COMMUNITY EDGE AI CONTAINER
One node. Local compute, network and resilience.
A CEAC is a deployable 20ft community data-centre node designed to bring useful AI compute and digital infrastructure closer to the people, organisations and services that use it.
The current Generation-1 design is built around approximately 100 households and combines AI acceleration, NVMe storage, networking, protected electrical systems, battery storage, directed cooling, monitoring, fire protection and physical security.
GENERATION-1 ENGINEERING MODEL
Interactive CEAC engineering model
Explore the current Generation-1 design in 3D. Drag to orbit, scroll to zoom and use the engineering controls to inspect the system.
Full-width interactive CEAC digital twin with orbit, cutaway, labels, clearance zones and P5-D airflow architecture controls.
Inside the current Generation-1 CEAC
The Generation-1 design brings compute, network, protected power, cooling, monitoring and safety systems together inside a serviceable 20ft infrastructure platform.
COMPUTE
- 4 × NVIDIA A10 GPUs
- 24 GB GDDR6 each
- 96 GB aggregate physical GPU memory
- 600 W combined maximum GPU board-power envelope
- NVMe local storage
- Rack server architecture
NETWORK
- Gateway
- Network switch
- Optical distribution frame
- Out-of-band management
- Upstream backhaul
- Community access network
PROTECTED POWER
- 10 kW online UPS
- 30.72 kWh nominal battery ESS
- Distribution and protection
- Hybrid inverter
- Metered PDU
COOLING
- Directed rack airflow
- Primary fan-coil cooling
- Independent DX failsafe
- Environmental monitoring
- Inlet/exhaust/room sensing architecture
SAFETY AND ACCESS
- Keypad + Heartprint access
- Equipment service access
- Fire detection
- Heat detection
- Alarm/beacon
- Emergency stop
- Battery separation
- Environmental telemetry
PHYSICAL SECURITY
Keypad + Heartprint access
CEAC combines a secure keypad with Heartprint, our newly developed contactless Sparse radar verification technology.
The keypad identifies who is requesting access. Heartprint then uses radar to scan the tiny movements created by the person’s heartbeat and verifies that the live Heartprint matches their enrolled access profile.
No camera. No fingerprint reader. No radar-based identity search.
Every attempt is recorded as PASS, RETRY or DENY, creating an auditable access trail before any physical unlock is considered.
Two-factor physical access, built for secure community infrastructure.
The infrastructure should outlive the GPU
CEAC is being designed as an infrastructure asset rather than a container built around one accelerator generation.
The shell, electrical systems, networking, security, cooling and site connection should have a longer useful life than the compute hardware. The compute layer should therefore be replaceable as accelerator technology changes.
Replace the compute sled — not the community infrastructure.
Put the right workloads at the edge
STRONG FIT
- Local LLM/chat
- Retrieval-augmented generation
- Embeddings and vector search
- Document processing
- Speech recognition
- Text-to-speech
- Sensor inference
- Vision inference
- Community services
- Education
- Local-government applications
- Privacy/locality-sensitive processing
- Time-shiftable batch inference
POSSIBLE / WORKLOAD DEPENDENT
- LoRA
- Small fine-tuning
- Compatible multi-GPU serving
LEAVE CENTRALISED
- Frontier model training
- Very large tightly coupled accelerator jobs
- Workloads requiring hyperscale/HPC interconnect
CEAC is principally a distributed inference architecture. It does not need to do every kind of compute to be useful.