The Arthur L. Irving Institute for Energy & Society

America's electric grid was mostly built between the 1930s through 1960s, and many of its components are now operating well past their intended design life. Today, that grid must power a 21st-century economy, facing surging electricity demand from data centers, electrification, and reshored manufacturing.

This creates a host of challenges: the grid needs to be upgraded, expanded, and made smarter. Here and across Europe, it was engineered nearly a century ago for one-way, predictable power flow, where electricity moved from large power plants to homes and businesses. That model no longer captures the efficiencies now possible with distributed energy resources, like rooftop solar panels and batteries, which can generate power locally and feed extra energy back into the grid.

What's needed is a smarter grid that supports two-way flow of electricity while remaining stable, secure, and fast enough to keep pace with the demands of a 21st-century economy. That tension anchored a panel convened September 23 at Climate Week NYC 2026 by SAP and Dartmouth's Arthur L. Irving Institute for Energy and Society. The discussion was held under the Chatham House Rule, so comments below are not attributed to individual speakers.

A Panel of Experts

The discussion was moderated by Torsten Welte, SVP and General Manager of Energy & Natural Resources Industries at SAP. He was joined by Geoff Parker, faculty director of the Irving Institute and the Charles E. Hutchinson '68A Professor of Engineering Innovation at Dartmouth; Bill Hendricks, President of Cognite Americas, an industrial data software firm; Philippe Piron, President and CEO of the Electrification Segment at GE Vernova; and Vish Ganti, President and COO of DERSec, a firm focused on securing distributed energy resources.

A Grid Built for One Direction, Now Running in Two

The panel opened with a structural reality: the grid was designed for centralized, predictable generation, in which large power plants pushed electricity outward to consumers. That model is now outdated. Distributed energy resources, or DERs, such as rooftop solar panels, home batteries, and electric vehicles, have turned millions of customers into participants who can both consume and supply power.

This bidirectional flow, panelists noted, introduces a volatility that the 20th-century grid wasn't designed to handle. Because the transmission system is so tightly synchronized, a shift of just one or two percent in load or generation can trigger instability.

This is a growing concern given how dramatically renewable and distributed capacity has expanded, both in the U.S. and abroad. Global clean energy investment approached $2 trillion in 2024, roughly double fossil fuel investment, according to the International Energy Agency. Panelists cautioned that physical infrastructure and market rules haven't kept pace with the variability this growth introduces.

The Data Silo Problem

A recurring theme was that utilities, generators, and industrial players sit on enormous quantities of data that could enable more efficient grid operations, but they lack incentives to share it. As one panelist put it, the operative question for any company asked to share proprietary data is simple: "What's in it for me?"

Utilities can unify data internally, but today's landscape is far more fragmented. It involves not just utilities and the organizations that manage regional electricity grids, but also distributed energy resource aggregators, also known as virtual power plants or VPPs, as well as individual asset owners. What's needed, panelists argued, is a shared incentive structure that helps the full energy ecosystem, meaning producers, consumers, and "prosumers" who do both, build a more cost-effective and resilient grid.

Creating Value from Data

Even when data is shared, panelists noted, it isn't automatically useful. For example, raw industrial data, they agreed, is essentially worthless on its own. Value only emerges once it's contextualized and turned into something actionable.

For grid operators, the challenge is not only to gain visibility behind the meter, but also to better utilize unstructured data: sensor feeds, asset logs, and telemetry that mean little without being mapped onto physical assets. Operators need to know what equipment is connected to what, and what a fault in one place implies for the rest of the system. The panel's consensus was that grid intelligence should start with organizing data around physical assets, rather than chasing individual AI use cases, an approach that too often produces fragmented, hard-to-scale pilots.

Human-in-the-Loop

Another point of consensus was that we are far from ready for an AI-run grid. AI is already reshaping grid operations, from predictive maintenance to demand forecasting to real-time price-spike detection. Still, no one on the panel felt we're ready for full AI autonomy. European operators described an environment markedly more cautious than their U.S. or Asian counterparts, treating AI strictly as a decision-support assistant given the catastrophic downside risk of a cascading blackout.

The refrain "human-in-the-loop" surfaced repeatedly, driven as much by liability concerns as by technical caution. Panelists agreed that someone must remain accountable for consequential grid decisions. One speaker noted the irony of operating assets designed to last 40 to 60 years or more, such as transformers, transmission lines, and substations, using AI models that can be retrained or replaced within months. That speaker called this mismatch living in "asymmetric time."

A Case for Open Standards

Cybersecurity emerged as a central anxiety, particularly regarding the fragmented software stack connecting utilities to customers, aggregators, and third-party device makers. Panelists described a landscape in which a single request, such as adjusting a smart thermostat, might pass through four or more intermediary systems, each representing a potential point of failure. One panelist announced a new initiative aimed at using public-key encryption—a security method that uses paired digital keys to verify identity and protect data—to secure data flowing between utilities and connected devices.

The broader argument was that open-source standards and interoperable protocols are necessary to prevent what panelists called concentration risk: a scenario in which a handful of vendors or AI models end up controlling critical infrastructure decisions without adequate auditing or regulatory oversight.

Capital, Talent, and the Limits of Speed

The panel closed on a note of financial and human constraint. Global grid and clean energy infrastructure investment has grown sharply. According to the International Energy Agency, over $300 billion was invested in grid infrastructure alone in 2023, with some estimates calling for that figure to roughly double by the early 2030s to keep pace with electrification. Panelists questioned how long that pace of capital deployment can be sustained, and framed it as a question for financial institutions as much as for engineers, since both need a realistic sense of what return on investment looks like in this new environment.

Compounding the funding challenge is a talent gap. Building and securing next-generation grid infrastructure requires specialized engineering and software skills that current workforce programs are not producing fast enough.

Physical constraints add another layer of difficulty, especially in the U.S., where new transmission projects can take seven to nine years to permit and build. Panelists were candid that there is no single fix. In the meantime, some operators are deploying mobile gas turbines, sometimes called "turbines on wheels," and flexible data center load agreements as stopgaps while permitting and transmission catch up.

Growing Trust

Asked how utilities and regulators might test AI-driven grid systems without risking reliability, panelists drew an analogy to autonomous vehicles: trust is built incrementally, through transparency and small-scale customer trials. Digital twins, or high-fidelity simulations of physical grid assets, were also cited as a helpful bridge, allowing operators to stress-test AI recommendations against realistic scenarios before deploying them in the real world. As one panelist observed, no matter how sophisticated the algorithm, the grid will ultimately obey the laws of physics.

The panel converged on a shared vision: building an intelligent, secure, and resilient grid is as much a challenge of economics, trust, and governance as it is of engineering.