The Algorithm
in the Oil Field

The size of the AI-in-oil-and-gas market depends almost entirely on how you scope it. Major analyst firms have published estimates ranging from $2.9 billion (Mordor Intelligence) to $5.29 billion (Grand View Research) to $6.69 billion (GM Insights) for the 2024 base year, with 2030–2034 forecasts ranging from $7 billion to $25 billion (Grand View · GM Insights). CAGRs cluster in the 13–18% range.

The spread is informative on its own. Energy AI today means several different things — seismic interpretation models, drilling optimisers, refinery RL controllers, methane-leak satellite analytics, grid forecasting, ESG reporting copilots — and analyst firms include different subsets in their addressable-market definition. Read any single number with that caveat in mind.

What is not in dispute: the spend is real, and the named operators (Aramco, ADNOC, BP, Shell, Equinor, ExxonMobil) are publishing concrete deployments and value-realised figures in their annual reports.

Of all the integrated oil companies, Saudi Aramco publishes the most concrete in-production AI figures. The company’s 2025 Sustainability Report states that its digital-transformation programme has generated over $1.8 billion in realised value from AI initiatives, spanning equipment reliability, reservoir recovery, safety, emissions, and infrastructure protection (Aramco Sustainability Report 2025).

A specific example Aramco discloses publicly: at the Uthmaniyah Gas Plant, one of the largest gas processing facilities in the world, the introduction of drones and wearable inspection technology has cut inspection times by approximately 90% (Aramco — Digitalization). The same digital programme combines machine learning, predictive analytics, computer vision, and generative AI across upstream and downstream.

Aramco has also taken a stake in the HUMAIN sovereign-AI venture launched in 2024 and operates a dedicated Google Cloud region in Saudi Arabia.

In November 2024, ADNOC announced ENERGYai — described by the company as the energy industry’s first deployment of autonomous, agentic AI in production. ENERGYai was built jointly with G42, Microsoft, and AIQ, and was trained on 80 years of ADNOC operational data. The agents are designed to autonomously perform complex tasks ranging from seismic analysis to real-time monitoring (ADNOC press release · Rigzone coverage).

The same November 2024 announcement covered a broader Microsoft–ADNOC–Masdar partnership on low-carbon AI data centres, anchoring the UAE’s positioning as both an energy producer and a sovereign AI infrastructure provider. AIQ joined Microsoft’s Cloud AI Partner Program and selected Azure as its primary cloud platform.

ENERGYai entered real-world testing at end-2024, with a three-year deployment programme. The combination of Aramco’s value-realised disclosures and ADNOC’s agentic-AI launch puts the GCC operators clearly ahead of most Western majors on enterprise-AI publicness — they are putting numbers in their reports.

The energy transition is fundamentally a forecasting and balancing problem — variable renewables only get fully usable when they can be scheduled. The cleanest publicly documented AI contribution to that problem remains DeepMind’s 2019 wind-energy study: a neural network trained on weather forecasts and historical turbine data made 36-hour-ahead predictions for a 700 MW Google wind fleet in central Oklahoma (90+ turbines), enabling day-ahead delivery commitments that boosted the economic value of wind energy by ~20% versus the no-commitment baseline (Google DeepMind blog).

The reason this study still matters is that the 20% number is independently verifiable, the methodology is published, and the result has been broadly replicated since across utility-scale renewable operators. The general technique — pairing weather-model output with operational telemetry to make better day-ahead commitments — is now standard practice at Ørsted, NextEra, EDF, Iberdrola, and most major utility-scale wind operators.

The same kind of forecasting machinery underpins grid balancing, virtual power plants, and demand-response programmes. National Grid (UK), CAISO, and the Saudi grid operator all run short-term load and price forecasting models, though the specifics are rarely made public.

The strategic point about the GCC operators isn’t the headline investment numbers — it’s that they are building and publishing the operational AI playbook for a hydrocarbon industry that has to defend its margins while funding the transition. Whichever playbook works at Aramco and ADNOC scale becomes the default for the rest of the industry within five years.

What is publicly documented today: Aramco’s $1.8B realised value across the digital programme, the Uthmaniyah 90% inspection-time reduction, the ENERGYai agentic-AI launch, the HUMAIN sovereign-AI venture, the ADNOC–Masdar–Microsoft data-centre partnership. None of that is forecast — all of it is in 2024–2025 corporate filings and press releases.

The Western majors (BP, Shell, ExxonMobil, Equinor, TotalEnergies) all run substantial in-house AI programmes too, but they publish less granular value-realisation data. That’s a reporting choice, not a capability gap. The interesting question over the next 24 months is whether the rest of the industry follows the GCC pattern of putting AI-value receipts directly in the annual report.