The problem
Equinor’s core business on the Norwegian continental shelf depends on keeping offshore platforms and onshore plants running. Its stated ambition is to hold production there at 2020 levels through 2035, around 1.2 million barrels of oil equivalent per day, and it calls AI crucial to that goal (Equinor).
Rotating machines such as turbines, compressors, pumps and generators sit at the center of that problem. When one fails without warning, the facility can face a sudden shutdown. Equinor says such shutdowns can lead to flaring and increased CO2 emissions, as well as lost production and safety risk. Monitoring hundreds of machines by hand, across many installations, is repetitive work at a scale where people miss early signals.
What they did
Equinor connected sensors on its rotating equipment to machine learning models that look for changes signaling a coming failure or maintenance need. According to the company, the system now monitors over 700 rotating machines with 24,000 sensors across all facilities, offshore and onshore (Equinor). The approach is predictive maintenance: act on the warning before the breakdown, rather than on a fixed schedule or after the failure.
The technology is deliberately unglamorous. Hege Skryseth, executive vice president for Technology, Digital and Innovation, said: “We primarily use traditional machine learning on our operational data” (Journal of Petroleum Technology). Generative tools such as copilots, chatbots and agentic AI are available to employees for other tasks, but the measured value comes from models trained on Equinor’s own industrial data.
Predictive maintenance is one of a portfolio. Equinor says it has a range of AI solutions in use and has identified over a hundred new use cases (Equinor). The two it highlights alongside maintenance are well and field planning, where AI generates thousands of alternatives so experts can focus on the best proposals, and seismic interpretation, where it reports a tenfold increase in interpretation capacity and 2 million square kilometres interpreted with the AI tool in 2025. Equinor has not named vendors for these systems or disclosed what they cost.
Results
All figures below are Equinor’s own estimates, published in its January 2026 announcement and not independently audited.
According to Equinor, predictive maintenance of rotating machines, fed by 24,000 sensors, has created value of USD 120 million since 2020, through improved safety, more stable operations and fewer sudden shutdowns (Equinor). Across all its AI use, the company says AI contributed value creation and savings of USD 130 million for Equinor and its partners in 2025.
Skryseth put the cumulative figure higher: “Since 2020, we have realized values of over 330 million USD with artificial intelligence in industrial processes”, she said, with 130 million of it in 2025.
The planning use case produced one large, concrete win. In phase 3 of the Johan Sverdrup field, AI-driven planning found a solution that no one had considered, saving the partnership USD 12 million, the company said. Trade press, including the Journal of Petroleum Technology and Riviera Maritime Media, reported the same figures from the company’s announcement, without independent verification.
What didn’t go to plan
Public sources report no setbacks, reversals or regulatory problems with Equinor’s predictive maintenance program.
The open risks are about measurement. The value figures are self-reported estimates, and Equinor has not published how it calculates them: what counts as an avoided shutdown, what baseline it compares against, or how much of the value accrues to Equinor rather than to its licence partners. The 2025 total explicitly covers Equinor and its partners. Avoided failures are counterfactual by nature, so the number depends on assumptions the company has not disclosed.
Cost is also missing. Without the investment in sensors, data platforms and teams, the return on the USD 120 million cannot be judged from outside. Finally, Equinor’s broader AI ambitions, including agentic AI, have no published results yet; the proven value so far comes from conventional machine learning.
Lessons
- Start with the asset that stops the business. Equinor aimed its models at rotating machines, where a single failure can shut a facility and cause flaring.
- Use the simplest model that works on your own data. Equinor’s executive says the value comes mainly from traditional machine learning on operational data, not generative AI.
- Instrument at scale before you model. Monitoring over 700 machines needed 24,000 sensors; the data infrastructure is the precondition for the savings.
- Count value across a portfolio. Maintenance delivered most of the cumulative value, while one planning decision on Johan Sverdrup saved USD 12 million.
- Publish your method with your number. Equinor’s figures are credible in direction but cannot be checked without a disclosed baseline and cost.
Adapt it in your organization
Is this for you?
- You run a fleet of critical rotating or mechanical equipment whose failure halts production.
- That equipment already carries sensors, or you can fund instrumenting it.
- You have maintenance records that link sensor patterns to past failures.
- An operations center or reliability team can act on an alert within hours.
- You can estimate the cost of an unplanned shutdown per hour or per event.
A 90-day version
Pilot: pick the 20 machines with the costliest unplanned downtime, run anomaly-detection models on their existing sensor data in parallel with current maintenance, and have reliability engineers review every alert. Owner: head of asset reliability. Success measure: at least half of the failures that occur in the period flagged in advance, with an alert volume engineers can review each week.
Metric to track
Unplanned downtime hours on monitored equipment, and the value of avoided failures calculated with a published baseline. Equinor reports USD 120 million since 2020 across more than 700 machines. For an inspection case in manufacturing, see the GE Aerospace case.
Failure modes to pre-empt
- Alerts no one owns, so warnings arrive but maintenance does not change.
- Too many false alarms, so engineers start ignoring the models.
- Claiming value without a baseline, which invites the same scepticism Equinor’s figures face.
- Starting with generative AI where conventional models on sensor data would do the job.
Questions to take to your leadership team
- Which ten assets cost us the most when they fail without warning, and are they instrumented?
- Who would act on a failure prediction tonight, and with what authority?
- How would we calculate avoided-failure value in a way our auditors would accept?
Executive FAQ
How much has AI saved Equinor?
Equinor says AI created value of over USD 330 million since 2020, including USD 130 million for Equinor and its partners in 2025, by its own estimate.
What AI does Equinor use for predictive maintenance?
Mainly traditional machine learning on operational data from 24,000 sensors on more than 700 rotating machines, according to the company.
About the curator & AI disclosure
Curated by Luciano Ayres
Luciano Ayres is Director of Artificial Intelligence at ArcTouch, with more than 20 years of experience in software and technology leadership and specialized studies in deep learning. Connect with him on LinkedIn.
AI-generated briefing. This case was produced through an automated AI research and writing workflow curated by Luciano Ayres. Sources are linked for independent verification; summaries and analysis may contain errors and do not replace professional due diligence.