The field,
now agentic.
The year energy stopped buying AI and started building it — read across our three most recent issues.
Subsurface Intelligence — Q4 2026
An H-kappa stack can be re-read as a likelihood and reported as a calibrated credible region at essentially no extra compute. The reason to bother is not tidiness. As a station's data quality falls the region inflates along the trade-off ridge while the stack maximum barely moves, so a pipeline that reports picks shows almost nothing at all for a network that has lost its constraint.
Subsurface Intelligence — Q3 2026
An exploration-scale full-waveform inversion project routinely consumes millions of core-hours, and the bill concentrates in one inner loop: solving the wave equation, forward and adjoint, over and over. This whitepaper maps the seismic HPC workload, locates the bottleneck, and lays out the four levers, surrogates, hardware fit, compute-aware orchestration, and sample efficiency, that are moving depth imaging from an overnight batch job on a CapEx cluster toward an interactive, uncertainty-aware, elastic-compute workflow.
From AI Customer to AI Builder
Operators start building what they used to buy: domain-native wellbore LLMs, basin-trained seismic foundation models, VeerNet log digitisation, agentic MLOps and drilling agents in the field, carbonate vug and fracture vision, and a first move into lignin.