Reserved compute deals put large, long commitments on both sides, often before the capacity exists. We draft and negotiate compute MSAs, reservation orders and cloud agreements for GPU cloud providers and for the AI companies buying from them. The focus is on what is committed, what happens when capacity is late or down, and how data and workloads are protected.
At a glance
- Practice
- AI Infrastructure
- Pricing
- Flat fee, quoted before we start
- Included
- 6 deliverables
- Process
- 3 steps, one named attorney
- Where we work
- Slack, email or a call
When founders call us
The moments that usually bring a company to us for this work.
- 01
You are a GPU cloud and need a compute MSA and order form you can use with every customer.
- 02
Your AI startup is about to sign a multi-month GPU reservation with a take-or-pay commitment.
- 03
A provider's capacity start date slipped and you want to know your options.
- 04
An enterprise customer wants to negotiate your cloud terms, including security and data handling.
What you get
Scoped in writing before we start. One flat fee for all of it.
- Compute MSA and reservation order forms for providers, or review of provider paper for buyers
- Capacity commitments, delivery dates, substitution rights and remedies for late capacity
- Take-or-pay, prepayment, ramp and early termination terms
- Availability SLAs, maintenance windows and service credits
- Customer data, model weights and workload confidentiality terms, with a DPA where personal data is involved
- Acceptable use, suspension rights and export control provisions
How it works
- 1
Clarify the commitment
We confirm what capacity, hardware type and term each side is committing to.
- 2
Draft or redline
We prepare provider templates or negotiate the provider's paper for buyers.
- 3
Stress test the exits
We check what happens if capacity is late, down or no longer needed.
Mistakes we help you avoid
The ones we see most often, and catch before they cost you.
- Avoided
Committing to pay for capacity from a fixed date with no adjustment if the provider delivers late.
- Avoided
Substitution clauses that let the provider swap in a different GPU generation or location without your consent.
- Avoided
Broad suspension rights that let the provider cut off a running training job on short notice.
- Avoided
Silence on who can access customer model weights and training data on the provider's infrastructure.
Compute Capacity & Cloud: questions founders ask
It is a commitment by a buyer to pay for a defined amount of GPU capacity over a set term, often with prepayment or take-or-pay terms. In return the provider commits to deliver and maintain that capacity. The details on delivery, substitution and remedies vary widely.
It means the buyer pays for the committed capacity whether or not it uses it. That gives the provider revenue certainty to finance hardware. Buyers should check what happens if the capacity is late, unavailable or does not meet spec.
If the provider processes personal data on your behalf, such as in training data or inference inputs, a DPA is generally needed under laws like GDPR and many US state privacy laws. Whether one is required depends on the data you run through the service.
Only if the contract allows it. Some agreements permit termination with a fee, resale of unused capacity or conversion to other services. We negotiate these options upfront, because they are hard to win once the commitment is signed.
Get a flat-fee quote for Compute Capacity & Cloud.
A short call with an attorney. You leave with a written scope and one fixed price.
Book a call