Seismic player working with national oil company to build large language model that will use TGS data to develop drilling prospects, the company's imaging and technology vice president Wadii El Karkouri tells Upstream in an exclusive interview.
Leading marine seismic company TGS is working to combine its vast multi-client data library with generative artificial intelligence and cloud computing in a project that it hopes will result in higher exploration success for operators across the world.
TGS is developing seismic foundation models, which the company’s executive vice president of imaging and technology, Wadii El Karkouri, likens to the “Chat GPT for the subsurface”.
“I think the biggest technology change is going to come from a new way of doing geophysical modelling,” El Karkouri told Upstream.
“What we are looking at right now is to collaborate with some cloud companies so that we can bring the best and largest multi-client data library and the best cloud technology together to generate new exploration ideas that nobody has thought about before.
The company is working with an unnamed national oil company to build a large language model that will use TGS data from 40 seismic surveys from around the world to generate tasks including prospect generation, mapping reservoir channels, mapping faults and looking for some prospectivity indicators.
Norwegian seismic giant TGS has acquired 65% to 70% of global multi-client data, El Karkouri notes, a breadth of information that could be put to work alongside cloud technology and generative AI.
“Right now, you have one person in a room collecting all the data and coming up with a recommendation. Going forward, it will be leveraging generative AI and general subsurface data, multi-client data, to generate recommendations based on previous projects,” he said.
While TGS has been combining multi-client libraries, cloud knowledge and generative AI for internal research purposes, this project is the company’s first commercial agreement and El Karkouri hopes it can provide a testing ground for future deals.
For El Karkouri who has worked for operators, marine seismic companies and cloud computing companies across the whole seismic value chain, he sees the benefit of drawing on the expertise of all sectors to drive technological advances.
“The more digital we bring into subsurface exploration, the more potential ideas for drilling better wells will be generated and so operators can increase their probability of success,” he said.
“For example, if you are trying to do salt identification, you would train your large language model around the Gulf of America and Brazil so that you can identify salt in Angola; the common factor has to be around geological features because that's the value in identifying those in a much faster and a more accurate way.”
The company is currently working with Google Cloud but is also weighing other potential providers.
Harnessing AI and cloud computing
El Karkouri is bullish on the project's potential to increase exploration success.
“If we assume that the success factor of exploration in the Gulf of America is below 20% today, but if we can bring that up to 30%, we are talking about billions of dollars of resources that could be found,” he said.
Harnessing AI and cloud computing will drive cycle time and cost reductions in seismic processing and interpretation, but El Karkouri argues for its deployment across the whole exploration value chain if it is to have a real impact on costs.
Companies across the energy sector are looking at how they can use AI to increase efficiency and success rates, shorten cycle time and save costs, with BP and Abu Dhabi National Oil Company putting AI at the heart of their operations.
But for real innovation, El Karkouri says the seismic industry should borrow from the technology world’s speed of innovation where research and development announcements are made almost weekly.
“We tend to wait and see for a little longer, but wait and see is not a strategy anymore when the average reserve life for companies today is between seven and 10 years and it takes about five years to discover something and produce it,” El Karkouri said.
The original article was published on Upstream Online on June 5, 2025.