ADNOC deploys AI on 120 drilling rigs, all announced numbers are “human” figures, not barrels

ADNOC deployed the Real-Time Operations Centre platform built on SLB’s DrillOps software across more than 120 land and offshore drilling rigs. It is not a trial and not limited to a single site: it is a full fleet. Every figure the company announced is a “human” number: engineering effort, response time, avoided downtime days, and no production figure for barrels or recovery rates.

What ADNOC announced and what the numbers actually mean

The company says engineering effort has dropped by 30 to 40 percent, that a single engineer now supervises two to three times as many rigs as before, and that incident response time has improved by 4 to 12 hours, avoiding up to two days of rig downtime per incident. The figures are self-reported, with no published methodology, no disclosed baseline, no contract value, and no ownership terms for the models trained on ADNOC data. Nevertheless, it is the largest publicly announced industrial AI deployment in the Gulf oil sector, and the difference is fundamental: most regional announcements describe intent, this one describes a fleet.

Savings are in people, not oil: the real story

There is no claim about barrels, recovery rates, or reservoir performance. Every metric describes how many people the drilling fleet needs and how quickly they notice a growing error. Deep-water and complex onshore drilling engineers are globally scarce and expensive everywhere. A national company aiming for five million barrels per day of crude energy by 2027, while expanding gas and petrochemicals in parallel, faces a hiring challenge before it faces a geological challenge. Software that lets the existing team cover twice or three times the workload removes a constraint that money alone cannot solve.

Marine downtime is the costliest disclosed item, and the 4-to-12-hour improvement in incident detection accounts for most of the savings. Detecting a problem before it escalates is unremarkable work, and that is where the money lies.

Sovereign-cloud clause: the second announcement hidden within the first

The deployment runs inside ADNOC’s sovereign cloud in the United Arab Emirates. Rakish Gaji, head of digital at SLB, described the environment as “the scalable digital foundation for workflow that enables AI across one of the industry’s largest drilling fleets.” ADNOC added that keeping sensitive information within Emirati jurisdiction enhances data security and operational resilience. Stripped of rhetoric, this is a statement about who owns the real-time operations data of a national asset. Drilling telemetry is among the most valuable commercial information an oil company holds: it reveals well performance, reservoir behavior, operational weak points, and broadly maps national production capacity. The Gulf has spent three years learning that access to technology can be limited by another country’s export policy. Applying that lesson to operations data before dependency forms aligns with how the region has approached buying compute, and it is the clause a competitor should study instead of efficiency ratios.

It did not arise from a vacuum: a multi-year program

The Panorama command centre dates back to 2017. The RoboWell autonomous well-control system, developed with the joint AI project AIQ, is an agentic system built with G42 and Microsoft around a large language model and specialized task agents, expected to cut the time to build complex geological models by up to 75 percent. Compared with that timeline, the Real-Time Operations Centre is not a leap. It is the first component in the program that reaches every rig in a single rollout, and accessing everything is the hard part. The trial was a purchase decision. Full-scale deployment changes the operating model, with training, escalation procedures, and responsibilities rewritten around a tool that did not exist last year.

What was not disclosed: the most important

Everything financial: no contract value, no capital cost, no economics split between ADNOC and SLB, and no statement on who owns the models trained on ADNOC drilling data. This last omission is the most important. Sovereign hosting determines where the data resides. It does not determine who benefits from what the system learns. None of this is independently verified. The efficiency figures are the operator’s own numbers, calculated with a unpublished methodology, against a baseline that no one outside the company sees. This is normal in corporate technical disclosures, and it remains a reason to treat the figures as directional indicators.

What will prove it works

Three observable things. Whether ADNOC’s drilling operating costs per rig move in the direction suggested by the efficiency claims, which will appear in results rather than press releases. Whether the engineer-to-rig ratio is disclosed as an actual number rather than a multiplier. And whether SLB sells the same configuration to another national oil company in the region, because a second buyer is the strongest evidence that the initial deployment succeeded. A fourth signal would be more telling than all of them: if ADNOC begins describing incidents captured by the system in detail, rather than savings percentages, that would indicate confidence rather than marketing.