Wear alerts for the pumps and motors outside a reactor's safety systems
TL;DR Bearingly fits vibration and temperature sensors to the cooling-water pumps, fans and motors that keep a nuclear station running but sit outside its safety systems. Its models flag wear weeks before a trip. Plant owners pay a yearly fee per monitored machine, starting with the industrial firms now funding captive small reactors.
India's nuclear fleet is small and old in parts - 24 reactors with 8,780 MWe across seven stations, the oldest at Tarapur[1]. That's 1.58% of national capacity, against a goal of 100 GW by 2047[2]. Getting there means many more units, and every unplanned outage at a baseload plant costs a bomb.
Funnily enough, a lot of what trips a station is ordinary machinery. Pumps, fans, motors and turbine auxiliaries wear out the same way they do in a cement plant. Research on fault detection in nuclear units shows the method works, but it also notes that raw plant data can't be shared between facilities and that the datasets aren't public[3]. So the bottleneck is data access, not modelling.
What it does, in regulatory terms: advisory condition monitoring of non-safety equipment only. Bearingly never connects to reactor protection, control or any safety-classified instrumentation. Its sensors are standalone and read-only, and a human planner decides every action. Anything closer to safety systems needs AERB qualification, which is out of scope.
I've reframed the seed, which pitched real-time predictive maintenance aimed at plant safety risks. Under the SHANTI Act the AERB is now a statutory regulator with licensing and inspection powers[4], and no young company will get an AI model into a reactor's safety loop. The useful version stays on the conventional side.
The kit is wireless vibration and temperature sensors on rotating equipment, an on-site gateway with no internet link, and models that rank machines by how fast their condition is moving. Engineers get a weekly list of which bearing to look at, how confident the model is and what the data showed. Every site keeps its own data on its own servers.
The buyer is the maintenance head at a plant, and the budget is the station's maintenance spend. The new captive owners matter most. Under NPCIL's small-reactor model, the industrial user pays all capital and operating costs for the plant's life, while NPCIL runs it[5]. That gives a big industrial group a reason to want fewer outages. Nobody publishes a price for nuclear condition-monitoring contracts in India, so per-machine annual fees are my working assumption.
The SHANTI Act received assent on 20 December 2025 and lets companies hold licences to build and run plants[4]. Draft rules opened for comment in August 2026[6]. Separately, large private conglomerates, BHEL and Indian Railways showed interest in captive Bharat Small Reactors, sending 687 queries on finance, tax and ownership[7]. New owners will pick their maintenance vendors early.
There is no public figure for spending on nuclear plant condition monitoring in India. The firmest numbers are the installed base, 24 operating reactors[1], and the list of large companies now studying captive units[7]. Treat each captive project as a possible anchor customer, not a market share figure.
One US nuclear services firm sells simulation, engineering and data-validation services to nuclear plants and claims work on 173 nuclear simulators[8]. NPCIL also runs its own maintenance teams and would build in-house tools if it chose. Indian industrial predictive-maintenance firms serve steel and cement, but I found none with a public nuclear reference.
The gap is a vendor that already passes a nuclear owner's security review.
| Difficulty | Hard (63/100) |
| Starting capital | ₹2–3.6 Cr |
| Time to launch | 12–18 months |
The sensors are standard. The hard part is getting in the door, which means vendor qualification with NPCIL, security clearance for staff on site and a cyber design that a nuclear operator will accept. Expect the first plant to take more than a year before any paid data flows. If nobody on your team has nuclear operating experience then you'll struggle.
You don't need a nuclear licence while you stay off safety systems, but every plant will make you prove that boundary.
Building this outside India? The categories above usually have a local equivalent, so check with your own regulator.
Never let the model decide maintenance or touch anything safety-classified. It advises on non-safety equipment, and a human planner makes every call.
Want to work out the AI layer for Bearingly? Talk to Yash →
I'd only try this with a team that includes former nuclear plant maintenance engineers and is willing to sell into steel and cement first to fund the wait. For a pure AI team it is a poor fit. Prove one thing before raising - a written trial approval from one nuclear site or captive developer.
Plan for roughly ₹2–3.6 Cr, mostly for sensors, gateways and the long trial cycle before a nuclear site signs.
Not a nuclear licence, as long as you stay off safety systems. You need written confirmation of that boundary, NPCIL's supplier approval and WPC type approval for wireless sensors.
Yes, but most teams start with conventional industry and add nuclear later. Each country's nuclear regulator sets its own rules for vendors on site.
The free Bearingly validation kit has the line-by-line capital breakdown, the effort scoring, the rules and clearances checklist, a 14-day validation sprint with pass/fail targets, a 90-day launch plan, 10 interview questions for plant maintenance heads, and 5 assumptions to test. Download the kit →
Thinking about building Bearingly or something close to it? The make-or-break step is getting past a nuclear operator's vendor and cyber review for the first site, and that's the kind of slow, fiddly problem I help founders plan for. If you need a hand, you know where to find me. Talk to Yash →
For illustration only. Names are fictional and ideas hypothetical; sourced facts are real but may be out of date. Nothing here is legal, financial or professional advice, and Yash Khatri is not personally liable for any use of it. Read the full disclaimer.