The infrastructure Physical AI is missing.
Robots, drones, autonomous vehicles and intelligent automation are moving into the real world. They need power, compute, communications and trust at the edge, linked across thousands of sites. That layer does not exist yet. CUBICAI is building it: self-powered Cubes with Zero-Touch management built in, and TrueNorth Trust Servers that run in our Cubes or in your own racks. Every site joins one distributed mesh, starting with the critical sites that cannot go down.
The missing layer: infrastructure for Physical AI
Without a new kind of infrastructure, Physical AI will not work at scale.
Physical AI is AI that senses, decides and acts in the physical world: the robots, drones and autonomous vehicles in the news, and the automation that increasingly runs grids, water systems, ports and factories. Investment is surging, and it is moving out of the data centre to the places where the work happens.
Data centres and the cloud were built for the internet, not for machines in the field. Physical AI needs power, compute, communications and governance at thousands of edge sites, working when the grid or the link fails and coordinated as one network. That missing layer is what we build, whether it arrives in a Cube or as a Trust Server in a customer’s existing rack.
The Physical AI revolution needs a physical layer: a mesh of self-managing, trusted edge sites. Every substation, pump station, port and factory becomes a node that cannot go down.
When automated systems fail, the cost is physical.
None of these involved CubicAI, and no single product would have prevented them all. Each shows why critical sites need local resilience, a clear rule for what machines may do, and a record of what they did.
Ukraine grid attacks
Attackers used stolen credentials to open breakers remotely. About 225,000 customers lost power for up to six hours, and operators restored service by hand.
Colonial Pipeline
Ransomware entered through an old VPN account without multi-factor authentication. The company shut the pipeline for five days as a precaution.
Rogers outage
A network change took down service for more than 12 million Canadian customers for about 26 hours. Many could not reach 911.
Cruise robotaxi
A driverless car did not detect a pedestrian beneath it and dragged her about 6 metres. An incomplete report to regulators led to a US$500,000 federal penalty.
CrowdStrike update
A faulty update from a trusted security vendor crashed about 8.5 million Windows devices worldwide.
Iberian blackout
Spain and Portugal lost power. The expert panel found many interacting factors, including gaps in voltage control.
Sources: E-ISAC and SANS (2016); Colonial Pipeline testimony to the US Senate (2021); Xona Partners report for the CRTC (2024); California DMV and US Department of Justice (2023, 2024); Microsoft and CrowdStrike (2024); ENTSO-E expert panel final report (2026).
Three industries are converging into one.
CubicAI sits at the point where energy, autonomous infrastructure and AI-driven operations meet: a single platform where each reinforces the others instead of being stitched together by integrators.
Energy
Storage, power conversion, grid services and VPP participation: the resilient power layer every edge site depends on.
Autonomous Infrastructure
Self-configuring, forklift-portable cubes that commission themselves on arrival and move between sites without a re-engineering project.
AI Operations & Maintenance
Physics-informed models that predict, self-heal and optimise, with performance managed in software, not service contracts.
Two ways in. One new infrastructure.
Where a site has nothing, the Cube arrives with power, compute, communications, Zero-Touch management and a Trust Server built in. Where a site already has a server room, the Trust Server goes into the customer’s own rack. Either way, the site becomes a node in the same mesh. The Cube hosts; the Trust Server decides.
The Cube family
Transportable, self-powered edge nodes: lithium iron phosphate storage, bidirectional power conversion, AI compute and private wireless in one ruggedised enclosure, designed for −40 °C to +50 °C.
- For yards, ports, streets, remote pump and lift stations, and off-grid sites
- For sites short of grid capacity
- Keeps the site sensing and computing through outages
TrueNorth Trust Server
TrueNorth, our sensor fusion and policy software, packaged as containers on one specified server. AI agents propose; a deterministic controller decides; every decision is recorded and can be replayed.
- Runs inside a Cube, in a customer’s own rack or data centre, or in a kiosk
- Sovereign by design: data is sensed, fused and governed on site
- The same software in every segment we serve
Cube figures are target specifications. The Trust Server is a product in development.
Sense. Propose. Dispose.
AI is good at suggesting what to do. Critical infrastructure needs certainty about what is allowed. TrueNorth separates the two: AI and optimisers propose, and a deterministic Policy Broker decides.
Sensor fusion
Cameras, environmental and acoustic sensors, SCADA and IoT telemetry fused on site into one governed picture of what is happening, without sending raw data off site.
AI and optimisers
Agents, forecasting models, fleet managers and operators suggest actions: dispatch the battery, release a video clip, reroute a robot.
Deterministic Policy Broker
Each proposal is checked against signed site rules, safety envelopes and who is asking. Default deny. Deterministic by design: the same inputs give the same decision.
Trust you can audit
A tamper-evident, replayable record of what was allowed, what was blocked and why, kept on site.
Inspectable
A rules engine, not a neural network or an agent, so it can be tested and submitted for accreditation.
On site
Every request authenticated and decided locally, from cached signed policy, with or without a link.
Labelled
Release decisions built toward NATO STANAG 4774 and 4778 labelling and attribute-based access control.
Outside software
Trips, fire and emergency stops stay hardwired, so software is designed never to block or delay them.
A new kind of infrastructure: one mesh across everyone’s hardware.
Cubes bring power, compute and communications where they are missing. Trust Servers in Cubes and in existing buildings, ours or other operators’, are designed to join them into one mesh, under one set of rules. That is something single-vendor systems are not designed to offer.
The trust mesh is design intent, to be proven in research and field trials from 2027.
Same building blocks. Three outcomes.
Critical Infrastructure and Edge AI Supercompute apply in Canada, Japan and Europe alike. Industrial Autonomous Systems brings the trust mesh to factories, warehouses and yards.
Critical Infrastructure
Trust, resilience, security, safety and sovereignty for cities, utilities, ports and defence. Trust Servers in plants and control buildings; Cubes at remote pump, lift and field sites; one mesh across both.
Edge AI Supercompute
Extending sovereign AI compute securely to the edge: clusters of Cubes and partner GPU racks on megawatt-scale sites, placed where the power is and able to flex with the grid.
Industrial Autonomous Systems
A neutral referee for many vendors’ robots and automation in one continuous workflow. The optimiser proposes; TrueNorth disposes. Trust Servers in customers’ data centres; Cubes in yards.
One family of cubes. Three tiers of AI.
All cubes share the Smart Base: designed for IP55, −40 °C to +50 °C and forklift portability. Connect two to four into a cluster to pool storage, power and edge compute. Figures are target specifications.
eV Cube
Dual-port DC fast charging for fleet depots and mobile charging applications.
Flex Cube
Pure storage configuration for cluster capacity, battery swap and grid-services buffering.
Edge Cube
24U of AI-ready rackspace for partner GPUs and private 4G or 5G, indoor or outdoor. A natural home for a Trust Server.
Power Cube
PCS for solar and grid plus genset backup for critical resilience and black-start capability.
The same intelligence, at every scale.
One AI-driven software platform manages cubes, clusters and mesh networks alike, pooling distributed hardware into one software-defined estate.
Cube
A self-contained, AI-managed unit: energy, compute and inference in one forklift-portable enclosure.
Cluster
Two to four cubes sharing storage, power and edge compute over a common bus: pooled capacity, orchestrated as one.
Mesh Network
Cubes and stand-alone Trust Servers federated across sites and operators: one trust mesh, designed to avoid a single point of failure.
Virtual Infrastructure
Highly distributed hardware presented as one software-defined platform: power, compute and AI where they are needed, under one set of rules.
Zero-Touch AI is the advantage.
Our edge is designed around deep device- and system-level AI models, the foundation of the whole platform. Four Zero-Touch pillars are designed to run natively on every cube, each taking over work operators now buy from system integrators.
Resilience
Keeps the site running when the grid or the network fails: islanded operation, self-healing failover, and physics-informed models designed to predict failures weeks ahead, with onboard thermal-runaway monitoring.
Security & Sovereignty
Zero trust on every cube: each request authenticated and decided locally against signed policy, and data sensed, fused and governed on site under local law. Built toward IEC 62443 security practice and NATO STANAG 4774 and 4778 data labelling.
Logistics & Mobility
Positioning, battery swapping between cubes and automated delivery scheduling, so the fleet reconfigures as demand moves.
Energy · Heat · Compute
Triple optimisation: workloads scheduled against battery state of charge, thermal envelope and grid signals at once, with every dispatch admitted by the Trust Server.
Metrics are prototype and design figures from CubicAI’s own work. Field results will be reported from 2027.
Zero-Touch is the whole point.
Every cube is designed to ship with the full suite running, so an integrator-ready cube is a complete answer, not a component a consultant configures at every site.
Sovereign AI, delivered where it matters.
We aim to work with utilities and telecoms to deliver sovereign edge AI, resilient energy and private communications: a layered stack from site to application.
Customer AI Training Cloud
Model training, complex reasoning, APIs.
Enterprise AI apps, hosted at the edge
Run third-party AI applications on-site.
TrueNorth Trust Server
Sensor fusion, policy and the decision record, with or without a Cube.
CubicAI Orchestration
Energy, compute and network coordination.
CubicAI Edge Cubes
Power, energy, edge and eV cubes: the site substrate.
Where uptime is non-negotiable.
Verticals where offline autonomy, governed automation and integrated power and compute are requirements, not nice-to-haves.
Applied Physical AI and Electrification for Leaders.
A programme for boards, executives and senior managers in utilities, cities, ports and critical infrastructure: what Physical AI can do, where it fails, and how to govern it, taught by people who have built and run these systems.
Executive briefing
A half-day session: the Physical AI landscape, lessons from real incidents, a live tabletop exercise, and the questions to ask vendors and your own teams.
Applied workshop
Two days applying Physical AI to your own sites: use cases, decision rights, security and a readiness assessment.
Leadership certificate
A longer programme for leaders responsible for Physical AI and electrification across their organisation.
Designed to work with technology from
Let's put a cube on the ground.
Pilot programmes, research collaborations, Physical AI education for leaders, technology integration and channel partnerships, for organisations that run critical sites.
Book a discovery call →Discovery Call
A short discussion of your requirements and use cases.
Technical Deep Dive
Architecture, integration and deployment planning.
Pilot Proposal
A custom deployment plan with clear success metrics.
Agreement
Finalise terms and agree support arrangements.