Data Unlocks AI for India’s Power Sector: Scaling It Takes Trust and Institutions Too
A Climate Collective Foundation's perspective from the AI for Climate Tech Dialogue, Delhi, October 2025
Context
Climate Collective Foundation convened senior leaders from utilities, government, policy institutions, think tanks, technology firms, and international organisations in Delhi, splitting the room into two parallel conversations at separate tables to map the AI-for-power ecosystem from two angles: one table on development (how AI solutions for the power sector actually get built), and the other on deployment (the pathways needed to take proven solutions to scale). The split was designed to help more AI solutions get incorporated into the power sector in a holistic and sustainable way, since development and deployment raise different questions, and we wanted each conversation address the questions that applied to it.
The Big Idea
India already has what it needs to lead globally on AI for Power: rapid RE growth, a national smart-meter rollout, and a public compute mission arriving at the same time. What stands between that potential and results at scale are three solvable gaps: data that utilities can’t yet share safely, trust that AI systems haven’t yet earned, and institutions that aren’t yet built to carry a pilot to procurement. The development table and the deployment table, working independently, converged on data and trust as shared diagnoses, a stronger signal than an opinion the room happened to share, and closer to a confirmed finding. Close these three gaps and the payoff compounds across the entire pipeline.
Three Key Insights
The single biggest constraint to scaling AI in India’s power sector is data availability and interoperability. Both tables converged on this independently: forecasting, DER orchestration, and asset-health tools are largely ready to deploy, but SCADA, AMI, GIS, and weather data remain siloed and non-standardised across utilities. The fix, already on the table – a sandbox or federated-access model where utilities keep control while enabling anonymised, rule-based validation – is ready to implement now. A related, easily missed finding: data centres are an underused flexibility asset that can also relieve grid load. In the US, shifting just 1% of data centre load timing could offset all projected new demand through 2030, a lever India could plan for.
Utility adoption of AI depends on explainability, reliability, a measurable payback, and someone willing to fund the gap before that payback shows up. Discoms weighed Return on Investment (ROI), reliability, and resilience gains; load dispatch centres weighed grid balancing and stability. Across both, explainability came up as a precondition for trust: regulators and utility staff need to see why a model made a call before they’ll act on it. Demonstrating that ROI case is itself expensive, since utility funding cycles are built around capital expenditure rather than experimentation, which is why participants called for patient, risk-tolerant capital (from philanthropy, DFIs, or corporate CSR) to fund pilots through to the point where the ROI case can actually be made.
The institutional gate, more than the technical one, is where we miss out on scaling. Even where the data and the trust exist, pilots routinely stall before they reach procurement. The pattern we've witnessed include: no pre-agreed cost-recovery mechanism, no regulatory filing pathway, and utility teams without the ownership or budget continuity to carry a pilot to scale. The fixes named were specific and unglamorous: a Regulatory Filing Kit to translate AI benefits into tariff-petition language and innovation cells embedded inside utilities to anchor ownership. Building that plumbing is squarely an institutional task.
What This Means for the Ecosystem
For startups, think tanks, and industry alike (well beyond data-infrastructure builders), there’s an entry point in each of these findings. The most foundational is still a data governance layer: standards, sandboxes, and federated access models that multiply the value of every AI pilot built on top of them. Two other leverage points matter just as much. Startups that can pair their models with credible explainability and evaluation tools have a real opening, since utilities and regulators are gating adoption on trust as much as performance. And the institutional gap, covering cost-recovery frameworks, regulatory filing support, and patient capital structuring, is as much a place for think tanks and intermediaries to add value as it is for utilities to fix internally. Programmes like ElectronVibe, which already broker startup-utility relationships at the operational level, are well placed to work across all three. For India specifically, and the Global South by extension, this is a first-mover opportunity: whoever builds working, replicable models for data trust, AI evaluation, and institutional pathways to scale sets the template every other market with legacy discoms and fragmented systems will want to adopt.
Open Questions
- Data-sharing and data access remain the foundational blockers across the board. What would broader, standardised data access for AI in the power sector actually need to look like?
- Patient capital was called for repeatedly, but most funding cycles in this space are built around capital expenditure rather than experimentation. What would it take for patient capital to become a sustainable feature of how pilots get funded?
- Evaluating AI claims credibly requires frameworks and capacity-building that don’t widely exist yet. What frameworks and capacity-building efforts are needed, and who is best placed to build them?
Where We Go From Here
Climate Collective Foundation will carry these insights into a forthcoming whitepaper with recommendations for ministries, utilities, and funders: the AI for Power Playbook, to be launched at Bangalore Climate Data Week, running 12 to 16 October. Reach out if you want to help design the data-sharing sandbox or bring a utility relationship to the table.