Frontier AI for Oil Reservoir Exploration

Overview

AI is changing how oil is found, and prices are being set before the evidence is in. This edition classifies 23 public AI claim records by evidence class, examines who may lawfully train a model on licensed seismic data, sets out every vendor and operator disclosure, and works through all 31 deals, contracts and financings since 2022 with the few prices anyone disclosed. It then gives the working tools an investment committee needs: a scorecard covering eighteen vendors, a break-even worksheet, a sixteen-risk register and 49 diligence questions. Where independent evidence does not exist, the guide says so and shows what was searched.

What this edition covers

Fourteen parts: where AI enters the chain from seismic survey to first oil; what "frontier" means, sorted by maturity; an evidence ledger of 23 public claim records; who owns what the models learn from; vendors and what their numbers include; operators' in-house programmes and the computing arms race; the transactions and the few disclosed prices; seven market-size estimates reconciled; the economics of a better prospect; the rules on reserves, AI law, chips and sovereignty; a sixteen-risk register; geothermal, minerals and carbon storage; the investment case; and twelve recurrent interpretation pitfalls.

Who this is for

Equity analysts covering oilfield services and seismic, venture and growth investors, private-equity and infrastructure funds, corporate development teams at oil and national oil companies, and lenders against reserves.

What you get

• Six evidence classes applied to every claim, including one we looked for and did not find
• An evidence ledger of 23 public AI claim records, classified
• The data-rights analysis: who may train a model on licensed seismic, across six release regimes
• All 31 deals, contracts and financings since 2022, with disclosed prices and their limits
• Tool 1 · Vendor Scorecard — eighteen vendors against five tests
• Tool 2 · Break-even Worksheet — what a product must achieve to pay for itself
• A sixteen-risk register: who carries each risk, and what to ask
• A 49-question diligence checklist with pass or concern fields
• Registers of 39 specialist companies and 117 SEC filings

Table of contents

1. Executive summary: The thesis, the numbers that frame it and the seven findings
2. Part 0 — The Map: Where AI enters the chain from survey to first oil
3. Part 1 — What "Frontier" Means: The technology, sorted by maturity
4. Part 2 — The Evidence Ledger: 23 public claim records, classified
5. Part 3 — The Data Is the Moat: Who owns what the models learn from
6. Part 4 — The Vendors: Capability, disclosure and what the numbers include
7. Part 5 — The Operators: In-house programmes and the computing arms race
8. Part 6 — Deals and Valuation: The transactions, and the few prices anyone disclosed
9. Part 7 — Sizing the Market: Seven estimates, three quantities, one reconciliation
10. Part 8 — The Economics of a Better Prospect: Where the value would come from
11. Part 9 — The Rules: Reserves, AI law, chips and sovereignty
12. Part 10 — The Risk Register: Sixteen risks, who carries each, and what to ask
13. Part 11 — Beyond Oil: Geothermal, minerals and carbon storage
14. Part 12 — The Investment Case: Where exposure can be owned, and how to underwrite it
15. Part 13 — Pitfalls: Twelve recurrent interpretation risks illustrated by the record
16. Tool 1 — The Vendor Scorecard: Eighteen vendors, five tests
17. Tool 2 — The Break-even Worksheet: What a product must achieve to pay
18. The Diligence Checklist: 49 questions, tear-out
19. Appendices — Timeline, glossary, method, sources and registers: 2000–2026 timeline; method and limitations; 333 sources; 39 companies; 117 SEC filings

Research and sources

Includes 333 numbered references, each linked and tagged by evidence class, plus a timeline 2000–2026, a glossary, a method and limitations note, a register of 39 specialist companies and an EDGAR corpus of 117 filings. Data cutoff 18 September 2026.

Licensing

Single-user licence. For use by the named licensee only. Redistribution, resale, posting to shared drives or internal circulation beyond the named licensee is a breach of licence. Team licences covering up to five named users are available.