Capacity gets you in. Intelligence keeps you ahead: Inside Hindustan Power’s AI rebuild

On 21 May, India’s power system did something it had never done before. As heatwaves gripped the country, peak demand touched 270.82 GW, a fourth consecutive daily record. Thermal plants supplied a little under two thirds of it and solar around a fifth, according to the Ministry of Power. The grid held.

But the headline number hides the harder story, which is what happens on either side of the peak.
Analysis by the energy think tank Ember found that on a single March day, solar and wind supplied 41 per cent of India’s midday generation. By then coal output had been pushed down by roughly 49 GW in six hours, only to be called back up just as quickly once the sun set and evening demand arrived.

Arman Puri, Director at Hindustan Power, quotes the same pattern, with coal rising by 51 GW in just three hours of the evening ramp, and draws a blunt conclusion from it. “Balancing movements of that size across millions of assets cannot be done by hand,” he says. “At that scale, coordination itself becomes the product, built on forecasting, automated dispatch and prices that change through the day.”

That sentence is a fair summary of how one of India’s older private power developers now thinks about technology. Hindustan Power, led by Chairman Ratul Puri, built its reputation on being early: it commissioned what it describes as India’s first 5 MW solar plant and Asia’s first 30 MW solar project, and it later added thermal capacity and a growing storage pipeline. Its Annupur plant in Madhya Pradesh began supplying 130 MW to Assam in October 2025, and in March 2025 the company signed a Rs 620 crore agreement with the Assam government for a 100 MW solar plant and a 100 MW battery system. Today the group has a portfolio of nearly 6 GW, operational and under development, spread across thermal, solar, storage and mining. Puri’s argument is that the next phase of competition will be decided less by who builds the most megawatts and more by who runs them best.

Technology stops being a separate plan
For a company with that much steel in the ground, the instinct might be to treat AI as a layer of analytics added after the fact. Puri says Hindustan Power has deliberately moved in the other direction.

“Technology is no longer a separate plan at Hindustan Power,” he says. “Our portfolio of nearly 6 GW, operational and under development, spans thermal, solar, battery storage and mining, and almost every significant decision across it now carries a technology question.”

The practical consequence shows up in project design. “When we design a solar or storage project, we define the data it will generate, and the teams that will act on that data, alongside the electrical and civil design rather than after commissioning,” he says. In other words, the data model is specified in the same meeting as the transformer.

Puri is also unusually direct about where he thinks the technology is heading. The largest AI labs now talk openly about superintelligence, and Meta and Microsoft have each set up dedicated teams to pursue it. Whatever the label, he says, “we plan on the assumption that these systems will sit inside our workflows, not beside them.” He sketches what that looks like on an ordinary day: invoices matched and checked before a person sees them, shift logs and sensor readings turned into maintenance work orders without being retyped, and mine plans and logistics schedules drafted overnight and approved in the morning by the teams that own them. “People will set objectives, resolve exceptions and remain accountable for outcomes,” he says, “while the systems do the first pass on nearly everything else.”

For CIOs, the more provocative claim follows. “That moves the strategic constraint from procurement to redesign,” Puri says. “For each function, the question is how quickly its processes can be rebuilt around these systems, and how directly the result shows up in cost per tonne, plant availability and turnaround times.” It is a pointed reframing. The scarce resource is no longer access to AI, which any competitor can license, but the organisational nerve to rebuild finance, procurement, operations and mining around it.

A grid that no longer runs one way
The urgency comes from the shape of the system itself. Puri describes the grid as a machine whose basic design assumption is quietly expiring. “For more than a century it was built to move power in one direction, from a few hundred large plants to millions of passive consumers,” he says. “That model is giving way to a distributed, connected network.”

The numbers he cites explain why. More than 50 lakh homes now have rooftop solar under the PM Surya Ghar scheme, and over 7 crore smart meters have been installed. “Rooftops, batteries and EVs can draw power at one hour and supply it the next,” he says, “and the grid has to see, price and balance all of them.” The International Energy Agency estimates that AI-based grid management could free up to 175 GW of transmission capacity on existing lines worldwide and cut outage durations by 30 to 50 per cent, which makes the case for software as an alternative to building new wires.

Where AI is already paying its way
The clearest returns at Hindustan Power so far are not in glamorous places. They are in coal logistics, a part of the business where paperwork and physical checks pile up. Across those operations, Puri says, AI now handles seal verification, document processing, movement monitoring and testing. Every month this saves more than 1,000 hours of manual work and close to 50,000 pages of paperwork.

The mines are next. The company is rolling out AI-based fleet management that matches excavators and dumpers in real time, along with predictive maintenance on heavy earth-moving equipment. The economics are straightforward. “Haulage accounts for roughly half the operating cost of an open-cast mine,” Puri says, “so even modest gains there are material.”

When a forecast becomes revenue
In renewables, the same logic is sharpened by regulation. “In renewables, forecast accuracy now shows up directly in revenue,” Puri says. From April this year, the Central Electricity Regulatory Commission halved the deviation tolerance band for solar projects to plus or minus 5 per cent, with wind moving to plus or minus 10 per cent. Grid-India’s simulations on operating plants suggested that solar projects would stay within the tighter band only about half the time.

“Every improvement in forecasting therefore reduces deviation charges,” he says. The regulator has also set out a phased path, ending in 2031, that will bring wind and solar sellers progressively closer to the rules that govern conventional generators, so the pressure to forecast well will only grow.

Storage plays by different rules, and it is scaling at startling speed. Mercom India Research counts 8.2 GWh of battery capacity added in the first half of 2026, against 98.4 MWh in the same period a year earlier, an increase of roughly 84 times. Ember estimates that around 10 GWh of midday charging would be enough to absorb surplus renewable output and keep coal plants above their technical minimum.

Under the standalone storage contracts now being awarded, Puri notes, the state load despatch centre decides when a battery charges and discharges, while the developer must guarantee at least 95 per cent availability and 85 per cent round-trip efficiency. That changes where the analytics effort belongs. “The analytics that matter are about battery health: temperature, cell balance and degradation, with faults identified before they affect availability,” he says.

In mining, the priority is trust in the numbers. Hindustan Power is rolling out drone-based LiDAR, with AI calculating volumes from each scan. Those volumes are reconciled against weighbridge data and fleet telematics across the pit, stockyard, rail siding and delivery point, while automated sampling and testing create a digital quality record for coal consignments. The accuracy matters, but Puri says the cadence matters more. “The larger gain from drones is frequency,” he says. A conventional survey can occupy a surveyor for a day or two, which is why most mines survey monthly at best. A drone flight takes under an hour, with results within a day, so measurement can move to a weekly cycle.

Puri has little patience for analytics that stop at the screen. “Many operations run expensive systems that produce detailed dashboards and change little on the ground,” he says. “The value lies in systems that recommend the next action and, over time, take it.”

The three capabilities that will separate winners
Looking ahead, Puri is specific about what technology leaders need to build. “Three capabilities will separate energy companies over the next five years,” he says.

The first is real-time operational data. “Battery operations, fleet control and grid services depend on minute-level data from the field,” he says, “while much of the industry still manages through daily and monthly reports.” A company that learns what happened yesterday cannot compete with one that knows what is happening now.

The second is integration. Puri expects core platforms such as SAP to remain the systems of record, with an AI layer above them that draws on several systems at once and coordinates work across them. That only works if equipment and software can exchange data freely, which is why he is watching the Ministry of Power’s move to mandate full interoperability across smart meters to prevent vendor lock-in. His advice is to copy the regulator’s homework. “Companies should apply the same standard to their own procurement,” he says, “with open interfaces and full ownership of raw data as baseline requirements.”

The third is talent, and here he points to a hiring signal. Job postings for forward-deployed engineers this year have run at more than ten times last year’s level, he says, because “enterprises have found that a model creates value only once it is built into the way work is done.” Hindustan Power’s answer is to embed people rather than centralise them. “We are placing forward-deployed AI engineers in every function, from finance and procurement to plant operations and the mines,” he says, “working alongside the teams that run them.”

His counsel to peers is deliberately unglamorous. “Technology leaders should begin with an operating problem and a measurable target,” he says. “They should build where the problem is specific to the business and buy where the function is standard. Every pilot should have a fixed window to prove its value before it is scaled or closed.” Pilots that never end, in his telling, are simply a slower way of failing.

The energy company of 2030
None of this, Puri insists, means software replaces steel. “Physical infrastructure will remain the price of entry,” he says. “Plants, mines, transmission lines and storage require capital, land and years of approvals, and software shortens none of that.” India will need a great deal more of all of it. Government projections put peak demand at 459 GW by 2035-36, and the Central Electricity Authority’s generation adequacy plan envisages 174 GW of storage, or 888 GWh, by then. Puri cites the CEA’s estimate of more than 230 GWh of battery storage needed by 2031-32.

He sees system reliability by 2030 resting on three things working in concert. The first is firm power. In non-solar peak hours this year, coal supplied close to three quarters of the electricity generated, and Puri expects thermal capacity to remain the flexible foundation of the system for many years. The second is storage. The third is flexible demand, which he considers the cheapest capacity of all. The cheapest new capacity will often be demand that can be moved, he argues: air-conditioning that pre-cools ahead of the evening peak, EVs that charge at midday, and industrial loads that shift by a few hours.

Coordinating a system with millions of participants requires shared digital rails, and this is where Puri’s attention turns to policy. The India Energy Stack, a Ministry of Power initiative whose task force is headed by Infosys co-founder Nandan Nilekani, is meant to do for electricity what UPI did for payments, with unique IDs for consumers and assets, consent-based data sharing and open APIs. “The physical build will deliver the capacity; the digital build will determine how much of that capacity can actually be used,” he says.

Then comes the prediction that should give every CIO pause. “Access to AI will be universal,” Puri says. “If the labs deliver even part of what they promise, raw intelligence stops being scarce. The advantage then moves to clean operational data, systems wired into physical assets, and organisations prepared to change how they work.” Models can be rented by anyone. Years of tidy sensor history, open interfaces and teams that have already rebuilt their processes cannot.

That logic extends to national security. “AI’s most immediate contribution to energy security is getting more out of what India has already built,” Puri says. Sharper forecasts reduce the reserves held against uncertainty, and predictive maintenance keeps capacity available through the peak. Over the longer term, he argues, security depends on domestic depth across the value chain, down to the software layer, and recent storage tenders under the viability gap funding scheme already require energy management systems developed in India. Hindustan Power, for its part, is building into biofuels and critical minerals on top of its existing mix.

The race after the build
India will keep pouring concrete, stringing lines and commissioning batteries, and the companies that do it fastest will earn their place on the map. But the evidence from the grid, the pit and the dispatch room points the same way. Capacity is becoming a commodity that anyone with capital and patience can assemble. Operating it, minute by minute, with fewer people doing first-pass work and more doing the thinking, is not.

Puri’s closing line is as much a challenge to CIOs as a forecast. “Capacity will still decide who is in the business in 2030,” he says. “How well that capacity is run will decide who leads it.”