Big Oil Benefits From AI Twice — Here’s How
By Güney Yıldız
click here to read this article at Forbes.com
*this article was not written by Roseland Oil & Gas
AI data centers are colliding with grid limits—and oil majors are responding with off-grid, gas-fired generation built specifically to supply them.
In the process, they can profit from AI twice: deploying AI to raise efficiency and output, then selling gas power directly to data centers, a loop that could reinforce fossil dependence for years.
The playbook operates on two reinforcing tracks. Deploy AI across operations to boost production and cut costs. Build dedicated power infrastructure selling natural gas electricity directly to data centers. The more efficiently AI helps extraction, the more gas gets generated to power more data centers.
ADNOC perfected the model first
Abu Dhabi National Oil Company deployed 30+ AI tools generating $500 million in value and abating up to 1 million tonnes of CO2 emissions between 2022-2023. November 2025’s AiPSO launch with SLB was deployed across eight oil fields, with ADNOC stating an ambition to scale across 25 fields by 2027—a shift from pilots toward operational deployment.
The strategic breakthrough: ADNOC’s partnership with Microsoft and Masdar creates a symbiotic loop. The partnership frames a potential ‘energy for AI / AI for energy’ loop: Masdar and ADNOC are working with Microsoft on solutions to support data-center/AI infrastructure and deploy AI across energy operations—with renewables positioned as part of the supply mix—while ADNOC also positions itself as an energy supplier to the AI economy.
AI optimizes fossil fuel production, which powers data centers, which run AI that optimizes more production.
American majors are building the infrastructure now
Chevron is reported to be pursuing its first ~2.5 GW natural gas plant in West Texas (expandable to ~5 GW) for an undisclosed data center customer, targeting 2027 operations.. CFO Eimear Bonner framed the logic: “We’ve got the gas.” TThe company has large Permian natural gas volumes, and pipeline constraints can make on-site power sales more attractive than marginal disposal.
If built behind-the-meter / largely off-grid, projects like this can reduce dependence on grid interconnection queues and can function as a form of regulatory arbitrage, depending on jurisdiction—locking in gas infrastructure before stricter data-center standards arrive.
ExxonMobil announced a 1.5 GW plant in December 2024—their first power plant not serving own operations. The company has communicated a target of ~$15 billion in structural cost savings by 2027 (relative to a 2019 baseline), with digital/AI initiatives presented as one contributor among several. Their AI procurement system delivered 40x ROI ($19 million) in 2024, proving internal AI pays for itself while external power sales generate profit.
Saudi Aramco is building sovereign AI with its Metabrain model—trained on 90 years of data, currently 250 billion parameters targeting 1 trillion. The partnership with Groq to establish the world’s largest inferencing data center in Saudi Arabia signals transformation from oil exporter to digital infrastructure provider.
The mechanism: waste gas becomes profit
The business model works through three channels converting environmental problems into revenue.
Stranded gas monetization: Associated natural gas that would be flared powers co-located data centers. Crusoe Energy operates 40 such facilities. But this creates perverse incentives—data centers justify continued oil production generating the gas.
Behind-the-meter generation: Dedicated plants supply power directly without touching public grids. This accelerates permitting while avoiding renewable standards and 8+ year interconnection queues.
Carbon capture integration: ExxonMobil estimates decarbonizing AI data centers could represent 20% of the total addressable market for CCS by 2050. Aramco’s CCS partnerships with Linde and SLB use identical logic. The technology exists but operates at minuscule scale relative to claims.
The environmental cost compounds
Data center emissions from electricity use will rise from 180 million tonnes today to 300-500 million tonnes by 2035 globally—remaining below 1.5% of total energy sector emissions. Training GPT-3 consumed 1,287 megawatt-hours, generating 552 tons of CO2.
Fossil fuels currently meet 60% of data centers’ energy demand. The IEA’s Net Zero scenario eliminates all non-emergency flaring by 2030, yet oil companies are building infrastructure extending fossil economics through the 2040s.
Five signals reveal strategic coordination
The 2027 convergence: Every major targets 2027 for operational capacity—Chevron’s West Texas plant, Exxon’s commercial scaling, ADNOC’s infrastructure expansion. This coordinated timeline suggests recognition that the window closes when renewables scale sufficiently.
Customer secrecy: Chevron’s undisclosed customer and Exxon’s vague “powering the AI revolution” language indicate deals with hyperscalers who don’t want public association with fossil-powered AI.
Operational AI maturity gap: Gulf producers deploy agentic AI at commercial scale while Western majors remain in pilot-to-commercial transition. Western companies become AI power providers, not technology leaders—controlling energy while ceding intelligence.
Regulatory preemption: Building off-grid now creates fait accompli before governments impose clean standards. Private contracts lock in fossil assets through 30-year operational lives.
The flaring paradox: Companies claim data centers reduce flaring, but the IEA states the solution is stopping oil production that creates stranded gas. Data centers don’t reduce flaring—they create demand justifying combustion of gas that should stay unproduced.
Why this matters beyond 2026
The dual strategy isn’t opportunistic positioning—it’s structural transformation where AI justifies continued fossil fuel extraction rather than accelerating energy transition.
Oil majors aren’t transitioning through AI. They’re using AI to create markets necessitating continued extraction. Infrastructure being built now operates for 30+ years. Chevron’s 2027 plant, Exxon’s dedicated facilities, ADNOC’s Microsoft partnership—these aren’t bridge solutions. They’re bets that AI demand outlasts political will to decarbonize data centers.
Big Oil discovered AI’s power appetite isn’t a challenge—it’s their growth lifeline. Every efficiency gain in extraction generates capacity to power data centers running that AI. The more successful AI becomes, the more entrenched fossil infrastructure becomes.
Whether this represents strategic vision or dangerous carbon lock-in depends on which arrives first: efficiency breakthroughs cratering AI power demand, or the point where 30-year commitments make fossil-powered AI irreversible.
By Güney Yıldız
click here to read this article at Forbes.com
*this article was not written by Roseland Oil & Gas