New whitepaperDemocratizing intelligence through community compute gardens
Reflex Logo
Reflex
TECHNOLOGIES
WhitepapersBook a site visit

Compute Where You Are

Own a share of a community compute center.
Or install compute in your own building.
We run it and pay you from what it earns.

Compute

The machines that produce intelligence, owned by the people around them.

Reflex develops community compute centers that members own together, and installs compute modules in facilities that want hardware of their own. Either way, Reflex operates the hardware, sells what it produces, and pays the owners.

A community solar farm, the ownership model compute centers follow
Community compute centers

We develop the center. Members own it.

Reflex builds an NVIDIA SuperPOD in a community center, business park, or campus, sells its output as model-as-a-service, and pays members from what it earns.

Hardware
NVIDIA Vera Rubin SuperPOD
Host sites
Community centers, business parks, campuses
Ownership
Shared, by member shares
Modeled revenue
~$800M / yr per center
Read the whitepaper
Wall-mounted Reflex compute nodes in a plant room
Facility modules

Install compute in a building you already have.

Air nodes for homes and business buildings, liquid-cooled nodes for plant rooms. Reflex installs, operates, and sells the capacity, and the owner keeps the income.

Hardware
Air node, liquid-cooled node
Host sites
Homes, offices, shops, plants
Ownership
The building owner
Install
Existing service, no data hall
See the products
Products

From one wall to a whole institution.

  1. Air node

    Sealed and air-cooled, quiet enough to hang on an occupied floor. It runs on the service the building already has, so there is no plant room to find and no structural work to book.

    Silicon
    16x L4 class
    Power
    ~1.4 kW
    Cooling
    Air, sealed panel
    Workload
    Inference
    Revenue
    ~$34K / yr
    A single sealed Reflex air node mounted flat against a wall
    One ST-16H air node — sealed, fanless to the room, on a standard bracket.
  2. Liquid-cooled node

    Direct liquid pulls an order of magnitude more heat out of the same footprint. Nodes mount in a plant room and run on a closed loop, or exchange heat into the building's chilled or district cooling.

    Silicon
    16x RTX PRO 6000
    Power
    ~12 kW
    Cooling
    Direct liquid
    Workload
    Inference, fine-tuning
    Revenue
    ~$144K / yr
    Five liquid-cooled Reflex compute nodes in a plant room, coolant and service lines run in beside them
    Liquid-cooled nodes on one plant-room wall.
  3. NVIDIA SuperPOD

    Eight NVIDIA Vera Rubin NVL72 racks on one fabric and one liquid loop. This is the hardware behind a community compute center, and the most efficient GPUs return the most to the people who own them.

    Silicon
    8x Vera Rubin NVL72, 576 GPUs
    Power
    ~2 MW
    Cooling
    Direct liquid, facility loop
    Workload
    Model-as-a-service at scale
    Revenue
    ~$800M / yr, modeled
    Rows of NVIDIA DGX Vera Rubin NVL72 racks
    NVIDIA DGX Vera Rubin NVL72. A SuperPOD runs eight racks.
A single sealed Reflex air node mounted flat against a wall
One ST-16H air node — sealed, fanless to the room, on a standard bracket.

Compute you configure
and actually own.

One node, sixteen GPUs
The ST-16H is the same unit on a garage wall and in a research rack. Scale by adding nodes, not by re-architecting.
Rent compute through our partners
Burst into partner capacity by the hour, on the same control plane.
On-demand and reserved terms
Regional placement and data residency
Member shares and monthly payouts
Own the hardware outright
Buy nodes and rack them with us or on your floor. Rented capacity converts to owned, and your utilization carries over.
AI model inference
Serve your own weights behind a managed endpoint with autoscaling replicas, batching, and per-token accounting.
shoptalk.com/compute/leases
JD
Filters
Clear
Search tenants
Window
Active terms
Q3 2026
Q4 2026
Lease type
All contracts18
Dedicated clusters7
Reserved nodes6
On-demand burst5
Status
Active
Provisioning
Renewing
+ New tenant lease
Compute
Tenant leases & fleet
Active Leases14 accounts
Committed336 / 384 GPUs
Active tenants18 accounts
Fleet utilization90.6% (464 GPUs)
Contract ARR$2.71M / yr
Tenant
Cluster / Node
Contract
Capacity
Status
Rate/mo
Ridgeway Materials Lab
Cluster Alpha (Pod 2)
LEAS-5108
96 GPUs (98%)
Active
$37,200/mo
Halstead College
Dedicated Rack 04 (4 nodes)
LEAS-5094
64 GPUs (100%)
Active
$24,800/mo
Brightwater Clinics
2x Node ST-16H
LEAS-5081
32 GPUs (86%)
Active
$12,400/mo
Marlow Street Residences
2x Wall Node ST-16H
LEAS-5077
32 GPUs (74%)
Active
$12,400/mo
Kestrel Design Studio
Wall Node ST-16H
LEAS-5063
16 GPUs (91%)
Active
$6,200/mo
Fenwick Robotics Dept.
Cluster Beta (Pod 1)
LEAS-5044
64 GPUs (93%)
Active
$24,800/mo
Coalport Logistics
Node ST-16H
LEAS-5029
16 GPUs (100%)
Renewing
$6,200/mo
Ellery Institute
Partner Burst Pool
LEAS-5012
32 GPUs (68%)
Active
$12,400/mo
Sablewood Estates
6x Wall Node ST-16H
LEAS-4998
96 GPUs (82%)
Active
$37,200/mo
Thorne Analytics
Node ST-16H Chassis
LEAS-4981
16 GPUs (100%)
Provisioning
$6,200/mo

Get Started