The energy sector has no shortage of AI ambition, but—according to a CERAWeek 2026 panel of researchers and industry leaders—it lacks the trusted data infrastructure needed to make that ambition real.
S&P Global's Levi Tillemann moderated the session "Data Sharing and Industrial AI: Reuse at Scale," with speakers:
The conversation drew a through‑line between academia and industry, with university researchers playing a central role in diagnosing the problem and pointing toward solutions. The speakers examined why so many AI initiatives stall at the pilot stage—and what it will take to move from promising experiments to durable, scalable value.
Tillemann opened with a pointed analogy: "Oil creates value only after it has been extracted, refined, and delivered." Industrial data works much the same way—except that, in most organizations, it never makes it out of the ground. It sits locked inside companies, siloed, and fragmented across supply chains.
Parker linked that observation to his research on platform economics and network effects. "Unlocking data creates opportunities for value exchange," he said. But companies, he added, currently perceive more risk than benefit in sharing what they know—which is precisely where university‑industry collaboration enters the picture. Parker and his colleagues have spent years working with companies to build concrete use cases that demonstrate the value of sharing data.
Parker offered a useful framework for thinking about what is actually at stake: data is a non‑rival resource. Unlike water, which is consumed when used, information can be shared at essentially zero marginal cost. "Data is like gold dust in dirt," he said. "Processes have gotten cheaper; now we can process, use, and share it." The challenge is no longer primarily technical—it is organizational, economic, and regulatory.
Otto offered a European perspective on the standardization challenge. Getting organizations to agree on common definitions—what a data point means, how it was generated, and what it represents—is laborious work. He used the electric‑vehicle industry as an example: charging data, driving patterns, battery status, and recycling eligibility all need to align across manufacturers, regulators, and recyclers before they become useful.
Otto introduced Large Industry Models (LIMs)—industrial counterparts to the large language models (LLMs). Where LLMs are trained on broad text, LIMs are trained on operational and engineering data specific to industrial domains. He noted that patterns can emerge from the data, suggesting that AI itself may help overcome the cumbersome manual processes that have historically blocked standardization.
He also highlighted the digital twin, which is a continuously updated digital representation of a physical asset, as a foundational element of data‑sharing architecture. In the EV context, a digital twin of a battery can track how a vehicle has been charged and driven, directly informing its recycling potential. He noted that regulatory frameworks, including emerging product‑passport requirements in Europe, are beginning to mandate this kind of structured data sharing, turning compliance into a forcing function for interoperability.
Phillips, who works with AVEVA—one of the world's largest industrial‑data companies—reinforced a point that resonated throughout the session: the value of data grows the more it is used and combined.
Extracting that value, however, is not straightforward. Getting real‑time operational data out of legacy systems, placing it in the right context, and making it shareable across an ecosystem of developers and service providers remains a significant challenge. Phillips described how AVEVA is applying AI to tasks like scanning physical assets to identify equipment—an example of AI helping to build a data layer that more sophisticated AI can later depend on.
He also highlighted an emerging frontier: combining industrial operational data with other streams—such as HR or workforce data—to answer questions like why one set of operators consistently outperforms another. He noted that more insight looking at large data sets together. These cross‑domain analyses are difficult to pursue within any single organization, let alone across company boundaries, which is part of why the data‑sharing question matters so much.
Tillemann posed the question that every data‑sharing conversation eventually reaches: What's in it for me? If a company shares data with a public entity or a partner, could a competitor benefit? What, precisely, is being given up?
Parker offered a practical distinction: directory data, which is stable, descriptive information, carries a different risk than flow data, which is constantly changing and therefore less risky to share. He went further, arguing that academia has a specific and underutilized role to play in resolving the valuation problem. Universities can evaluate datasets, model various sharing combinations, and map outcomes back to inputs—giving companies an evidence‑based way to quantify what they stand to gain or lose.
Tillemann put it plainly: "If we can say, 'share this data with a vendor, get a 3 % efficiency gain,' then we can speak to businesses."
An audience member from the oil‑and‑gas sector confirmed how real this issue is. Even within a company's own client base, every piece of data tends to be treated as proprietary. The industry, they noted, "Isn't good at sharing." Tillemann acknowledged that while some owner‑operator relationships do produce mutually beneficial data‑sharing arrangements, they are far from universal.
A recurring theme was the wide variation in data readiness across firms. Consumer‑technology companies have spent years thinking carefully about the trade‑offs between user data sharing and the value users receive in return. In the industrial space, that level of sophistication is largely absent.
Otto agreed, "There is much reluctance and low data readiness." Organizations need both a robust data layer and resource‑management capabilities before they can confidently decide what to share and with whom.
Parker explained what this means for less‑prepared organizations: "Low‑data‑readiness firms are in a tough spot." But he also pointed to the opportunity. Helping companies build that readiness—through research, frameworks, and patient, use‑case‑driven engagement—is exactly the work that university‑industry collaboration is built for.
A final comment from the audience illustrated the technical dimension of the problem: time‑series data, the backbone of industrial operations, presents its own cleansing and contextualization challenges. Raw data can be valuable, but it also contains noise that must be interpreted carefully. The consensus: data needs to be cleaned, placed within an asset framework, and then examined across metadata—with security and shareability built in from the start.
The session closed with a shared sense of what success could look like. "Data sharing," Otto said, "needs to move beyond hype and deliver measurable value." That will require ecosystem stakeholders—companies, regulators, technology providers, and universities—to work in genuine alignment.
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.