How is AI being deployed in the oilfield? Is it unlocking more production, or deflating cost-curves? And who benefits? Answers to all of these questions are evidenced by reviewing 40 case studies of AI in the oil industry, from the past year, as described in technical papers from the Society of Petroleum Engineers.
“There’s an app for that”. The oil patch contains a vast number of domain-specific problems. Hence most of the case studies in our data-file address some highly specific problem, by absorbing large quantities of training data, then automating tedious and manual processes, to be over 80% faster, or sometimes instantaneous. LLM interfaces also help, because historically engineers spend 60% of their time on low value-add data-wrangling.
Examples of deploying AI in the oil industry include automating drilling (the single most targeted area); reservoir simulations and well placement; optimizing gas-lift, water-flooding, or gas compression; tailoring drilling fluids; predicting wellbore instability, casing corrosion, or screen-outs in hydraulic fracturing; benchmarking maintenance; generatively designing offshore facilities, better drillbits, or decommissioning plans; using video to detect HSE violations; optimizing offshore helicopter routes or engine idling on offshore vessels; digitizing legacy documentation.
“Thousands of AI models” are seen being trained in our forecasts for AI energy consumption. This was speculation, when we originally made our forecasts!! But note each of the examples above is going to require its own model being developed/refined. In one fascinating case study, Aramco joins forces with NVIDIA for complex reservoir simulations, on H100 GPUs, with inference times dropping to “milliseconds” rather than hours-days. Another study sees AWS optimizing gas processing facilities in Australia.
“I’d pay for that”. Oil Services are among the most prolific developers of these AI apps. One company stood out in particular. Many of the studies develop rapidly query-able LLMs, enabling junior engineers to take on more work while following best practice. In another study, for example, casing corrosion risks are inferred, automating a task that used to take experts 30+ hours. You could even imagine a suite of AI “apps” uplifting the revenue potential for select Oil Services.
Production uplifts are also seen. Real-time optimization of gas-lift has raised output 1-4% in case studies from two leading producers. Better well placement can uplift EOR by 6.5% in a study from an Oil Services firm. Preventing lost time incidents could avoid 20-60% of today’s downtime.
NOCs were highly active in developing these AI apps, especially ADNOC and Aramco. Widespread deployment also connotes higher production. The data-file contains a breakdown of the companies that seemed most active in deploying AI to the oilfield, across producers, services firms and venture-stage technology companies.