AI and break-even: why large producers expect gains and small ones don't

Every producer is told that AI will lower costs. The useful question for a board is who expects it to, and by how much. The Dallas Fed's Q4 2025 Energy Survey, collected December 3-11, 2025 from 131 energy firms, asked E&P executives exactly that: by how much will AI lower your firm's break-even price for new wells over the next five years?

45 E&P firms answered: 8 large (10,000 barrels per day or more) and 37 small. The headline is not the average. It is the gap between the two groups.

Large producers: 25% expect no reduction, 38% expect $0.01-1.00 per barrel, 25% expect $1.01-2.00 and 13% expect $4.01-5.00. So 75% expect some gain. Small producers: 70% expect none, 11% expect $0.01-1.00, 8% expect $1.01-2.00 and 11% expect $2.01-3.00. So 30% expect some gain. Across all 45 respondents, 62% expect zero.

Economic reading, ours and not the survey's: AI in drilling and production carries heavy fixed costs: data engineering, integration with field systems, scarce talent. A firm with more wells and more rigs spreads those costs over more barrels. That is a scale economy, and it would explain why expectations rise with size.

A second mechanism is access. Large producers can negotiate with service companies and software vendors, and can run pilots on several assets at once. A small producer often meets AI as a feature inside someone else's product, where the gain accrues partly to the vendor or to the service company that drills the well.

Implications by type of actor. Large producers: expectations are high, so the risk is overpromising; fix a measurable target per asset. Small producers: expecting little may be rational, but 30% do expect gains, and those firms probably found a narrow use case. Service companies: the survey suggests small clients are skeptical, so selling on a priced outcome beats selling on a platform.

Limits and counter-arguments. The large-firm group is 8 executives, so one answer moves the result by 12 points. Expectations are not savings: executives may be optimistic or cautious. The question covers new-well break-evens, not operating cost or uptime, where gains may be easier to show. And a single respondent comment in the survey says AI already lowered effective well costs through general productivity gains, which a break-even question does not capture.

So what should an executive do? Treat AI as a scale question first. If you are large, set dollar-per-barrel targets. If you are small, buy a narrow capability, demand a measured baseline, and refuse multi-year platform contracts.