Five energy companies. One pattern. The decision always came first.
In 2015, Francesco Starace sent every employee at Enel a survey and discovered that his most senior leaders were the least digitally capable people in the company.
The further up the hierarchy you went, the further behind you were.
A different CEO might have buried that finding. Starace — who had taken over Europe’s largest utility a year earlier — reorganized the company around it.
He launched a strategy called Open Power that broke the vertical silos between departments and oriented Enel around digital capability rather than existing hierarchy. The logic was stark: if the people with the most authority to approve change are the least equipped to understand what needs changing, the company is built wrong.
By 2019, more than 50% of Enel’s energy came from zero-emission sources. Enel X, a new business unit built around digital services, was steadily overtaking the legacy retail business. By 2020, Enel was the most valuable utility in Europe.
The survey wasn’t a diagnostic. It was the provocation that produced a different company.
One sector. Three bets. One pattern.
While the tech press wrote about Silicon Valley, a handful of companies in one of the world’s oldest, most capital-intensive, most heavily regulated industries were doing something more interesting than adopting AI.
They were redesigning how they operated around what AI made possible.
Offshore wind farms. Upstream oil platforms. Continental electricity grids. Not obvious candidates for transformation. And nobody wrote the profile.
Three companies in particular — Ørsted, Shell, and BP — faced a version of the same problem in roughly the same era. Expensive, remote assets. Enormous quantities of sensor data nobody was fully using. Maintenance regimes built around calendar schedules and human inspection, not prediction.
They each answered it differently.
Ørsted built it.
In 2017, Ørsted divested its upstream oil and gas business and became a pure-play renewable energy company.
That is not an IT decision. That is an identity decision. And it changed everything about what came next.
With 1,300 offshore wind turbines to operate, Ørsted needed to run a fundamentally different kind of company than anyone had run before. Each turbine carries thousands of sensors. Each minute produces vast quantities of data. The operational question — not the technology question — was what to do with all of it.
Working with Microsoft, Ørsted built predictive maintenance AI across its offshore fleet. Engineering compute time for new wind farm foundation design dropped from weeks to four to eight hours. The company launched an internal program called “Democratise AI” to spread capability through the organization rather than concentrate it in a specialist team. By 2023, it had extended AI to 5.5 gigawatts of US land-based wind, solar, and storage assets.
CIO Michael Biermann put it plainly: “Each turbine is equipped with thousands of sensors and each minute, each hour, they produce vast quantities of data that we can analyse and optimise.”
That is not a technology implementation talking. That is an operating model talking.
Shell bought it.
By 2019, Shell was running pilots with C3.ai on critical valves in a refinery in the Netherlands. By March 2022, that had turned into monitoring more than 10,000 pieces of equipment across upstream, manufacturing, and integrated gas assets globally.
The underlying infrastructure ingested 20 billion rows of data weekly from more than 3 million sensors. It ran nearly 11,000 machine learning models in production and made over 15 million predictions every day.
The reported outcomes: a 20% reduction in unplanned downtime. A 15% reduction in maintenance costs. £1M+ in annual savings per site.
Those are real numbers. But the more interesting fact is the scale ambition from the outset. Shell wasn’t running a proof of concept. Shell was industrializing a platform it had bought.
Dan Jeavons, Shell’s VP of Computational Science and Digital Innovation, framed the 10,000-asset milestone as an internal target the company had set for itself in 2021 and hit on schedule. The deployment schedule was the point.
BP funded it.
BP made neither an internal-build bet nor a platform-purchase bet. BP made venture bets.
In 2017, BP Ventures put $5 million into the Series A of Belmont Technology to deploy an AI geoscience platform — nicknamed Sandy — for subsurface analysis. Sandy was designed to cut the time BP engineers spent on subsurface data collection and interpretation by 90%.
BP also invested $20 million in Beyond Limits, applying AI technology originally developed for NASA deep space missions to offshore exploration and production.
David Eyton, BP’s Group Head of Technology, described AI as central to BP’s digital strategy for upstream operations. By the end of 2018, BP’s digital models had reduced offshore visits and generated roughly $45 million in operational savings in Trinidad alone.
Three companies. One problem. Three radically different sourcing models — build, buy, fund. All three produced real outcomes.
The lesson is not which model was right. The lesson is what made all three of them work.
In each case, the company had decided AI wasn’t a tool to evaluate. It was a capability to build into the operation. The sourcing debate was downstream of that decision.
Xcel commits first, builds after.
The American case makes the pattern explicit.
Xcel Energy, under CEO Ben Fowke, became the first major US utility to commit to 100% clean electricity. That commitment — announced in December 2018 — created a very specific operational problem. Renewable energy, particularly wind, is variable in ways that coal and gas are not. If you can’t predict when the wind blows, you have to keep backup generation on standby. That standby generation is expensive, carbon-intensive, and quietly undermines the commitment you just made.
Xcel attacked the forecasting problem directly. The company partnered with the National Center for Atmospheric Research, feeding NCAR sensor data from hundreds of wind turbines to train machine learning models for high-resolution wind forecasts.
The result: a 40% reduction in wind forecast margin of error. $60 million in annual customer savings. 250,000 tons of carbon emissions avoided per year through higher renewable energy utilization.
Xcel didn’t deploy AI because AI was interesting. Xcel deployed AI because it had made a public commitment that required better forecasting to honor. That is the same move Ørsted made when it divested oil and gas: decide what kind of company you’re going to be, and let AI become the delivery mechanism for the decision.
The commitment came first. The technology was in service of the commitment.
The pattern, plainly.
Five companies. One sector. Wildly different sourcing models, different geographies, different asset types.
One consistent pattern underneath all of it.
The companies that got durable transformation had made a prior decision. About what kind of company they were going to be (Ørsted). About the scale at which they would operate (Shell). About which bets they’d back rather than build (BP). About what public commitment they would honor (Xcel). About which leaders had the authority to shape the next decade (Enel).
AI was how they delivered on that decision.
The companies that got expensive pilots — and every sector has them, this one included — made the opposite move. They evaluated AI as an object of interest. They ran controlled experiments. They measured the experiment against the existing operation rather than asking whether the existing operation was the right baseline.
Adding AI to an existing operation gives you a smarter version of what you already had. Redesigning an operation around what AI makes possible gives you something different.
The gap between those two things is the gap between the case study and the footnote.
What this means in 2026.
Your organization is not more regulated than Enel. Not more capital-intensive than Ørsted. Not more operationally complex than Shell. Not more publicly committed to a hard goal than Xcel.
None of these companies were born digital. They carried legacy infrastructure, union workforces, regulatory obligations, and capital structures that made experimentation genuinely costly. They were not operating in a forgiving environment.
What they shared wasn’t a technology advantage. What they shared was a prior decision about what the organization was for — and a willingness to let AI reshape how they got there.
The transformation nobody noticed wasn’t about AI. It was about a leadership team that had decided, clearly and irrevocably, what kind of company they were going to be. The technology just caught up.