Video
Heavy industry still checks its most critical equipment by sending people inside it. We build robots that go in instead, from parts we reuse, so every robot is cheaper and faster to build than the last.
Every year, a coal power plant shuts each boiler down for an overhaul. Before anyone can check the thousands of tubes inside, a contractor builds scaffolding up the inside of the furnace, a structure dozens of metres tall. Putting it up and taking it down takes about ten days, and the plant earns nothing while it stands.
Once the scaffolding is up, about a hundred people work inside, many of them hanging at height in harnesses, taking spot readings by hand. Between them they read less than 1% of the tubes.
In November 2025 we spent a boiler's entire month-long overhaul at MEIL's Anpara plant in Uttar Pradesh, watching every section being inspected. The boiler went back into service on 5 December, inspected exactly as the plant's procedure requires. Eight days later, a tube burst.
A burst tube takes a unit offline for about a week. At ₹4 Cr a day plus repairs, that is roughly ₹25 Cr (at a large plant like Tata Power Mundra, a cold restart adds about ₹8 Cr more, mostly diesel), for one tube nobody read. The same pattern runs through steel furnaces, refineries, cement plants and nuclear stations: people inside, spot checks, and no tube-by-tube history.
Inspection robots exist. Gecko, a US company now valued at US$1.25B, sends magnetic crawlers up open boiler walls, and drones fly through open spaces. But a boiler is not one space. It is several, each with its own way in:
A robot that covers three of these sections saves nothing, because the scaffolding stays up for the fourth. The plant gets its days back only when the whole boiler is covered.
So we don't build one robot that does everything. Every inspection robot needs a way to sense, a way to reach and a way to move or hang inside the asset. We build each of these once, as a module, and combine them into a robot for each section. A new section needs a new combination, not a new robot.
| Module | What it does | Status |
|---|---|---|
| Sensor stack | EMAT wall thickness, IR time-of-flight camera, lidar and 3D mapping | Built, pilot underway |
| Linear slide link | Carries the sensor head across a workspace | Built, pilot underway |
| Cable suspension | Skyclimber-type platform for tall, hanging sections | Built, pilot underway |
| Magnetic crawler | Carries the arm across tube banks | In development |
| Two-link magnetic crawler | Moves through pipes and headers | Two prototypes built |
| Caged drone | Flies inside open volumes | Planned, next year |
Sensor stack + linear slide link, with its own second link and joint.
Sensor stack + linear slide link, on the magnetic crawler.
Sensor stack + linear slide link, on the cable suspension.
| Section | Robot | Customer | Pilot |
|---|---|---|---|
| Radiant tubes, annealing line | Radiant tube platform | JCAPCPL | Nov 2026 |
| Economiser coils | Economiser crawler | Adani Power | Dec 2026 |
| Waterwall, through soot-blower port | Waterwall arm | Tata Power, Mundra | Feb 2027 |
| Header | Two-link magnetic crawler | Tata Power, Mundra | In discussion |
Tata Power and JCAPCPL have issued purchase orders for their pilots. JCAPCPL came on Tata Power's referral and issued its order 1.5 months after the first call. The radiant tubes in a steel annealing line have the same access problem as the pendant coils at the top of a power boiler, so a robot built for a steel plant inspects part of a power plant too.
Each pilot covers one section, so on its own it won't shorten a shutdown. We aim to cover every major section of a coal-fired boiler within two years.
For the plant, the value is the ten days it gets back. We charge a fifth of it.
India has about 630 utility coal boilers and the world about 6,500, and the fleet is still growing: India plans to go from 224 GW of coal to 283 GW by 2032. The same modules then carry into the rest of heavy industry's confined spaces: refinery heaters, steel furnaces, heat-recovery steam generators, storage tanks, ship tanks and cement preheaters.
A US$50M-a-year business needs 69 boiler overhauls, 1.3% of the world's 5,200 a year, at our global price of US$729k each. Gecko built a US$1.25B company on the open surfaces; the congested sections are still unserved.
JCAPCPL, Adani Power and Tata Power, section by section.
Thermal suite commercially ready and certified.
Whole-boiler inspection at 5 plants: ₹45 Cr (~US$5M).
Once a plant gives one vendor the entire boiler, switching means requalifying every section at once.
The same tubes, measured at every shutdown. Wear rates and remaining life come from that history; a newcomer does not have it.
Every section we prove is one a competitor still has to build, then wait for a plant's shutdown to test.
Build simple modules and put each to work. Combine them until we take on the whole asset. Use the data to predict failures before they happen.
Outage day: 500 MW × 24 h × 83% PLF × ₹4/kWh ≈ ₹4 Cr. Scaffold days, outage length and costs from plant operators. Global figures at a US$243k fleet-average outage day. FX ₹96.8 / US$. Full workings in the Market Analysis and Economics documents.
One boiler overhaul costs ₹50 L to inspect today and is worth ₹40 Cr in saved outage days. Across ~6,500 coal boilers and ~800k confined industrial assets, that is US$30.7B a year at stake.
FX ₹96.8 / US$ (RBI reference rate, 8 Oct 2026). Global figures are USD-native.
Boilers cause at least 60% of outages; tube leaks cause at least 75% of boiler outages.
workers killed in confined spaces in the US, 2011–18. About 60% of them died trying to rescue someone else (BLS / NIOSH).
500 MW unit, one annual overhaul, India.
We price against outage days, not UT crew rates. That turns a ₹50 L line item into a ₹8 Cr ticket.
India: 630 utility boilers + 1,500 industrial; global: 6,500 utility + 30,000 industrial. Tube leaks = boilers × 0.8–1.0 leaks/yr × 6 days × outage-day value. Scaffold = overhauls × 10 days × outage-day value. Outage day: India ₹4 Cr (operator average); global US$243k.
Each new asset class reuses the EMAT link and one of three mobility platforms already built.
India confined space: ₹240 Cr spend + ₹528 Cr value = ₹768 Cr (US$79M), bottom-up across 7 asset classes. Refining 258 → 310 MTPA by 2030 (MoPNG); cement ~700 → 915–925 MTPA by 2030 (CMA; CRISIL).
Inspection data decides where maintenance and integrity money is spent. We own that data in the zones nobody else can reach, then sell into the US$25–31B integrity budget.
India pricing is 20% of value created. Outside India, customers also pay a premium for taking people out of the confined space.
Series D US$125M at unicorn valuation (Jun 2025); ~US$350M raised; >500k assets on platform; FY23 revenue est. US$34.5M.
Series B (Dec 2025); ~US$20M raised.
Series B (2026).
Gecko built a unicorn on open steel surfaces. The congested zones, through soot-blower ports, between economiser coils, into platen and radiant sections, are still unserved.
| Assumption | Value (low / base / high) | Basis |
|---|---|---|
| India utility coal/lignite units | 610 / 630 / 650 | 603 units in CEA CO2 database FY23 + new 660/800 MW units |
| Overhauls per boiler per year (India) | 1.0 (minimum) | Abhiyantr customer data |
| Global coal units | 6,000 / 6,500 / 7,000 | ~2,200 GW ÷ ~340 MW average (GEM) |
| India access + NDT per overhaul | ₹20 L / ₹50 L / ₹1 Cr | Genco tenders 2024–26: scaffold ₹13–75 L, UT/NDT ₹9–46 L |
| Global access + NDT per overhaul | US$50k / 120k / 250k | Gecko robotic price US$50–100k + scaffolding |
| Tube leaks per unit per year (India) | 0.5 / 0.8 / 1.2 | NTPC 692 leaks FY22–Sep 2025 across ~170 units |
| Days lost per tube leak | 4 / 6 / 7 | Operator data; forced outages run up to a week |
| Outage day (India) | ₹4 Cr (₹4/kWh, 83% PLF, 500 MW) | Operator data |
| Scaffold days per overhaul | 10 (erect + dismantle), all on critical path | Operator data; to be confirmed in pilots |
| Global coal CF / power price | 50% / US$60 per MWh | Fleet average |
| Confined-space asset counts | See section 04 | Refineries (OGJ), cement plants (Global Cement), fleet (UNCTAD); tanks/vessels estimated |
Five industries, one pattern: critical assets inspected by hand, from scaffolding or on wires, with no tube-by-tube history.
All figures use ₹96.8 = $1 (RBI reference rate, 8 Oct 2026). Conservative assumptions throughout.
Automation attacks the four biggest costs of a boiler shutdown: scaffolding, people, downtime and missed defects. Then the data keeps paying.
bought before each shutdown, because boiler condition is unknown until it opens. Tube-level data, and an AI model that marks probable defect areas ahead of time, let plants order only what will fail.
~90% of power-plant MRO inventory is used less than once a month; carrying cost is 20–25% of inventory value a year. Power Engineering
of conventional readings needed under risk-based inspection at Equistar's ethylene plant, saving $500K+ upfront and halving vessels opened per turnaround. Full coverage moves plants from calendar to risk-based, reactive to predictive. AFPM
saved in one case by safely delaying a replacement. Better inspection means better maintenance; on multi-crore assets, even one extra year of life is worth a great deal. Vysus Group
Thinned, fouled or leaking tubes reduce heat transfer. Early detection keeps the boiler near design efficiency between shutdowns. Oxmaint
uptime from predictive maintenance, with maintenance cost −5–10% and planning time −20–50% (Deloitte); maintenance cost −10–40% in high-asset industries (McKinsey).
Every inspection adds to the asset's condition history: the basis for predictive maintenance, RLA and regulator reporting. Pooled across plants, it becomes an industry benchmark.
We are still in price discovery. The working model is outcome-based: once the suite is mature, we charge 20% of the value from reduced downtime.
₹4 Cr a day; customer saves ₹40 Cr.
₹10 Cr a day; customer saves ₹100 Cr.
Pre-2028: pilot fees only, kept minimal; the purpose is lock-in, not profit. A first ₹1 Cr is possible from pilot cheques by end '27, but is not counted on.
How we size the savings. Current inspection timelines come from customer conversations, cross-checked against our own scaffolding and shutdown estimates for each asset.
Future vision. Today we price at the plant level, on downtime alone. As conversations move to corporate and national headquarters, our full value comes into play: asset life, fleet-wide planning and risk reduction. Pricing will scale with that.
Thermal first. Private plants first. Pilots, then perfect the product, then sell net-net inspection: ₹45 Cr ($5M) in the first year of commercial operations.
Radiant tube (suspension + arm), economiser, in-pipe, soot blower, hanging pendant; then header + drone.
Exit: paid pilots → customer lock-inProduct-heavy year: mechanics, software, AI. In-house test rigs for customer demos. Pilots continue; international discovery starts.
Exit: thermal suite ready + certifiedPer-part and net-net inspection. Offer per-part early, push to net-net.
Exit: 5 net-net plants · ₹45 CrDiscovery only, no selling yet.
India and international, until end-to-end inspection is perfected.
Not decided yet. Thermal is the multi-year focus.
Other robots inspect a surface. We inspect the whole asset. Only full coverage takes the scaffolding down, and the players who could build that either cannot reach the congested sections or have no reason to.
| Player | Open surfaces waterwalls, pipes, tanks | Open volumes | Submerged | Congested sections between coils, headers |
|---|---|---|---|---|
| Gecko RoboticsMagnetic crawlers + data platform, as a service | ||||
| Flyability Elios 3Caged drone, sold as equipment | ||||
| Detect TechnologiesDrone + monitoring software | ||||
| PlanysUnderwater vehicles | ||||
| Utility R&D, e.g. NTPC NETRAWaterwall crawler, one fleet | ||||
| OEMs and contractorsManual, from scaffolding | Wherever people can safely stand | |||
| Abhiyantr | The whole asset, section by section | |||
No Indian customers named; largest overseas customer is ADNOC.
Cannot fly between coils or into headers. Sold as equipment: the plant still does the inspection.
NTPC's waterwall crawler patent was filed in 2011. Plants still inspect from scaffolding.
A new team starts at month zero, and every robot it builds still has to wait for a plant's scheduled shutdown.
None of the reasons is about capability. Each group is more likely to become a partner, customer or acquirer than a competitor.
Their economics depend on one platform deployed across large areas. Each congested section is a small, separate engineering problem, and Indian plants are not where they are growing.
Partner or acquirerInspection is an O&M cost bought as a service. An in-house team learns from one fleet: Adani Power's 18.3 GW is 8% of India's 224 GW coal fleet.
Customer, strategic investor, referral sourceScaffolding and manpower are 80% of what they bill for a shutdown. Robotic inspection removes those days.
Subcontract inspection to usBuilt for pipelines and monitoring, or underwater structures. Boiler internals mean starting the module library and the shutdown cycle from zero.
Adjacent playersOnce a plant gives one vendor the entire boiler, switching means requalifying every section at once.
The same position-tagged tubes, measured at every shutdown. Wear rates and remaining life come from that history; a new vendor does not have it.
Every section we prove is one a competitor still has to build, then wait for a shutdown to validate.
Three robots, one core. Each new robot starts from four proven modules: sensor stack, linear slide link, magnetic crawler and cable suspension.
| Module | Waterwall arm | Economiser crawler | Radiant tube platform | BOM |
|---|---|---|---|---|
| Sensor stackEMAT + vision · shared across pilots | ||||
| Linear slide linkShared across pilots | ||||
| Magnetic crawlerReusable module | ||||
| Cable suspension systemReusable module | ||||
| Second link + insertion jointRobot-specific |
Sensor stack + linear slide link, with its own second link and insertion joint.
Sensor stack + linear slide link, on the reusable magnetic crawler.
Sensor stack + linear slide link, on the reusable cable suspension.
Next robots start from the same four modules, so every new section costs less engineering than the last.