SOVEREIGN ON-PREM AI
Your AI.
Your premises.
Your rules.
The FalloutBox Sovereign Cluster runs large-model inference entirely on-site. Zero cloud dependency. Zero data egress. Nothing leaves the building.
WHY FALLOUTBOX
One cluster. Everywhere it matters.
Sovereign by architecture
Every token is generated inside your walls. Nothing leaves the building — not prompts, not weights, not logs.
Data-center class, closet size
A mini-rack that sits under a desk yet serves an entire facility.
Built for real-time, whole-team inference
Dual compute modules drive fast local inference for whole teams, not just one user.
Scale by stacking
Start with one cluster. Add capacity by adding boxes — no re-architecture, no migration.
Runs anywhere there's power
Solar-friendly draw; deployable in greenhouses, classrooms, warehouses, and field camps.
Human-in-the-loop control
FBOS proposes; your people approve. Every automated action is auditable.
WHERE IT RUNS
One box. Ten very different jobs.
Coral Farms & Aquaculture
Live sensor fusion and reef/tank monitoring at the water's edge. Fully on-prem.
AI Café
Sell AI time by the minute alongside espresso. Customer data never leaves the shop.
Vocational School
One solar-assisted rack powers an entire coding and CAD lab.
Automated Supply Chain
Inventory that proposes its own reorders. A human approves every one.
Personal Assistant
A whole-home voice assistant, processed locally. Nothing leaves this home.
Shared Co-Working Space
AI-time kiosk at the counter; one box serves every desk.
Disaster Response
The ruggedized mobile case: compute, comms, and coordination with infrastructure down.
Neo Factory
Line-side inference for inspection and control, with no round trip to the cloud.
Relief Camp
Registration, translation, and coordination for displaced communities, fully offline-capable.
Token-Maxxing
Spare inference capacity, metered and sold by the token instead of sitting idle — settled automatically, no cloud middleman.
SEE IT RUN
Watch the film.
How it works
Plug it in.
Connect power and network. FBOS auto-configures in minutes — no IT team required.
Runs on day one.
Arrives preloaded with a curated model set — model-agnostic architecture, so you're never locked to one vendor's weights. Add or swap models anytime.
Serve everyone.
Workstations, tablets, and kiosks connect over your LAN. No internet required.
Coming next: tune and fine-tune your own model weights directly on-device.
> SELECT HARDWARE PROFILE
Three tiers. One platform.
Mobile and Cluster Rack run the same NVIDIA Grace Blackwell chip — no watered-down "mobile" silicon. Node count and packaging are the only differences, so nothing you learn on Mobile stops working when you scale to a Cluster Rack. FBOG is a third, larger tier currently in design — a different, data-center-class silicon platform for solar/wind-sited deployments, not a bigger Grace Blackwell cluster.
The sovereign node you carry.
- NVIDIA Grace Blackwell (GX10-class), 128GB unified memory
- Runs the full FBOS stack locally — zero cloud dependency
- Off-grid microgrid ready — pairs with a modular solar panel + battery (EcoFlow DELTA 3 Plus-class), Home Assistant-integrated for live power telemetry
- 10Gbps Ethernet LAN, 8TB base storage
- Built-in 5G/Wi-Fi hotspot with OpenVPN client/server support (e.g. GL.iNet Mudi 7-class) — connectivity independent of any fixed ISP
- Meshtastic-ready — mesh connectivity to the local LLM via Home Assistant (planned)
Draft price: TBD — hardware-cost-plus, contact for current BOM
Order terms (draft): reserving is an expression of interest, not a payment — no funds are collected up front. Final price is confirmed against a real, current bill of materials before any order ships.
Same platform. Scaled to a data center in a box.
- 2x Grace Blackwell units — same platform as Mobile
- 256GB aggregate unified memory (2x128GB)
- 200Gbps backplane between nodes, plus a dedicated 10Gbps switch and OpenVPN-capable router for the rack's shared site uplink
- 10Gbps Ethernet LAN + 8TB base storage per node
- Half-rack enclosure, rack-mount form factor
- Meshtastic-ready — one mesh gateway serves the whole rack, mesh connectivity to the local LLM via Home Assistant (planned)
- Off-grid microgrid ready — modular, per-node solar panel + battery (EcoFlow DELTA 3 Plus-class) for sites without reliable grid power
Pre-sale cohort: $35,000 per finished FBR2 — hard cap of 5 seats
Cohort terms: $17,500 deposit (50%) holds a seat and starts the build; remaining $17,500 due on shipping. Targeting ~14 days from deposit as the supply chain finalizes — a target, not a guarantee. First-come among qualified buyers. Contact only on this site for now — no embedded payment yet. A seat is not reserved until the deposit is actually received.
> FBR2 pre-sale — express seat interest
Data-center-scale, solar/wind-sited. A different silicon class, not a bigger cluster.
- Data-center-class silicon: NVIDIA RTX PRO 6000 Blackwell Server Edition (96GB GDDR7 ECC per GPU) — not more GX10/DGX Spark units
- Modular 8-GPU “cube” building block; tower ladder from a 1-cube pilot (8 GPUs) up to a 3-cube flagship (24 GPUs)
- Optional H100 SXM “quad” module (real NVLink fabric) for workloads that need coherent memory beyond one 96GB card
- Designed for solar/wind-sited, DC-coupled deployment — installs behind the site inverter, skipping the AC/DC round-trip a grid-connected data center pays
- Same FBOS software stack and sovereignty posture as Mobile and Cluster Rack, on a different hardware tier
Estimated first-pilot cost: ~$107K-153K in bare components, or ~$165K-237K turnkey — both explicit estimates, not a quote
Status: DESIGN — not built, not costed against a real BOM yet. Targeting a first single-tower pilot around December 2026. The cost range above is a first-pass estimate pending a real WooTech BOM/vendor quote. Site-partner thread: 90-day flagship partner seek (solar farm / power co-op, data-center operator, or capital-constrained mining site) — see the FBOG partner section below. No partner contacted or confirmed.
> FBOG 90-day partner arrangementFBOG is the data-center-scale third FalloutBox SKU (different silicon class from FB01 / FBR2). WooTech is running a 90-day partner-seeking arrangement, with intent to begin around December 2026 — matching the existing design target, not a newly invented ship date. This is a site / anchor-partner conversation, not a consumer pre-order.
Flagship-caliber anchor, not a random first pilot
The operator model is WooTech's equivalent of Span's relationship with PulteGroup: Span (maker of XFRA, the distributed residential mini-data-center already documented as a competitive reference) partnered with a major homebuilder as its flagship / anchor implementation partner for a multi-home proof-of-concept — not a one-off random install. WooTech is seeking the same class of anchor partner for FBOG: a serious site operator who can host and prove a dense single-tower deployment, not just any first taker.
Who we want to talk to
- Solar farm and/or local power co-op (land, generation, interconnect already in place)
- Data center operator evaluating a dense single-tower co-location path
- Bitcoin / crypto mining operation with energized land and grid interconnect but without the multi-million-per-MW capital for a full AI retrofit
Mining-operation angle is grounded in documented 2026 industry reality: capital-constrained miners often already hold the scarce assets (energized land, substations, interconnect) but not the ~$8–15M/MW for a full AI retrofit. No specific miner, farm, co-op, or colo is named or contacted here.
What is already documented about the hardware
- GPU density per tower (narrow claim only): FBOG's 3-cube flagship carries 24× RTX PRO 6000 Blackwell GPUs vs. XFRA's confirmed 16 GPUs per unit — a verified 50% higher GPU count per physical tower. This is a GPU-count-per-tower comparison only — not a blanket claim that FBOG is 50% better on throughput, cost, or efficiency.
- Mix-and-match module types per site: two separate module types share the external tower shell and cooling design — an 8-GPU RTX PRO 6000 Blackwell module (bulk / primary) and a 4-GPU H100 NVLink module (specialty, for large single-model tensor-parallel workloads beyond 96GB). They are not a shared-backplane / any-slot-takes-either design; a single shared backplane across both GPU types was considered and rejected.
"Token Factory" participation (direction, not a term sheet)
FalloutBox's longer-horizon monetization work already frames an electron economy / token factory path: metering and selling excess or idle AI inference capacity (e.g. via x402), with an adjacent mining "24/7 token factory" angle for sites that skew toward mining infrastructure. The partner opportunity under discussion is to participate in ownership of token-factory revenue generated by the site's AI compute (and/or mining) — not only a one-time hosting fee.
Exact ownership percentage, legal entity, and commercial structure are undefined. This is the shape of the ask to negotiate — not a fixed equity split or signed structure.
What is still not true
- No specific partner has been contacted under this arrangement.
- No LOI, term sheet, deposit, site agreement, or ownership percentage exists.
- FBOG remains DESIGN status — not built, not costed against a WooTech BOM, not a committed public launch.
- Nothing on this page implies a partner, site, pilot, or revenue share is confirmed.
Contact only — no embedded payment or application portal. Use the button above; we don't list the address in plain text here to keep spam and bot traffic down.
> FBR2 PRE-SALE COHORT
Reserve interest in an FBR2 seat
5 SEATS MAXSmall pre-sale cohort for the FalloutBox Cluster Rack (FBR2) — SKU 2. This page is a contact / lead-capture step only. There is no embedded payment yet. Expressing interest here does not reserve a seat, collect a deposit, or confirm a ship date. A seat is held only after a qualified buyer's $17,500 deposit is actually received and recorded.
Operator-set commercial terms for this cohort. Read them before you reach out.
Unit price
$35,000 USD per finished FBR2 (Cluster Rack)
Deposit to hold a seat
50% up front — $17,500. Build work starts when the deposit is received.
Balance + timeline
Remaining $17,500 due on shipping. Targeting ~14 days from deposit receipt as the supply chain finalizes — a target, not a guaranteed ship date.
Seats
Hard cap of 5 seats total. First-come, first-served among qualified buyers.
Qualification still applies (fit, intent to take physical hardware, ability to fund deposit and balance). First-come among qualified buyers — not a vanity waitlist. No sale, deposit, or shipment is implied by submitting this form.
Model selection: for this initial 3–5 seat cohort, we'll work with you — within reason — to tailor the preloaded model set to your workload, or you can simply start with our curated, recommended model list. Either way, the platform stays model-agnostic and you can add or swap models later.
Opens your email client with this filled in. No payment processor on this page.
On-prem vs. cloud
Own your AI.
See the FalloutBox Sovereign Cluster running live at your facility.