GeofabriqRFGet in touch

OverviewFor occupants

Your AI,on your floor.

Private models on the building's own network, on servers you own or lease, run for you around the clock. Your data never leaves the building, and the work keeps going when the internet does not.

Tell us what you would run
Network
Your building's LAN
Hardware
Yours, owned or leased
Routing
Pinned to the building
Continuity
Behind the generator
Assurance
SOC 2 before client data
03Who it is for

For the work you keep in the building.

Each kind of occupant buys it differently. These are the three we are building the first rooms around.

  1. 01

    Hospitals

    Video analytics across your cameras, bed-flow and operations models, research on de-identified data, and the fleet manager for your delivery robots, all behind your generator. Clinical AI arrives through your EHR's governance or an ISV that deploys on-premises.

    Where
    Racks in your data center, or a Node in your building.
    Assurance
    A HITRUST report before any patient data. A business associate agreement as standard.
  2. 02

    Universities

    You probably own a room and a cluster already. Buy Serve on it: the platform, tenant isolation, metering and a model estate your faculty can fine-tune, for the enclaves and persistent serving that shared supercomputing allocations cannot take.

    Start
    A no-cost research memorandum of understanding.
    Assurance
    HECVAT and TX-RAMP in hand before any paid subscription.
  3. 03

    Firms in towers

    A 150- to 400-person firm needs four to eight GPUs for a restricted-matter enclave, assistants and document AI, with retrieval trimmed to your document permissions. Your servers move with you if you leave the building.

    Hardware
    Bought or leased through our occupant lease program.
    Keys
    Customer-managed, with break-glass alerts you can see.
02What you get

Private by construction.

Not a cheaper GPU-hour. A Node is bought for privacy, continuity and the network, and every proposal says so plainly.

  • Your data stays in the building

    Models run on your servers, on the building's network, behind its firewall. Nothing is sent to a hyperscaler to be reasoned over.

  • A pin you can audit

    No routing to frontier models, no pooling with other tenants, no resale of your capacity, unless you turn it on. The pin is contractual and audited.

  • Work that survives an outage

    Inference runs behind the building's UPS and generator, with failover between Nodes in the metro, so AI-assisted work does not stop when the internet does.

  • Capacity you own

    Your GPUs, on your lease or your books, serviced like an elevator. No queue, no egress fees, no surprise reprice.

  • Open models, licensed where it counts

    Open-weight models carry most private work. Licensed or routed frontier models run only where you allow them.

  • Numbers shown plainly

    Every proposal compares your all-in cost per GPU-month with the cloud, including where the cloud is cheaper.

01How it works

From a workload to a running enclave.

  1. Scope

    Size it

    We size the work with you: enclaves, named applications, cameras, robots.

  2. Order

    Own it

    Servers bought or leased in your name, with a Serve subscription on top.

  3. Commission

    Start it

    Your own non-client workloads first; client data after the assurance report.

  4. Run

    Keep it

    Around-the-clock operations, metering and billing, for the life of the term.

LLobby

Hold your placein the first rooms.

Tell us what you would run. A memorandum of understanding holds your place in Austin's first Nodes.

For building owners →For partners and investors →Overview →