H10 is a GPU-ready private AI pod for your home, estate or private office. Racks, 208/480 V three-phase power, liquid cooling, UPS backup and segmented networking, engineered as one outdoor system that you own.
AI is moving from a tool you visit to a system that reads your documents, watches your cameras and helps you decide. Running it on hardware you own changes who can see that data, what it costs to keep running, and whether it works when the network doesn't.
Family records, investment and legal files, health documents and camera footage are processed on hardware you own, under access rules you set.
If the internet goes down, local inference, knowledge search and home automation continue. Updates wait for a maintenance window you approve.
Agents that run all day, nightly indexing and media generation run on fixed hardware you own, instead of a bill that grows with every request.
Management, AI and storage, and home devices live on separate network zones. Nothing is reachable from the internet by default.
Search and summarize family documents, photos, archives and records, with permissions for each household member.
Run approved AI workflows over investment, legal, research or business material inside a controlled environment.
Sustained model inference, fine-tuning, media generation or agent development on GPU servers you choose.
Connect local AI to selected cameras, sensors, automation, backups and media services, each on its own network.
H10 is infrastructure, not a consumer chatbot. Cloud services and external search still need the internet; H10 keeps the workloads you choose local.
Choose a pod, add compute and options, and watch the price and site requirements update as you go.
We source, integrate and test the servers for you. GPUs are passed through at market price with no markup; you pay only for integration and testing.
Already own servers or have a preferred vendor? We review power, cooling, connectors and shutdown integration, then certify them for your pod.
Route A: add servers below at market price. Route B: choose "Bring your own servers". You can mix both. Power figures are planning values; final loading follows the server vendor's data.
Options for longer outages, redundant cooling, and the conditions at your property.
Service plans are agreed separately when you sign. See what each plan covers.
We'll review your configuration, schedule a site survey, and send a firm quote.
Every pod is monitored around the clock by our team. Here is exactly what each plan covers.
| Essential | Priority | Concierge | |
|---|---|---|---|
| Monitoring | 24/7 by our team | 24/7 by our team | 24/7 by our team |
| Remote response | Next business day | Within 4 hours | Within 1 hour |
| On-site visit | Within 48 hours, billed per visit | Within 48 hours, included | Within 48 hours, included |
| Preventive maintenance | Remote health checks | Annual on-site service | Quarterly on-site visit and restore drill |
Response times start when an alarm fires or you report an issue. On-site response covers the San Francisco Bay Area; elsewhere, visits are scheduled with you and travel is billed at cost. Plan pricing is set in your service agreement when you sign.
A weather-resistant outdoor pod about the size of a 10-foot shipping container, plus a separate utility skid. Inside are two 19-inch racks for GPU servers and storage. The skid holds the UPS and battery, cooling distribution and power protection. It is the room, power, cooling and network your AI servers need, delivered as one tested system.
For many tasks the cloud is fine. H10 is for work where data control matters, where workloads run all day, or where several people and systems share AI at home or in a private office. You decide which workloads stay local and which, if any, use outside services.
Yes, for local workloads. Approved local models, knowledge-base search, storage and home automation keep running. Cloud APIs, external search and remote support are unavailable until the connection returns.
That depends on the servers you choose. An inference node with four RTX PRO 6000 Blackwell GPUs has 384 GB of GPU memory, enough for large open-weight language models, document search and image generation. 8-GPU H200 and B200 nodes in H10 Pro handle the largest open models and fine-tuning.
H10 runs on 208 V or 480 V three-phase power. Standard typically needs a 50 A three-phase breaker at 208 V; Pro needs about 100 A at 208 V. Most homes have 120/240 V single-phase service, so a utility upgrade or phase converter may be needed. Our site survey confirms this before you commit, and a licensed electrician signs off the final design.
The design target is 45 dBA or less at night at the nearest bedroom or neighbor you agree with us, about the level of a quiet library. The optional acoustic package adds barriers and a night mode. We measure on site after installation.
The UPS carries your critical loads for 10 to 15 minutes (30 with the extended option). The controller checkpoints or stops jobs, shuts servers down in order and keeps pumps running until hardware is cool. A generator interface is available as an option.
No. You can have us supply them (Route A), with GPUs passed through at market price and no markup, or bring your own (Route B). For your own servers, choose "Bring your own servers" in the configurator. We review each server's dimensions, weight, power input, connectors, cooling mode and shutdown method before installation.
Our team monitors every pod 24 hours a day. Depending on your plan, we respond remotely within 1 hour, 4 hours or the next business day, and a technician is on site within 48 hours when a visit is needed. See the service plans above for what each plan includes.
You own the administrator credentials, encryption keys and every decision about whether data leaves the site. Remote support uses time-limited access with multi-factor sign-in and audit logs, and only when you allow it.
Usually yes. Even though the pod is prefabricated, a home installation can require electrical, structural and fire review. Our permitting and engineering support option prepares the package for your jurisdiction.
No. It is an estimate in US dollars. GPU servers follow market prices, and the final quote depends on the site survey, permits and local code. You get a firm quote after the survey. Service plans are priced separately in your service agreement.