Huburb Radar

AI Agents

AI agents are software systems that use a language model to plan a sequence of steps and then take actions — calling tools, writing files, querying databases — instead of only producing text.

Foundation Models

Large models trained on broad data that can be adapted to many downstream tasks rather than being built for one purpose.

Inference Optimisation

The set of techniques — quantisation, batching, caching, distillation and specialised silicon — that reduce the cost and latency of running a trained model.

Liquid Cooling

Moving heat away from servers using liquid — either cold plates attached to chips or full immersion in a dielectric fluid — instead of moving air.

Humanoid Robots

General-purpose robots with a roughly human form factor, intended to work in spaces built for people rather than in purpose-built cells.

Autonomous Vehicles

Vehicles that perceive their surroundings and drive without a human operator, ranging from robotaxis to yard trucks and delivery pods.

Small Modular Reactors

Nuclear reactors in the roughly 50–300 MW range, built from factory-produced modules rather than assembled entirely on site.

Long-Duration Storage

Storage that discharges for ten hours or more — iron-air, flow batteries, thermal and mechanical systems — as opposed to the four-hour lithium standard.

Direct Air Capture

Machinery that pulls carbon dioxide directly out of ambient air for permanent storage or industrial use.

Computational Protein Design

Using machine learning models to predict protein structure and to design new proteins with intended functions.

Earth Observation

Satellite constellations that image and sense the planet continuously, across optical, radar and hyperspectral bands.

Smart Glasses

Eyewear with cameras, microphones, speakers and sometimes a display, used as a hands-free interface to an AI assistant.

Northwind Agents

Agent platform for back-office workflows with human approval built in.

Toronto, Canada · Founded 2023

Cirrus Inference

Inference serving stack aimed at cost per token rather than peak throughput.

Austin, United States · Founded 2022

Thermafold

Direct-to-chip cooling kits designed for retrofit rather than new builds.

Rotterdam, Netherlands · Founded 2022

Meridian Atomics

Factory-built small modular reactors for industrial sites.

Ottawa, Canada · Founded 2020

Ferrovolt

Multi-day storage using abundant, low-cost materials.

Perth, Australia · Founded 2021

Aerelith Carbon

Direct air capture co-located with geothermal heat and storage.

Reykjavík, Iceland · Founded 2020

Atlasworks Robotics

General-purpose robots for handling tasks in existing warehouses.

Munich, Germany · Founded 2022

Orbital Cartography

Radar imaging constellation with an analytics-first product.

Glasgow, United Kingdom · Founded 2020

Halcyon Quanta

Error-correction-first quantum hardware programme.

Delft, Netherlands · Founded 2018

Helix Foundry

Designed proteins with automated wet-lab validation in the loop.

Boston, United States · Founded 2021

Veritas Ops

Monitoring and containment for autonomous software agents.

Tel Aviv, Israel · Founded 2023

Compute meets the grid

AI capacity planning has become an electricity problem. Interconnection queues, transformer lead times and cooling water now determine build schedules more than chip supply does.

Agents need governance before autonomy

The constraint on deploying AI agents in serious workflows is not capability but accountability. Audit trails, scoped permissions and approval steps are becoming the product.

Inference, not training, is the cost line

As AI products reach volume, attention is shifting from training runs to the recurring cost of serving. Routing, compression and purpose-built silicon are where margin is won.

General-purpose robots are starting narrow

Humanoid platforms are being sold as general labour but deployed on single repetitive tasks, because that is where data, safety cases and payback periods are tractable.

A premium is returning for firm power

Buyers with round-the-clock load are paying for generation they can schedule. That is pulling capital toward nuclear, geothermal and multi-day storage rather than additional intermittent capacity.

Measurement is becoming the product

Satellites, sensors and models are converging on a market for verified answers — emissions, crop yield, construction progress — rather than raw data feeds.