CATALYST / DEDICATED INFRASTRUCTURE
Dedicated T3/4 GPU regions in the USA. Secured space, power and GPU allocations, with a region-delivery team accountable from deposit to acceptance.
Built for AI labs, frontier workloads and enterprise scale.
Full operations + 24/7 support
01 / THE ENVIRONMENT
Start with your workload, not a fixed cloud menu. We align the hardware, facility and delivery plan around what your team needs to run.
Dedicated GPU capacity and bare-metal access. Catalyst oversees procurement and delivery. Our established neocloud operating partner deploys the infrastructure and provides full operations and 24/7 support.
Contiguous space options are available for customers who need a cohesive footprint. Plan deployment and expansion together, with site and fabric boundaries agreed up front.
Configure non-blocking RoCE networking, CPU and control nodes, NFS or parallel storage, local NVMe, private connectivity and observability around the workload.
NVIDIA B200 · B300 · GB300 NVL72
AMD MI325X · MI350X · MI355X · MI450
Need a contiguous footprint or a specific configuration?
OPERATIONS / BUILT FOR PRODUCTION
Cobalt combines Catalyst Capacity’s procurement expertise and end-to-end delivery oversight with infrastructure deployment, full operations and 24/7 technical support from an established neocloud operator.
Partner credentials and detailed service-level commitments are available during technical due diligence.
Day-to-day management of your GPU environment, with ownership and responsibilities agreed before deployment.
Around-the-clock support, direct technical escalation and coordinated incident response.
Contracted availability, response targets and service credits, with clear measurement and accountability.
Managed-service scope, pricing and exact SLA commitments are confirmed in your proposal and signed operating schedule.
02 / ACCOUNTABLE DELIVERY
An end-to-end region-delivery TPM at your disposal, from allocation to customer acceptance.
Our team’s leadership includes the founding product manager of Oracle Dedicated Region, alongside GPU infrastructure, capacity-planning and technical program management expertise.
Baseline your design, site, acceptance criteria and committed delivery dates.
Hardware, power, cooling, fabric and storage brought together to your specification.
Burn-in, fabric validation and a documented handover package.
Test your workload for three days. Not satisfied with performance? Receive a full deposit refund. Service start and billing follow acceptance.
*4–16 weeks from cleared deposit to handover for agreed RoCE configurations. InfiniBand adds a minimum 3–4 weeks. Customer testing follows handover. Final dates are agreed per order.
Have a handover date your team needs to protect?
03 / BUILD YOUR COMMITMENT
A transparent starting point for your procurement team. Choose a GPU, set your capacity and model your commitment.
32 GPU nodes · 256 GPUs
Applied toward contract charges
Three days to validate performance. Not satisfied? Get your full 30% deposit back.
One email to connect. Your configuration comes with you.
Capacity TCV = GPU count × USD/GPU-hour rate × 730 hours × months. Deposit = 30% of estimated TCV, not an additional charge. The calculator models 24/7 reserved capacity using 730 hours per month, equivalent to 8,760 hours per year. Actual billing follows the signed order’s calendar-hour or fixed-month convention.
Published rates use a three-year base commitment. Each additional year reduces the GPU-hour rate by $0.10 across the entire contracted term: $0.10 less for four years, $0.20 less for five, and $0.30 less for six. Rates remain indicative, subject to the final quote. Taxes, separately priced storage, CPU nodes, managed services, connectivity and financing costs are excluded. Eight GPUs per node is a reference configuration for the listed server-based shapes, subject to OEM confirmation. GB300 NVL72 and MI450 require a custom quote and a validated GPU-to-node mapping. No hourly rate or node conversion is assumed for either platform.
CAPITAL THAT UNDERSTANDS COMPUTE
Specialist GPU financing arranged through our partners, aligned with your infrastructure commitment and deployment plan.
Talk to Us about financingIndicative financing rate
Target start-to-finish timeline
Subject to lender underwriting, eligibility, complete documentation and final terms. The ~8% rate is indicative; the lender will confirm the annualized rate, APR where applicable, fees, security and repayment schedule. The 1–2 week process is a target, not a funding guarantee. Financing is separate from the capacity estimate.
04 / COMMERCIAL CLARITY
Six principles behind your Cobalt agreement. Performance thresholds, delivery remedies and operating commitments are tailored to your deployment and documented before you commit.
Test performance for three days after handover. If you are not satisfied, receive a full refund of your 30% TCV deposit.
Service start and recurring service billing follow acceptance. Delivery alone does not trigger billing for unaccepted capacity.
Committed handover dates and late-delivery penalties are specified in the service order, with agreed remedies for prolonged delays.
Hardware, fabric, storage and site specifications are recorded in the order. Material substitutions require your written approval.
The draft protects customer data, model weights and intellectual property. No model training or competitive use without explicit authorization.
An established neocloud operator delivers deployment, full operations and round-the-clock support. Catalyst provides delivery oversight. Service levels, response targets, escalation and credits are documented in your order.
Request a draft MSA tailored to your organization and deployment. Our team will review your requirements, customize the agreement and send it to you directly.
Prepared by our team, not an automatic download. Drafts are for discussion, not a binding offer. Legal review and completed commercial schedules are required before signature.
05 / BEFORE YOU COMMIT
Bring your architecture, workload and deadline. We’ll work through the rest with you.
Talk to UsNodes. Cobalt supports requirements from 12 to 256 GPU nodes. Rack footprint depends on the selected server, power, cooling and network design. GB300 NVL72 and MI450 need an agreed node definition and ordering increment; neither is automatically treated as an eight-GPU server.
Cleared deposit to node handover for an agreed RoCE configuration. The three-day customer testing window follows handover. InfiniBand adds a minimum 3–4 weeks. Your service order records committed dates, dependencies and late-delivery remedies.
You have three days after complete handover to validate performance with your workload. If you are not satisfied with performance, notify us during that window to receive a full refund of your 30% TCV deposit. The test plan helps structure validation, but a refund is not limited to a failed numerical benchmark. The order records the testing clock, notice process and refund timing. Service start and recurring billing follow your acceptance.
The service order includes committed handover dates and a late-delivery penalty schedule. The parties agree triggers, amounts, caps and payment or credit mechanics, plus prolonged-delay remedies, before deposit. Delivery penalties are separate from operating SLA credits.
Yes, contiguous space options are available. We confirm the facility footprint, expansion allocation, reservation period, pricing and lead time in your proposal. Contiguous space does not automatically mean a single fabric or unlimited expansion; those are explicit design decisions.
GPU shape, CPU/control nodes, local NVMe, NFS or parallel storage, fabric, private connectivity and monitoring can be tailored. Bare-metal access is included. Catalyst Capacity leads procurement and end-to-end delivery oversight. An established neocloud operator deploys the infrastructure and provides full operations and 24/7 technical support. The final quote and operating schedule define managed-service scope, pricing, root/BMC access, firmware, networking and security responsibilities. Partner credentials are available during technical due diligence.
Our established neocloud operating partner provides full infrastructure operations, 24/7 technical support and best-in-class SLAs. Your operating schedule names the responsible teams and sets availability commitments, incident severity levels, response and update targets, escalation contacts and service credits. Detailed commitments are available during technical due diligence, with exact metrics and service inclusions confirmed in your proposal and signed order.
We arrange GPU-specialist financing through partners at an indicative rate around 8%, with a target 1–2 week process from start to finish. Approval, rate basis, fees, collateral and repayment terms depend on underwriting and the final lender offer. Financing approval does not itself reserve capacity, and borrowing costs are not included in this calculator.
We source Tier III / IV facility options in the USA, with Canadian options available. The specific facility’s resilience design, certification status, audit evidence, power and cooling are confirmed during diligence. T3/4 is not a blanket certification of Catalyst or every available site.
No. These are starting USD/GPU-hour rates and modeled capacity commitments. The final quote confirms site, term, GPU/node configuration, inclusions, taxes, delivery dates and contract terms. No capacity is booked and no money is collected by this page.
YOUR NEXT REGION STARTS HERE
