S06.1 · Energy, Power & Climate
Exploration and production of oil and gas reserves, sized by capex given full commodity-price exposure; AI is compressing subsurface and drilling costs.
Upstream oil and gas production — exploration and extraction — is sized by capital investment rather than revenue, and for good reason: revenue in a price-taking segment measures the commodity, not the business. Global upstream capex reached $570B in 2025 (IEA World Energy Investment 2025), up from roughly $450B in 2021, and the IEA base case puts forward growth at only 1-2% a year, with upside tied to LNG-export and power/datacenter gas demand. The structure — fragmented, price-taking — is not changing. What is changing is technical: agentic AI is compressing the cost of subsurface interpretation and drilling engineering, and moving competitive advantage away from geoscience talent toward data and compute scale.
Access, not capital, decides where an operator can play. National oil companies control roughly 60% of global reserves, and the regime of royalty terms, leasing rules and NOC-versus-IOC access governs entry basin by basin. Production concentrates in the US Permian, the Middle East (led by Saudi Aramco and ADNOC), offshore Brazil and Guyana, and West Africa. The producer base itself is fragmented — the top 20 companies, a mix of NOCs and international majors, hold roughly 40% of global output, with a long tail of thousands of independents beneath them.
Upstream sits at the head of the value chain: it feeds midstream gathering and pipeline networks and downstream refining, and buys drilling and completion services from the oilfield-services industry. The economics are capital-intensive and fully exposed to WTI and Brent pricing, with break-evens of $35-55/bbl for US shale against $20-30/bbl for Middle East conventional production. Permitting and leasing add a further gate. Revenue is largely merchant rather than contracted, so producers wear the price cycle with little insulation.
Producers have two directions to diversify, and they solve different problems. Integrating into midstream gathering and processing is defensive — it secures well takeaway. Expanding into LNG export or new international E&P acreage is expansionary, aimed at reserve replacement.
AI's effect runs through cost, not revenue — in a price-taking segment there is nowhere else for it to go. Agentic AI automates subsurface interpretation, drilling-parameter optimization and reservoir modeling, compressing the geoscience and engineering headcount that has been a major fixed cost of exploration. The gain lands as lower finding-and-development cost, and the moat shifts in step: away from proprietary seismic-interpretation talent, toward the data and compute scale needed to run these models well. Drilling-automation tools are already in commercial use; broader adoption of subsurface AI is expected within 2-5 years.