Sunday Essay — The Diesel Ledger on India's Towers
India's mobile towers have been chasing diesel-free operations since 2011; fifteen years on, Indus Towers has solar on fewer than a fifth of its roughly 270,000 sites and still loses money on every unit it re-bills to operators. The AI deployed across the estate is primarily a bookkeeping technology — fuel sensors, smart meters, automated contractor photo checks — that makes diesel auditable rather than scarce.
What is a mobile tower in western Uttar Pradesh supposed to do when the grid feeder drops in the middle of a weekday afternoon, the battery bank was sized for an outage half as long, and the pump down the road has been told to stop filling jerrycans? Every answer the industry gives this year involves software. Most of them, read carefully, involve something older first.
when the feeder drops
The jerrycan detail is real. On 13 April 2026 the Digital Infrastructure Providers Association, the lobby for India's tower companies, told the Department of Telecommunications and state authorities that outages of four to six hours a day had spread across several states, as TelecomTalk reported from an Economic Times story the same day. Districts in Uttar Pradesh, Maharashtra and Madhya Pradesh had restricted diesel sales in containers during the fuel scare that followed the West Asia conflict, and some fuel stations had stopped serving telecom refuelling crews. DIPA said diesel is needed "on a routine basis to run DG sets at mobile tower sites." I could not find DIPA's own filing, so treat the hours figure as the association's claim relayed by the trade press. It is still the clearest statement I have seen of what a tower runs on when things go wrong.
The site's options are few. It drains its batteries. It starts the generator if there is fuel in the tank and the tank has not been siphoned. It runs on solar if someone bolted panels to the shelter roof and the sun is out. Or it goes dark, and the operators renting space on it lose every call and data session in the cell.
So the honest first answer to the opening question is hardware: a bigger battery, a panel, a full tank. Software decides how well that hardware is used and, more importantly in India, whether anyone can prove what happened to it. That second job is where the AI in this story earns its keep, and the industry is underselling it by calling it energy optimisation.
what fifteen years of promises bought
This argument is older than the current cycle, which is a reason to take it slowly.
In July 2011 the GSMA's Green Power for Mobile programme published a report in which Indus Towers, then the world's largest tower company with more than 110,000 towers, described its solar-hybrid trials. The same report put Indian telecom's diesel consumption at about 2 billion litres a year, with roughly 60% of tower power coming from generators and only 40% from the grid. Three months earlier, on 12 April 2011, the Telecom Regulatory Authority of India had issued its green telecom recommendations: at least half of rural towers and a third of urban towers on hybrid power by 2015, all rural towers and half of urban towers by 2020.
Six years later, in August 2017, Indus announced that over 62,000 of its sites, about half its portfolio, had become "green sites", defined as sites that ran without diesel as power backup. It claimed 200 million litres of diesel saved over five years. Its chief executive, Bimal Dayal, said: "Our vision is to achieve diesel-free operations." No date was attached.
Now the present. On its 28 October 2026 earnings call for the quarter to 30 September, Indus reported about 270,500 macro towers and roughly 49,000 sites with solar, up from 259 MW of installed solar to 297 MW in a single quarter. A week later its chief executive, Prachur Sah, told Business Standard the company had passed 50,000 solar sites and aims for 100,000 by FY29, out of a base he put at around 280,000 towers.
Lay those numbers side by side and the shape is plain. The 2011 regulator wanted every rural tower on hybrid power by 2020. The 2017 company said half its sites ran without diesel backup. The 2026 company has panels on under a fifth of its towers and still has a diesel line in its cost base large enough to report an 11% cut as the headline of its energy discussion. The 2017 definition of a green site (no diesel backup, usually because the battery and grid could cover) is a different measure from the 2026 count of solar sites, and the tower base more than doubled in between. Even so, a vision of diesel-free operations stated nine years ago has become a target of 100,000 solar sites three years from now. Progress, yes. Slower and more conditional than any of the press releases implied at the time.
sensors before panels
What changed in the last three years is what the tower knows about itself.
Sah's list, across the call and the interview, is specific. Smart meters on approximately 140,000 sites over three years. Fuel sensors on generators that report tank levels in real time, so a refill can be calculated from the reading. About 250,000 sites transmitting live data. A fault on a solar site raises a repair ticket automatically. And AI image and video analytics used for "site validation, safety, maintenance, and supervision," which on the call he put more bluntly: "AI-led image analytics is strengthening governance, vendor accountability, and maintenance quality."
Take that last sentence as a field engineer would. Vendor accountability means the photograph a contractor uploads after a battery swap or a panel clean gets checked by a model before the invoice is paid. Governance means the fuel the company paid for can be matched against the fuel the sensor saw arrive. The 2011 GSMA report did not have a line on remote monitoring. In 2026 it is the backbone.
Here is my position, plainly. The AI on India's towers is first a bookkeeping technology, and that is its most valuable use. It makes diesel visible, it makes contractors provable, and it turns a dispersed estate of shelters into something a finance team can reconcile. Panels and lithium cells are cutting the litres. The software is making the cut auditable, and in a business where the energy bill is passed through to customers, auditable is worth more than a few extra percentage points of efficiency.
The diagram makes one point the prose can only gesture at. Every physical arrow into the site is a hardware purchase: a grid connection, a panel, a cell, a generator. The analytics sit on a side branch. They change nothing about where the electrons come from. They change who gets dispatched, which invoice is paid, and what the customer is billed.
the radio draws the load
The strongest technical objection to my framing comes from inside one of Indus's largest customers.
Bharti Airtel's case study on the GSMA site, undated but describing a live programme, says energy is about 29% of its annual network operating cost and the radio access network accounts for roughly 80% of that energy. Its answer is a vendor-agnostic AI tool that studies cell-level traffic and extends sleep windows when traffic is low, across around 200,000 sites, with an initial target of roughly ₹100 crore in annual savings and a longer-term potential of 8% to 10%. Put simply: if four-fifths of the energy is drawn by the radios, the real energy lever is a model that switches radios off, not one that checks a contractor's photographs.
The objection is fair on physics, and it should be weighed. Airtel's FY26 ESG disclosure shows how lopsided the energy picture is. The company reports 17,900 kilolitres of diesel burned in its own generators and 9.79 TWh consumed at third-party network sites, against 1.69 TWh of total energy in its own operations. Most of Airtel's energy is drawn at someone else's tower, by Airtel's radios. Trimming that draw by a tenth would matter.
Two things keep me from accepting the objection as a rebuttal. The first is evidence quality. The 8% to 10% is Airtel's own projection, and a comparable number, Ericsson's 10% to 12% RAN energy reduction across pilot sites from January 2023, comes from a vendor describing its own product on an unnamed operator's network. I would treat both as ceilings measured under favourable conditions until an operator publishes an audited before-and-after on a large, mixed estate. The second is that a sleeping cell still sits on a site that has to be powered when it wakes. A radio-side saving of a tenth does not change how many hours a site in a weak-grid district spends on battery or diesel; it changes how fast the battery drains. The two levers stack. They do not substitute, and the supply-side bookkeeping is the one that decides whether a given site stays up during an afternoon outage.
twenty-two point six hours, on average
The other objection is more basic, and it comes from the government.
On 18 March 2026 the Press Information Bureau published an overview of India's power sector stating that average daily supply in rural areas rose from 12.5 hours in FY14 to 22.6 hours in FY25, and in urban areas from 22.1 to 23.4 hours. The same document puts the national energy shortage at 0.03% through December 2025. If that is the grid, then the premise of this essay (that towers burn diesel because the feeder keeps dropping) looks out of date. Diesel would be a management failure, and fuel sensors would be fixing a theft problem.
I think the Ministry's numbers are true and incomplete, and the incompleteness matters for exactly this use. An average of 22.6 hours says nothing about how the missing 1.4 hours arrive: one clean break at night or ten flickers through the day. For a tower, that distinction is the whole game, because every interruption cycles the battery and every long one risks a generator start. The PIB document reports no interruption-frequency or duration indices of the kind utilities elsewhere publish as standard.
The best independent evidence I know points the other way. Prayas (Energy Group) in Pune ran its Electricity Supply Monitoring Initiative from 2015 with plug-in voltage monitors at about 400 locations across 23 states. In Uttar Pradesh, over 2016 to 2019, it found rural sites averaging more than 240 minutes of outage a day and up to ten outages daily, with about 37% of outages lasting over three hours. That data is old and the grid has improved since, as Prayas itself said. But it is the right kind of measurement, minute by minute, at the consumer end, and it describes the kind of supply DIPA was complaining about this April.
So the disagreement resolves into a measurement question. The Ministry measures hours supplied, averaged. A tower needs interruptions, counted.
whose savings
There is a commercial twist that explains why Indus talks about cost per unit rather than margin.
Tower companies in India re-bill energy to their tenants. On the October call the chief financial officer, Abhishek Maheshwari, reported an energy margin of minus 3.6% in the quarter, against minus 4.6% in the previous one. In other words, Indus still spends more on power and fuel than it recovers from operators for them. Sah told analysts he would focus "more on the cost initiatives and less discussion about the margin," and described the aim as making "a meaningful impact on a cost per unit that we take for energy from our customers." When an analyst, Kunal Vora of BNP Paribas, pointed out that power and fuel expenses had risen about 3% against tower growth of about 6%, Sah put it down to operational work.
Follow the money and the AI dividend splits. Some of it closes Indus's energy loss. Some of it, by design, flows to Airtel and Vodafone Idea as a lower energy tariff. Airtel, for its part, says it is working "in collaboration with Indus Towers" on the same problem, and its chief executive Gopal Vittal said in May that the goal is "to completely eliminate diesel from our operations." That is a customer telling its landlord, in public, what it expects the landlord to deliver.
This is why bookkeeping is the right frame. When energy is a pass-through, the thing both sides need is a shared, trusted record of what each site consumed, from which source, and why. The model that checks a fuel delivery against a sensor reading is the clause that keeps the tariff argument civil.
a grid map hiding in a tower fleet
Put the two halves of the argument together and something useful appears.
Prayas built India's best independent picture of supply quality with roughly 400 monitors. Indus says around 250,000 of its sites stream live data and some 140,000 carry smart meters. Every one of those meters, by construction, records when grid power arrived and when it stopped. The tower estate is very probably the densest consumer-end record of grid reliability in the country, spread across every state and down to villages a discom inspector rarely visits.
None of that data is public, and there are good reasons: commercial sensitivity, tenant confidentiality, and the plain fact that a tower company has no obligation to embarrass the utilities it buys from. But the energy debate this essay describes cannot be settled without it. DIPA says four to six hours of outage. The Ministry says 22.6 hours of supply. Both could be right in different districts. The tower companies hold the numbers that would show which.
I would go further and say that publishing an aggregated, district-level interruption series would serve the tower industry's own case. It would show regulators where batteries and solar make sense and where only a better feeder will do. It would give the 100,000-site solar target a map to follow. And it would turn a decade of "green site" announcements, measured in definitions that keep changing, into something an outsider can check.
The batteries are coming either way. Indus has said it will replace lead-acid with lithium-ion across all its sites over the next few years, and on 6 October it signed a memorandum with Cygni Energy to explore battery storage manufacturing of up to 1.5 GWh (Cygni's chief executive, Venkat Rajaraman, spoke of up to 2 GWh; the article does not reconcile the two). Bigger, faster-charging cells change how long a site can ride through an outage. They do not change the need to know how long the outages are.
where this leaves the ledger
Indian telecom has been trying to get diesel off its towers since at least 2011, when the industry burned about 2 billion litres a year and the regulator set hybrid-power targets for 2015 and 2020 that were missed. Indus declared half its sites diesel-free in backup terms in 2017 and set no date for the rest. In the quarter to September 2026 it reported an 11% year-on-year cut in diesel use, solar on roughly 49,000 of about 270,500 towers, and an energy business that still loses money on every unit it re-bills. The physical gains come from panels, lithium cells and, where the feeder holds, the grid. The AI that the company describes is mainly an accounting and supervision layer: meters, fuel sensors, automatic tickets and image checks on contractors' work. On the demand side, Airtel and Ericsson project radio-sleep savings of roughly 8% to 12%, figures that remain the companies' own. The government reports rural supply averaging 22.6 hours a day; the tower lobby reports outages of four to six hours in several states; the only independent minute-level data, from Prayas, is several years old. The meters that could reconcile those accounts already sit on the towers.
Tarry Singh is the founder and CEO of Real AI, an enterprise AI advisory and deployment firm working with global enterprises on production agent systems, model risk, and AI sovereignty strategy. He also leads Earthscan for Energy AI startup, and is a founding contributor to the EU-funded HCAIM and PANORAIMA programmes for responsible AI education across European universities. He writes at tarrysingh.com.