How artificial intelligence is reshaping the energy sector—and what it will take to make that transformation real.
When more than 11,000 energy leaders, researchers, technologists, and entrepreneurs gathered in Houston for CERAWeek 2026, one theme wove through virtually every conversation: artificial intelligence is no longer a horizon technology for the energy sector. It has arrived. The question now is whether the institutions, data systems, markets, and human organizations surrounding it are ready to meet the moment.
Dartmouth's Irving Institute for Energy and Society sent one of its largest-ever delegations to the conference, and faculty members appeared in panels spanning organizational transformation, industrial data infrastructure, grid modernization, electricity market design, and wildfire risk management. Taken together, their contributions reveal a coherent and urgent argument: the technical capabilities of AI are advancing faster than the cultural, organizational, and regulatory systems needed to deploy them responsibly and at scale. Closing that gap is a critical challenge for the energy industry and the transition to low-carbon sources we hope to accelerate.
The deepest obstacles to AI adoption in energy are not technical. They are human.
Geoff Parker, Faculty Director of Dartmouth's Irving Institute, joined executives from Microsoft, Schneider Electric, Williams Companies, and Accenture on a panel titled "Creating AI-Ready Organizations." While AI tools are advancing at breakneck speed, most energy firms are still running on siloed data and workforces trained for a pre-AI world. Fewer than 10 percent of companies today can point to a clear AI-generated return on investment—not because the technology doesn't work, but because the necessary organizational structure and operating models haven't been built.
Microsoft's Rob Schapiro described "frontier firms"—companies that have moved beyond having an AI strategy to having AI woven throughout their operations, with agentic systems that act autonomously and self-optimize. Getting there requires two kinds of leadership: transformational leadership that provides top-down vision to overhaul entrenched processes, and innovative leadership that empowers frontline teams to experiment and iterate in safe-to-fail environments.
Debbie Pickle, Chief Human Resources Officer at Williams Companies, added that senior leaders must model humility—openly adopting AI tools, acknowledging missteps, and inviting feedback—to normalize a culture in which AI is a teammate rather than a replacement. Williams Companies tracked a $7 million bottom-line improvement from a pilot team that adopted AI-augmented forecasting, demonstrating that well-tracked experiments are often the path to demonstrable business impact.
Parker contributed two frameworks that clarified the stakes. The first distinguished "above-the-line" roles—those with the agency to shape outcomes and direct AI—from "below-the-line" roles that are largely execution-focused. The organizational imperative is to help employees move above the line through continuous learning. The second addressed risk: not all energy assets carry the same profile. Lower-risk infrastructure such as solar installations presents fundamentally different considerations than higher-risk assets like production wells. Risk-specific guardrails must be built into AI strategies from the outset—a piecemeal approach can accelerate technical debt and introduce security vulnerabilities that undermine long-term value.
If organizational culture is the first barrier to AI adoption, data infrastructure is the second — and for incumbent energy firms with deeply siloed operations, it may be the harder problem to solve.
On a panel focused on data sharing and industrial AI, Parker offered a clarifying analogy: data is like gold in ore—the challenge isn't finding it, but separating and refining it into something usable. The cost of processing it has dropped dramatically. The challenge is no longer primarily technical—it is organizational, economic, and regulatory. Most industrial data sits locked inside companies, siloed across supply chains. Companies perceive more risk than benefit in sharing what they know, and without concrete evidence of value, that perception is difficult to change.
Parker, joined by Boris Otto of the Fraunhofer Institute and Noel Phillips of AVEVA, explained why this matters so much for AI: the value of data grows the more it is used and combined. A single company's operational data has value; that same data benchmarked against peers and analyzed across asset types has far greater value. Universities have a specific role to play here. Academic researchers can design value measurement methods, evaluate datasets, model sharing scenarios, and map outcomes back to inputs—giving companies an evidence-based way to quantify what they stand to gain. Putting a number on that gain, as the moderator put it, is the kind of concrete calculation that can shift corporate behavior.
Otto introduced Large Industry Models—industrial counterparts to large language models, trained on operational and engineering data rather than broad text—and argued that AI itself may help overcome standardization challenges that manual processes would take years to resolve. He also highlighted the digital twin as a foundational element of data-sharing architecture, noting that emerging regulatory frameworks in Europe are beginning to mandate structured data sharing, turning compliance into a forcing function for interoperability.
A parallel transformation is underway in the physical infrastructure the energy sector depends on. The electricity grid is being asked to do things it wasn't designed to do, and AI is both driving that demand and offering tools to manage it.
Dartmouth Associate Professor of Engineering Junbo Zhao joined a panel on smart-grid modernization with the CEO of Voltus and a senior executive from ABB. The numbers tell the story: data-center electricity consumption has jumped 30 percent over five years, electric-vehicle charging has surged 45 percent, and building electrification is accelerating. Zhao identified three forces reshaping the grid: bidirectional power flow enabled by advanced power electronics, high-speed secure communications via fiber and 5G, and accelerating electrification that simultaneously increases demand and creates new flexibility through demand response. Digital twins help operators anticipate capacity constraints and manage the grid proactively rather than reactively.
Dana Guernsey '06, TH'08, CEO and co-founder of Voltus, offered a striking observation: about 20 percent of the existing grid operates at full capacity less than one percent of the time. A smarter grid doesn't just reduce that inefficiency—it lowers energy costs, enables load growth, and accelerates lower-emissions solutions simultaneously. She predicted that within five years, the large technology companies currently straining the grid with AI data centers will become some of its most important stabilizers, participating in demand-side management and deploying storage at scale.
The barriers are familiar: funding for transmission upgrades, cybersecurity risks from new market participants, and regulatory frameworks still written for a one-way, static grid. Modernizing those rules, the panel agreed, is essential.
Assistant Professor of Engineering Cong Chen brought a different lens to the grid challenge, using AI agents—simulated humans powered by large language models—to model how real households make decisions about energy use.
Utilities struggle to predict how customers respond to dynamic electricity prices, especially during rare high-impact events like outages. Most models treat all customers as a uniform group. Chen's team assigned AI agents distinct personalities—a data-savvy graduate student, a cautious grandmother, an emotionally driven actor—and asked them to decide whether to charge or discharge a home battery in response to daily price signals. In a simulated blackout, the graduate student rushed to sell power for profit; the grandmother saved energy for emergencies. These behavioral differences have real consequences: in simple markets, customers act predictably; in volatile markets, their choices diverge, making grid planning harder.
Because AI agents can be run quickly and varied endlessly, they give utilities a low-cost way to study behavior during rare events without waiting for those events to occur. Chen also proposed new pricing formulas to eliminate the volatile price swings that currently force generators to receive out-of-market subsidy payments—replacing them with mechanisms that keep all payments inside the market and encourage honest cost reporting. Behavioral simulation paired with fairer price signals points toward a grid more responsive to real human preferences and more stable under stress.
One Dartmouth presentation addressed a dimension of the grid that is easy to overlook when the focus is on expansion and optimization: the growing risk that electrical infrastructure can trigger catastrophic wildfires.
Associate Professor of Engineering Vikrant Vaze presented research conducted with his PhD student, Spencer Bertsch, on optimizing Public Safety Power Shutoffs (PSPS)—the practice of proactively de-energizing sections of the grid when environmental conditions create elevated fire risk. PSPS events reduce fire risk but carry real social costs: disrupted medical devices, spoiled food, and hardship for the communities utilities are trying to protect. The question Vaze and Bertsch are investigating is how to make shutoff decisions more precise.
Their research focuses on three environmental factors that interact dangerously with electrical infrastructure: low humidity, dry vegetation, and strong winds. Using calibrated simulations, the team analyzes how these factors combine to elevate risk across a grid. Two California case studies illustrated the value of calibration. In the 2020 Blue Ridge Fire, a calibrated model produced results only modestly different from standard approaches. But in the more severe Mineral Fire that same year, the calibrated simulator revealed substantially different risk assessments—with important implications for which grid sections should have been shut down and when. The more severe the event, the more consequential the gap between the calibrated model and the normal model.
The work connects directly to themes running through the Irving Institute's broader CERAWeek presence: the value of better models depends on the quality of the inputs feeding them, and operators need more precise, dynamic tools for managing complex systems under uncertainty.
Across these five sessions, Dartmouth's presence at CERAWeek 2026 reflected a coherent view of what the energy transition requires. It requires organizations willing to redesign their cultures and workforce practices around AI as an embedded capability. It requires data infrastructure that moves industrial information out of silos and into shared ecosystems—with universities playing a specific role in building the evidence base that makes sharing rational. It requires a modernized grid with updated rules to match the bidirectional, distributed, data-intensive system now emerging. It requires modeling tools sophisticated enough to capture the behavioral richness of real humans. And it requires the clear-eyed recognition that the grid remains a physical system operating in a physical world—one where dry winds and aging infrastructure can combine to produce disasters that demand precise, risk-aware modeling to anticipate and mitigate, because better outcomes depend on better inputs, not just faster computation.
The panelists returned repeatedly to one conclusion: these challenges are too complex and interconnected for any single company or sector to solve alone. Collaborative ecosystems—where firms contribute genuine strengths and rely on networks for the rest, where industry and academia build evidence together, and where regulators and operators align around shared standards—are not idealistic. They are necessary.
For Dartmouth's Irving Institute, CERAWeek 2026 was both a showcase and a working session: a chance to bring research into contact with the practitioners who need it, and to bring the complexity of practice back to the researchers and students who will shape what comes next.
CERAWeek is regarded as one of the most influential annual conferences in the energy sector, drawing more than 11,000 energy leaders, researchers, technologists, and entrepreneurs from around the globe.
At the 2026 conference, Dartmouth faculty members appeared in panels spanning organizational transformation, industrial data infrastructure, grid modernization, electricity market design, and wildfire risk management.