing before you agree to anything.
Can every commitment be sold?
No. Some agreements allow capacity to be resold, reassigned or used on another party’s behalf, and others restrict it. Part of our review is reading the relevant terms with you. If we can’t find a structure that works within them, we’ll say so.
Which providers’ capacity do you look at?
Committed or reserved GPU capacity on clouds such as AWS, Microsoft Azure and Google Cloud; unused capacity or commitments with AI model providers such as OpenAI, Anthropic and Google (Gemini); and on-prem or colocation GPU hardware. If yours doesn’t fit any of these, describe it anyway. Nivergrid is independent and isn’t affiliated with or endorsed by any of these providers.
How do you handle confidentiality?
We’ll sign a mutual NDA before you share contracts, invoices or usage data. The form on this page asks only for high-level information, and you’re welcome to ask for an NDA before telling us anything else. We use what you share only to evaluate and, if you go ahead, complete a transaction. See our privacy notice.
What do you need from us to make an offer?
To start: the provider, the type of capacity, its approximate size or value, and how much of the term remains. For GPUs, the GPU model or instance type and region help; for AI models, the unit of the commitment, such as throughput units or prepaid spend. During review we’ll typically ask for recent utilization and the relevant agreement terms. We never ask for passwords, API keys or account credentials.
How are offers priced?
Case by case. An offer depends on the type of capacity, how much term remains, where it runs, the provider’s terms and demand for that capacity. We’ll explain what an offer is based on.
How long does it take?
It depends mostly on how quickly details can be shared and on any steps the provider’s terms require. Once we’ve seen the details we’ll give you a realistic timeline, and your written offer includes it.
Can we sell part of a commitment?
Possibly. Tell us how much you expect not to use, and we’ll look at whether a partial structure works within the provider’s terms.
Are we obligated to accept an offer?
No. Requesting and reviewing an offer doesn’t commit you to anything. Nothing is binding until both sides sign a written agreement.
07Get an offer
Start with a confidential conversation.
Tell us what you’re holding. We’ll review it with you and come back with a clear answer: a written offer, or a plain no.
What happens next
- We read what you send and reply by email.
- If it looks like a fit, we put a mutual NDA in place and review the details with you.
- You get a written offer, or a straight answer that it isn’t one we can buy. You decide.
Prefer email? Write to hello@nivergrid.com.
Thank you. We’ve received your details.
We’ll reply by email to arrange next steps.
We couldn’t send that automatically.
Please email your details to hello@nivergrid.com and we’ll pick it up from there.
Sell your capacity
A rough picture is enough to start. Please don’t include confidential contract terms — we’ll exchange those under NDA.
Fields marked * are required.
We couldn’t send that automatically.
Nothing is lost. Your answers are ready in an email — use the button below to send them to from your own mail app.