Energy and utilities firms investing aggressively in AI despite mixed returns

AI is reshaping workflows across the energy value chain, with aggressive and wide-ranging adoption. AI's permeation has not always been strategically planned or necessarily well-managed, and short-term returns have not proven consistent. Regardless, 451 Research's Voice of the Enterprise: AI & Machine Learning, Use Cases 2026, a survey of IT and line‑of‑business professionals in utilities and oil & gas organizations, shows that, despite mixed outcomes from existing initiatives, investment plans over the next year remain ambitious.

Utilities and oil & gas organizations have a strong appetite for automation. For domain-specific applications, respondents reported that AI has been applied to most use cases, with the bulk of remaining use cases expected to see AI investment over the next 12 months. A similar pattern holds for general-purpose applications; 54% of those had seen some level of AI investment, with a further 35% expected to see AI adoption within the next 12 months. Compared with other industries, energy respondents have invested more rapidly in autonomous, agentic capabilities and have placed greater emphasis on use cases that require little to no human intervention. These ambitious plans persist despite uneven near-term returns. Only 43% of AI projects initiated over the past year are forecast to deliver a return on investment within 12 months of launch. This investment trajectory has also advanced despite ongoing data privacy, quality and availability constraints. As these data challenges suggest, energy and utilities companies continue to face challenges embedding domain context into AI workflows, which is hampering the technology's value.

Energy and utilities AI adoption rapid and widespread

Across all AI classes assessed in the survey, organizations report meaningful levels of integration. The share of energy and utilities companies reporting that an AI type is "fully integrated" across the organization is 35% for generative AI, 27% for agentic AI, 24% for rules-based AI and 23% for pattern recognition, with much of the remaining respondent base indicating the technology is "adopted by specific departments or projects." These integration levels are expected to climb over the next 12 months to 49%, 38%, 35% and 41%, respectively.

This rapid advancement also extends into where the technology is set to be deployed. On average, respondents estimate that just 9% of their organization's surveyed use cases will remain untouched by AI within the next 12 months. The most popular general use cases included summarization (78% of companies), data management (66%) and data visualization (66%). In terms of industry-specific use cases, predictive maintenance (75%) and energy demand forecasting (74%) by utilities companies, and supply chain and logistics optimization by upstream (71%) and downstream (73%) companies have all seen significant attention.

This permeation is not occurring via a single sourcing route. Over the past 12 months, AI capabilities were sourced from internal development (49%), delivered as a custom solution by a third party (46%), added as an upgrade or add-on to an existing application (46%), embedded in hardware (45%), purchased as part of a new application (37%) and co-developed with a partner or consortium (37%). Budgets dedicated to AI initiatives are substantial, with 64% investing more than 10% of total IT budgets, including 7% investing over 30%.

Energy and utilities companies have a strong appetite for automation

Agentic AI uptake among energy and utilities companies appears to be outpacing the broader market. Despite the early-stage nature of the underlying technologies, 27% of respondents reported that agentic AI is already integrated across their organization, and a further 33% said it is "adopted by specific departments or projects." This compares with 18% and 27% in the overall survey population, respectively. Given this emphasis on agents, often associated with greater autonomy, it is perhaps unsurprising that energy and utilities companies also skew toward AI initiatives that require less operator intervention. About a quarter of AI projects funded over the past 12 months were scoped as fully autonomous, meaning their target state was for AI to execute actions without human intervention. A further 25% were intended to be conditionally autonomous, with AI performing actions under predefined conditions or rules.

When unintended outcomes are experienced in autonomous AI systems, most commonly the internal technical team is held responsible by energy and utilities companies, which was defined as the team that builds or deploys the system (36%). Other popular accountability frameworks primarily hold the AI vendor responsible (26%), or distribute accountability across technical and business teams (24%).

AI initiatives challenged by data and context problems

Over a third, 37%, of AI initiatives initiated in the past 12 months are perceived to have already been "successfully deployed," as in "currently operational and delivering business value." More than a third, 34%, of projects were classified as "terminated," including 9% "decommissioned post-deployment." This energy and utilities termination rate is higher than in the overall survey population, where 26% of initiatives were terminated on average.

Data challenges appear to be contributing to these outcomes. System integration, data quality and data access were the technical challenges most commonly identified by respondents. Among organizations reporting data access challenges, 54% said the issue had "materially affected timelines, scope or performance" of AI initiatives. A further 15% said it had "halted or led to failure of the initiative." Respondents in energy and utilities experience these data-related constraints more acutely than peers in other industries do.

The need for domain-specific context likely contributes to the data problems energy companies are encountering. When asked about the limitations of generative AI models and services, "situational awareness" and "customizability" were cited more frequently by energy and utilities respondents than by those in other industries. Much of the operational context required to execute tasks may not be well represented in the training corpus of general-purpose foundation models, whether grid topology or asset health signals. Complexity in storage formats and schemas can also complicate retrieval workloads, as energy companies commonly rely on a patchwork of systems.

ROI and organizational expectations

While less than half of AI initiatives over the past 12 months are projected to deliver near-term ROI, overall impact is meeting organizational expectations. On average, just 43% of AI projects were classified as on track to achieve ROI within 12 months of launch. The majority, 63%, of respondents forecast that most of their projects will not deliver positive ROI within that time frame. This does not necessarily mean that these projects are considered to have failed. Few organizations perceive AI initiatives as having actively underperformed against the objectives they are tracking. Across every objective assessed, a plurality of respondents said objectives had broadly been met, with 40% even stating that projects exceeded expectations for process efficiency and employee productivity.

Companies achieving higher ROI rates were more likely to describe a mature AI discipline, being far more likely to cite "having a clear, documented AI strategy aligned with our core business goals. We have dedicated roles and a structured career path for specialized AI professionals." They also tended to prioritize AI initiatives based on business impact, and reported fewer constraints related to cybersecurity talent shortages or integration difficulties.

This article was published by S&P Global Market Intelligence and not by S&P Global Ratings, which is a separately managed division of S&P Global.