RRB Partners Compute · GPU capacity for AI inference
Multi-year B300 capacity, below market.
RRB Partners places AI companies on dedicated NVIDIA B300 GPUs in purpose-built inference modules, through our data center partners. You get firm delivery dates, pricing below listed long-term rates, and contracts that protect you.
Two sides that need each other
AI companies
Serving models in production means GPUs running around the clock for years. On-demand pricing is expensive, and the newest GPUs are hard to get.
Data centers
Lenders finance GPU modules against signed, long-term customer contracts. Without committed customers, new capacity doesn't get built.
RRB Partners
We bring the two together: AI companies get committed capacity at better prices and with protective terms; data centers get the long-term customers that let them build.
Committed capacity costs about half of pay-as-you-go
NVIDIA B300, listed price per GPU-hour (US$)
Listed prices across 35 providers, GetDeploying, October 2026. Your price depends on size, term and start date. We'll give you a range after an intro call.
What you get
Below listed market rates
Multi-year capacity we place is typically priced below the cheapest listed long-term reservations for the same GPUs.
Firm delivery dates
B300s are scarce. We can see our partners' upcoming capacity, so you reserve a dedicated cluster with a start date, not a waitlist.
Built for inference
Purpose-built modules of 1,000+ NVIDIA B300 GPUs with redundant power, designed for serving models in production around the clock.
Contracts that protect you
Uptime credits, ramp-up schedules instead of paying for everything on day one, hardware refresh rights and clear exit terms.
Help getting approved
Long-term capacity needs strong credit. We help structure deposits, letters of credit or credit insurance, so you need less cash up front.
Direct contracts
You sign directly with the data center, and we stay on hand for expansions and renewals.
The capacity
Our partners build modular data centers designed for one job: running AI models in production. Each module is a dedicated, self-contained cluster, so your capacity isn't shared with someone else's training run.
- GPUs
- NVIDIA B300 (Blackwell Ultra)
- Memory
- 288 GB per GPU
- Module size
- 1,000+ GPUs
- Contract sizes
- From 128 GPUs to multiple modules
- Power
- Redundant, for round-the-clock inference
- Terms
- 2 to 5 years
- Locations
- United States; Canada being evaluated
Who it's for
Our partners finance each module against long-term contracts, so the best fit is a company with steady, growing inference demand.
- You serve AI models in production: assistants, agents, voice, video, search, coding tools or model APIs
- You need at least 128 GPUs, ideally 512 or more
- You can commit for 24 months or longer
- You want capacity in the next 3 to 12 months
- US-hosted capacity works for you (Canadian capacity is being evaluated)
- You have meaningful revenue or funding, or can offer a deposit, letter of credit or guarantee
How it works
- 1
Apply
Ten minutes: your company, the capacity you need and your financial position.
- 2
Intro call
We confirm the fit and walk you through options, timing and pricing.
- 3
NDA and documents
Accept a mutual NDA, then upload financials to your secure, encrypted document room.
- 4
Proposal
We prepare your credit package and bring you a proposal from our data center partner.
- 5
Sign directly
You contract directly with the data center. We support you through go-live and beyond.
Questions
Who is the data center partner?
We name our partners once you've applied and accepted a mutual NDA. They run purpose-built AI inference modules at US sites.
What's the minimum?
128 GPUs for 24 months. Larger, longer commitments get the best pricing and the earliest dates.
What documents will you ask for?
Usually your last two years of financial statements (or management accounts), recent bank statements, and details of your latest funding round. Nothing until you've accepted the NDA.
How is my information protected?
Documents are encrypted before they're stored, on servers in Canada, and seen only by RRB Partners and, with your permission, the data center you're applying to. You can ask us to delete them at any time.
Is training a fit?
The modules are built for inference and fine-tuning. Large pre-training runs are a weaker fit, but tell us what you need.
Tell us what you need
Ten minutes to apply. We reply within two business days, and nothing is shared without your permission.