Own Your AI Stack: a whitepaper for energy operators
AI in oil and gas will hit $17B by 2030, yet half of enterprise pilots never ship. The fix is engineering execution, not better models.
by Tannistha Maiti · Senior AI Researcher
Get the full reportPDF · 18pp
The AI in oil and gas market is projected to grow from $7.2B in 2025 to $17B by 2030, yet up to half of all enterprise AI projects never reach production. The cause is not model quality. It is the missing engineering layer between a notebook that works and a workflow that ships.
I
The opportunity
AI spending in oil and gas is compounding faster than almost any other category of operational technology in the sector, and this is a story about who owns the intelligence layer on top of the world’s most expensive industrial data.
Market context
$7.2B
AI in oil & gas market, 2025
$17B
Projected market, 2030
18.5%
CAGR through 2030
61%
Share held by upstream
II
The technical landscape
A working model is roughly 15% of the journey. The other 85% is the engineering stack underneath it, six layers that turn a notebook into a workflow that ships inside the operator’s security perimeter.
Model · 15%
Engineering stack · 85%
VIPlatform & deploymentInside the operator's security perimeter; full ownership transfer at scale.
VAgentsUpload, run, visualise, report. Agentic orchestration of multi-step subsurface workflows.
IVAI / MLHulde foundation model plus five domain modules. Versioning, drift, MLOps.
IIIData unificationSubsurface, operational, and sensor data on a single ontology.
IIData engineeringIngestion and provenance for LAS, SEG-Y, raster logs, image logs.
IHPCSeismic-scale compute provisioned for inference and retraining.
Pilots don’t stall because the model is weak. The working model is only ~15% of the journey; the other ~85% is a six-layer engineering stack (HPC to Data engineering to Data unification to AI/ML to Agents to Platform/deployment), and a project ships only when every layer below the model is built to production grade. Drag the build line up the load-bearing column: with all six built the model reaches the production ceiling; with any gap below it the model detaches into POC purgatory, the ~50% that never ship. The ~15%/~85% split, the six layers and the ~50% figure are the whitepaper’s own; the equal-sixths column sizing is schematic.
Closed vendor platform
Operator subsurface data trains shared models
Reservoir intelligence becomes vendor IP
Workflow redesign owned by vendor services arm
Switching costs accumulate with every well ingested
Operator-owned full stack
✓Open-weights foundation model (Hulde, RAIL licence)
✓Data and models remain inside operator perimeter
✓Workflow redesign owned by the asset team
✓Capability transfer is the explicit end state
A closed vendor platform and an operator-owned stack diverge on one axis: who owns the intelligence as wells accumulate. Every well ingested into a shared-model platform converts more operator reservoir IP into vendor asset and ratchets switching cost up, one-way: ingested data can’t be un-ingested (drag the slider back and the lock holds). Toggle to the operator-owned stack and IP stays inside the perimeter, switching cost stays flat. The 4,700-well portfolio vs 200-well training set and the named platforms are the whitepaper’s own; the IP-bled fraction and cost magnitude are schematic (the direction is sourced).
III
Our approach
EarthScan is an AI engineering company. We build and operate the full stack, six capability layers from HPC up to deployment, and we hand the keys back to the operator at scale.
“A general-purpose LLM optimised to sound confident is the wrong tool for a million-dollar well decision. For subsurface workflows, hallucination resistance is a prerequisite, not a feature.”
Five production modules
Vug Detection
Fracture & Bedding Detection
VSP Analysis
Well Log Detection
Geomechanical Analysis
Hulde foundation model
Hominis family, open weights
RAIL licence, no data lock-in
Procedural, not narrative
Hallucination-resistant
Six capability layers
HPC
Data engineering
Data unification
AI / ML
Agents
Platform deployment
IV
Case examples
Operator B, a Gulf region independent, engaged EarthScan to deploy Well Log Detection across a 47-well evaluation portfolio, cutting evaluation time from roughly 14 hours per well to under 90 minutes with geoscientist review retained for sign-off.
Operator B, Well Log Detection (47 wells)
Manual evaluation per well
14 hrs
Agentic workflow per well
<90 min
About an 89% time reduction, roughly 600 expert hours freed across the portfolio, with the operator retaining the platform, the weights, and the data.
The engineering layer didn’t replace the geoscientist, it moved expert time off rote evaluation onto judgment. On Operator B’s 47-well portfolio, per-well Well-Log-Detection evaluation fell from 14 hours to under 90 minutes (~89%), reclaiming ~600 expert hours in 8 weeks, while geoscientist review was explicitly retained for anomaly cases and final sign-off. Sweep across the portfolio: each 14-hour bar collapses to a <90-minute sliver and an amber “review retained” band rides every converted well, never shrinking to zero.
V
Implementation roadmap
We deliver in three phases: Discover, Pilot, Scale, paced to the operator’s risk appetite rather than the vendor’s revenue model.
PilotWeeks 5-12
One production module integrated into the asset workflow: ingestion and provenance, a domain model on Hulde, an agentic workflow, MLOps scaffolding.
VII
Conclusion and next steps
The decision in front of operators is not whether to adopt AI; it is whether the AI deployed on your asset is owned by you or by a platform vendor whose interest is maximising billable services on your subsurface data.
By the numbers
$7.2B→$17B
AI in oil & gas, 2025 to 2030
~50%
Enterprise AI projects that never ship
75%
Seismic build-time cut by ADNOC's ENERGYai
12 wks
Discover + Pilot to production
References
[1]Knowledge Sourcing Intelligence. AI in Oil and Gas Market (2025). Market sized at $7.2B (2025), forecast $17B by 2030, 18.5% CAGR.
[2]Mordor Intelligence. AI in Oil and Gas Market (2025). Upstream accounts for 61% of market; ADNOC ENERGYai cut seismic model-build times by 75% in 2024.
[3]International Energy Agency. Energy and AI Special Report (2025). Missing internal expertise identified as the dominant adoption barrier.
[4]IDC (2025). Up to 50% of enterprise AI projects collapse before production.
[5]Devon Energy, CERAWeek 2025. Reported 25% productivity gains from AI drilling optimisation.
[6]McKinsey & Company. The State of AI (2025). Workflow redesign identified as the single strongest EBIT correlate of AI value capture.
This is the short version
The full 18-page analysis, in your inbox
The complete roadmap, the risk framework, the reference architecture, and the full data behind every exhibit above, sent as a PDF. Or read the full whitepaper.
