NO BLACKOUTS. A grid drawing heavily on renewable energy can use AI to rapidly fill demand-supply gaps
| Photo Credit:
Ravi Muchhapothula
There is a photograph that energy companies love to show at conferences. A gleaming control room, banks of screens, operators watching dashboards fed by thousands of sensors across pipelines, wells and grids. Everything visible, everything connected, everything intelligent.
The KPMG Global Tech Report 2026, which surveyed 258 technology leaders across the oil and gas, mining, chemicals, power and utilities, and renewables sectors in 22 countries, tells a more complicated story. The screens are real. The dashboards exist. But the data feeding them is often incomplete, poorly governed or trapped in legacy systems.
One in five energy companies is achieving a return of more than 200 per cent on its technology investments. The majority, 57 per cent, are at break-even. The cross-sector average for technology ROI sits at 200 per cent. Energy lags, and has been lagging for a decade of heavy investment.
The pilot trap
Almost every large energy company has run AI pilots. Many have run dozens. Predictive maintenance. Production optimisation. Automated document processing. These pilots often work. They demonstrate value. And then they stop.
Twenty-nine per cent of energy companies are still in the piloting phase, running AI projects without clear returns. Executives expect that to fall to 2 per cent within a year, a claim so confident it functions as a test of whether the sector’s commitment to scale is real or rhetorical.
The obstacle is rarely AI. Around 60 per cent of energy executives say legacy systems are blocking full returns on their technology investments. An oil refinery built in the 1980s runs on control systems designed back then, robust and long-lived and wholly unequipped to share data with cloud platforms or machine learning models. Connecting them is slow, expensive and operationally risky. Upgrades are deferred, and the gap between the modern systems built on top and the ageing ones underneath — what one KPMG partner calls “digital debt” — keeps widening.
India makes this visible at scale. The National Thermal Power Corporation and the Oil and Natural Gas Corporation run sophisticated digital programmes. The State distribution companies that actually deliver electricity to homes are still fighting bad meter data, billing failures and grid losses, which in the worst-performing States exceed a quarter of the power generated. For them, the conversation about agentic AI is not yet relevant. It is about whether the underlying data is reliable enough to build on.
What the numbers hide
The financial returns that do exist come largely from the back office, finance, procurement, HR and compliance, where generative AI accelerates document-heavy work. Over 50 per cent of energy organisations report that AI contributes 31–40 per cent of their total financial benefits. That is substantial, and a long way from the story of AI changing how energy is produced and delivered, which remains, for most organisations, ahead of them.
At the operational front-end, AI’s most categorical value is in grid management. A coal or gas grid is predictable: Burn more fuel, get more power. A grid drawing heavily on solar and wind must balance variable supply against variable demand across thousands of connection points in real time, at a speed beyond human operators. India’s target of 500 GW of non-fossil fuel capacity by 2030, formalised in its Nationally Determined Contribution to the UNFCCC, will make this problem acute. The technical case for AI-driven grid management in India is as strong as elsewhere in the world. The implementation challenge, given the grid infrastructure and data quality across much of the country, is correspondingly harder.
Cutting corners
Nearly three-quarters of energy executives say that prioritising speed and cost-efficiency leads to trade-offs in security, scalability and data standardisation.
The consequences in energy are physical. A cybersecurity gap in a pipeline invites infrastructure disruption. A poorly governed AI model in grid management can contribute to blackouts. Improved cybersecurity management is the single most anticipated benefit from technology investment, ranked above revenue growth and operational efficiency.
AI sharpens the threat on both sides. It enables faster, more accurate threat detection. It also puts sophisticated attack tools within the reach of bad actors. Intrusions into Indian grid infrastructure have been documented by cybersecurity researchers, though attribution in specific cases remains contested. As India’s grid becomes more connected and AI-dependent, the attack surface expands. The gap between national cybersecurity policy and the operational reality at smaller State utilities is consequential in the country’s energy transition.
The missing skill
Ninety-six per cent of energy leaders believe that managing AI agents will be a key workforce skill within five years. The current generation of AI tools recommend; the next will act. An agentic system will not suggest adjusting a valve — it will adjust it. Managing that shift requires people who understand what these systems are doing, when to override them, and where human judgment remains indispensable.
India has engineering depth and a growing data science community. What it lacks, at scale, is the overlap: People who understand both the physics of energy systems and the architecture of AI. Shreyansh Upadhyay, KPMG India’s AI for Energy chair, puts it plainly. The biggest challenge is in building models that are context-aware and grounded in the physical behaviour of actual machinery, not just historical data patterns. That combination is hard to hire and harder to build quickly.
Published on July 13, 2026
