Nanoprecise Sci Corp

Industrial maintenance has historically run on one of two schedules, and both waste money. Reactive maintenance waits for equipment to fail, which converts a cheap planned repair into an expensive unplanned outage with production losses attached. Preventive maintenance services equipment on a fixed calendar regardless of condition, which means replacing components that had years of life left and occasionally still missing the failure that arrives early.

Condition-based maintenance is the alternative, and it depends on detecting the signature of a developing fault. A bearing beginning to degrade changes its vibration profile long before it seizes. A misaligned shaft, a failing lubricant film and an unbalanced rotor all produce characteristic patterns. Nanoprecise pairs wireless sensors with machine learning models that identify those signatures and estimate how much operating life remains, giving the operator time to schedule the work.

The company raised $52.1 million in 2025 to expand that platform. Its natural market is industries running large fleets of rotating equipment in locations where sending a technician to inspect something is itself expensive, which describes Alberta’s energy and processing sector almost exactly, along with mining, pulp and paper, and utilities worldwide.

This is a good example of a category Alberta is unusually well placed for. Industrial AI requires two things that rarely coexist: machine learning capability and deep familiarity with the physical equipment being modelled. Alberta has an AI research base in Edmonton and one of the continent’s densest concentrations of heavy rotating equipment within a few hours’ drive of it.

The deployment constraints shape the product more than the algorithms do. A sensor installed on a pump in a remote field has no mains power and often no reliable network, so it has to run for years on a battery and transmit sparingly. That forces analysis onto the device itself: rather than streaming raw vibration data to a server, the sensor processes locally and sends a small result. Edge computing is a fashionable phrase in most contexts and a hard physical requirement in this one.

Selling into industrial operations carries its own difficulty. Maintenance teams have justified scepticism about software promising to predict failures, having watched previous generations of condition monitoring generate false alarms until everyone stopped listening. Trust is rebuilt one correct call at a time, and the metric that matters commercially is not detection accuracy in the abstract but the ratio of true warnings to false ones, because a system that cries wolf is worse than no system at all.

Hazardous-area certification is a further barrier that favours patience and penalises shortcuts. Equipment installed where flammable gas may be present has to be certified not to provide an ignition source, and that certification is expensive, slow and specific to each jurisdiction. It is also the reason a competitor cannot simply ship a cheaper sensor into this market.

The category has a tailwind unrelated to technology. Experienced maintenance personnel are retiring faster than they are being replaced across heavy industry, taking with them the tacit judgement that let them hear a bearing going bad from across a plant. Instrumentation that captures some of that judgement is not replacing people so much as compensating for people who have already gone.

The value calculation is unusually easy to make in this category, which is a considerable commercial advantage. Unplanned downtime at a processing facility has a known hourly cost, often a very large one, and a failure that halts production has consequences extending through the supply chain. Against that, a monitoring system that prevents a single significant outage in a year has paid for itself many times over. Buyers can run the arithmetic themselves, which removes the need to sell a vision.

The technical difficulty sits in the gap between detecting an anomaly and diagnosing a fault. Identifying that a machine has changed is comparatively straightforward. Determining which component is degrading, how quickly, and how much operating life remains requires models grounded in the physics of the specific machine type, not merely statistical deviation from a baseline. That domain knowledge is what separates a useful system from an alarm generator, and it is accumulated slowly through observed failures.

Data accumulation therefore compounds into a defensible position. Every failure a deployed fleet observes improves the models for every other customer running similar equipment, and a company with years of observed failures across many sites can diagnose faults that a newcomer with better algorithms and no history simply cannot. In machine health monitoring the moat is the dataset rather than the mathematics.

The applicable market is broader than Alberta by a wide margin. Rotating equipment is everywhere heavy industry exists: mining, pulp and paper, water utilities, power generation, chemicals, shipping and manufacturing worldwide. A company that proves its technology on Alberta oil and gas assets has validated it in one of the most demanding environments available, in terms of remoteness, temperature range and consequence of failure, which travels well as a reference into every other market.

All startups →Latest news →