The questions people ask before they call.

GPU One is STN’s fully managed, dedicated GPU cloud, built for teams running AI in production rather than teams still testing ideas. We own and operate the infrastructure, and we manage the whole stack around it: compute, networking, storage, security, observability, and 24/7 monitoring from our own security operations center. Your team works on models and products instead of running servers.

We did not start as a GPU provider. STN has spent nine years running managed private cloud, data management, and security operations for enterprise and regulated customers, and GPU One grew out of that. The practical difference is that the operational discipline came first. Monitoring, access control, compliance, and incident response were already how we work, so they are built into the platform rather than added after a customer asks for them. We also do not hand you a cluster and walk away. We ask what you are actually trying to build, then design and operate the environment around it.

New deployments run on NVIDIA B300 and GB300 NVL72. Long-term contracts include a built-in upgrade path to NVIDIA Vera Rubin, which reaches mass availability in Q1 2027, so a multi-year commitment does not lock you to a single hardware generation. We continue to host earlier-generation hardware for customers who already own it.

Your GPU capacity is reserved and dedicated to you. You are not competing for cycles with other tenants, so the performance you plan around is the performance you get. Network and access isolation is enforced separately: every customer gets a dedicated virtual firewall instance tied into your own identity provider, with your environment isolated at the network layer and multi-factor authentication required for access.

We run Ethernet end to end, with no InfiniBand. Our clusters use NVIDIA Spectrum-X switching with a 400 Gbps interconnect on current systems and 800 Gbps on the next generation, and 200 Gbps Ethernet to storage. Ethernet gives us a single fabric to operate, monitor, and secure across every site, which is a large part of why our uptime and support posture holds up at scale.

GPU One is SOC 2 Type II certified and HIPAA compliant, and we are currently pursuing FedRAMP, starting at Moderate. Security is a standing business line at STN, not a checkbox: we operate our own security operations center, run dedicated Palo Alto firewalls with enforced multi-factor access, and integrate with your identity provider for single sign-on. A due diligence package is available on request.

No, and that is the point. Managed services are included in the rate, not sold on top of it. That covers 24/7 monitoring, provisioning, performance tuning, and capacity planning, along with a complete managed Slurm or Kubernetes deployment, networking, and observability delivered to your team. There is nothing to remove for a discount, because the operating model is the product.

Yes. If you would rather own the hardware and keep it on your balance sheet, you can buy the systems and we will host and operate them under the same managed services model. Hosting is priced on power draw rather than per GPU. It is a common structure for teams that have already made the capital commitment and want the operations handled.

Three groups. Frontier labs training foundation models, who need large-scale capacity, a clear hardware roadmap, and a provider willing to commit power and space years ahead. AI builders and platform companies running training and inference at production scale, who need performance they can plan around and cost they can forecast. And enterprises moving AI into production, particularly in regulated sectors like healthcare, life sciences, and robotics, who need secure and compliant infrastructure without building and staffing it themselves.

We start by understanding the workload, the data, and your risk and compliance requirements, then design the architecture around them. From there we handle secure connectivity, your container and scheduler environment, identity and access integration, and observability, and we validate the whole thing in practice before you rely on it. After go-live we tune as you use it and plan capacity ahead of your growth rather than after it.

Our minimum economic deployment is roughly 64 GPUs, or one rack, though we can go smaller for a proof of concept. Lead time depends on the site and the configuration, so the honest answer is that it is worth asking early. Capacity in this market moves quickly, and giving us advance notice is the single biggest thing that determines whether we can meet your date.

Talk to us directly. We will walk through what you are building, what you need, and what it costs. No pressure and no discovery theater, just a straight conversation about whether we are a fit.