Strange New Worlds
Every company gets more efficient. No company gets permanently more profitable. That is the story of AI and operating margins.
The prevailing thinking amongst CEOs is AI can cut costs by reducing head count, while raising output, and expanding operating margins expand. This is already happening, and they’re not wrong, BUT…
A high operating margin is not a stable state – it’s a signal. It signals that there is money to be made in the market. Capital follows that signal. Newbies show up with lower price points. Existing competitors ramp AI investment to close the efficiency gap. Customers gain bargaining power as alternatives multiply. The margin that AI created starts compressing before your next annual report is filed. Sustaining innovations spread through an industry, raise the productivity floor, and then get priced into competition. Disruptive innovations produce durable excess returns for innovators because they create new competitive terrain. That’s not an opinion. That’s science, the same as water’s wet, the sky is blue, and the Dallas Cowboys suck. The confusion comes from the fact that most CEOs fundamentally don’t understand how computers work, let alone AI.
Terminal operating margin – the margin a business can sustain in perpetuity – is fundamentally a competitive equilibrium concept. It reflects what a company earns when the market has fully adjusted to its cost structure and its competitive set.
If AI raises efficiency for all players in an industry, the terminal margin should not change much. The productivity gains become the new baseline. Prices adjust downward as competition passes efficiency through to customers. Input costs for AI tools themselves rise as vendors capture value. The equilibrium reasserts itself at higher productivity but similar profitability.
Here is the constraint that rarely gets named: AI is trained on historical data. It is, by design, a system for solving known problems well. Feed it enough past examples and it will find patterns, compress costs, and optimize processes at a scale no human team could match. What it cannot do is identify a market that does not yet exist, or recognize a business model that has no precedent in the training data. Those discoveries have no historical examples to learn from. They require something different. They require…wait for it…humans.
Specifically, the kind of human who can sit with ambiguity, notice what customers cannot articulate, translate the incoherent ramblings of execs, and imagine competitive terrain that hasn’t been drawn yet. The irony of the AI era is that the companies who treat human employees primarily as costs to automate away are optimizing their sustaining innovation. They will get more efficient. They will not get more disruptive. The companies that invest in the human capacity to explore strange new worlds, to seek out new life and new markets, to boldly go where no AI has gone before, those are the ones building a durable advantage.
