Many AI budgets today look a lot like oil businesses before Rockefeller.
A company spends $200,000 a month on AI infrastructure. The invoice arrives from OpenAI, Anthropic, or Bedrock. Finance sees the total. Engineering sees usage. Beyond that, the picture often becomes much less clear.
Leadership can usually answer how much was spent. What becomes harder to answer is what the organization received in return. Which teams generated the spend? Which products depended on it? Which workflows created value? Which ones consumed resources without producing meaningful outcomes?
That wasn't a major problem when AI was primarily experimental. Teams were testing ideas, building prototypes, and learning where the technology fit inside the business. Efficiency wasn't the primary objective because learning itself had value.
As AI moves deeper into operations, the expectations begin to change.
Support teams use it to resolve customer issues. Engineers use it to accelerate development. Product teams embed it directly into customer experiences. Once AI becomes a meaningful operating expense, leadership inevitably starts asking a different set of questions than they did during the experimentation phase.
The challenge is that most organizations still measure AI through consumption. They know how many tokens were used and what the invoice says. Those metrics are useful, but they're incomplete. They describe activity rather than outcomes.
Rockefeller would have recognized the problem immediately. The issue isn't spending money. The issue is spending money without a clear unit of accountability.
For Rockefeller, that unit was the barrel. Once he understood the economics of a barrel moving through the business, he could identify inefficiencies, compare facilities, and make better decisions than his competitors.
For AI, the equivalent unit is still being defined. Tokens tell us how much intelligence was consumed, but they don't tell us how much useful work was completed. What matters is understanding the cost of a resolved support ticket, a completed analysis, a successful agent run, or a feature delivered faster because AI was involved.
The organizations that manage AI effectively over the next decade will likely be the ones that make that transition. They'll stop viewing AI primarily as a consumption metric and start treating it as an investment that needs to be connected to business outcomes.
Rockefeller didn't build Standard Oil by obsessing over spending less money than everyone else. He built it by understanding where every dollar went and what it produced. As AI becomes a larger part of operating budgets, that same discipline is becoming increasingly important.