Dr. Aditya V Kashyap, AI and Innovation Leader, driving enterprise transformations through trusted strategy, governance and bold leadership.

getty

There has never been a cheaper time to solve a problem. Raw computation, the input that constrained a generation of business decisions, has fallen in price faster than almost any resource in economic history.

When OpenAI opened its first commercial language model to developers in 2020, it charged $60 to process a million tokens. Stanford's Human-Centered AI Institute found that the cost of querying a model scoring at GPT-3.5's level on the MMLU benchmark fell from $20 per million tokens in November 2022 to $0.07 by October 2024, more than a 280-fold reduction in roughly 18 months. Analysis that once took a research budget now sits inside a subscription.

And yet organizations are not becoming effective at anything like the rate their tools are becoming cheap. The gap between what a company can calculate and what it can accomplish has, if anything, widened. If the scarce input were computation, abundant computation should have produced abundant performance. It has not, which means we have been measuring the wrong constraint.

The confusion comes from treating two different problems as one. Computation solves an individual task: It forecasts demand, prices a risk or detects a tumor. Coordination is the separate work of aligning many such solutions, produced by different people, systems and institutions, into a coherent outcome. A firm can hold a perfect demand forecast and still miss the quarter because procurement, logistics and finance acted on different versions of it. The forecast was a computation problem, and it was solved. The failure was a coordination problem, and it was not. The two are different in kind, and cheapening the first does nothing for the second.

Once you look for it, coordination failure is visible wherever computation is abundant. American healthcare is the clearest case, because no sector has more data or more analytical firepower per dollar of output. A 2019 review in JAMA estimated that the U.S. wastes between $760 billion and $935 billion a year, roughly a quarter of all health spending, and attributed $27 billion to $78 billion of it to failure of care coordination: avoidable admissions, duplicated tests and complications that arise not from a lack of clinical knowledge but from the difficulty of aligning physicians, specialists, pharmacies, insurers and patients across fragmented workflows. The diagnosis is rarely the bottleneck. The handoff is.

Supply chains tell the same story. Manufacturers today hold forecasting capability that would have astonished planners a generation ago, yet they remain acutely vulnerable to disruption, because knowing what demand will be is far easier than synchronizing suppliers, carriers, customs authorities and ports across continents and jurisdictions. Financial markets rest on the same foundation. Their resilience is not principally a matter of computational speed but of coordinated architecture: exchanges, clearinghouses, custodians, payment systems and regulators executing against shared rules so that trust survives stress.

In March 2020, the U.S. Treasury market, one of the deepest and most heavily analyzed in the world, seized up not for lack of computation but because a rush to sell overwhelmed dealers' capacity to intermediate the trades, and only massive Federal Reserve intervention restored it. Markets rarely break for want of processing power. They break where the coordination between institutions gives way.

This points to a distinction I have come to think is the central fact of modern organizational life. Every institution runs on two kinds of intelligence. Computational intelligence is the capacity to process information and produce answers. Coordination intelligence is the capacity to align independent actors, each with their own incentives, information and authority, toward coherent collective action.

For most of the past century, these advanced together, and it was easy to mistake one for the other. They are now advancing at very different rates. Computational intelligence is being commoditized, available to anyone with a browser and a credit card. Coordination intelligence cannot be bought by the token, because it is not a tool an organization acquires but a capability it accumulates, expressed through its governance, incentives, decision rights, the standards that let its parts interoperate and the trust that lets independent actors rely on one another's work. That is why it stays scarce, and why it separates the organizations that compound value from those that merely process it.

The strategic consequence follows directly. The advantage of faster analysis is being competed away, because everyone buys from the same rapidly cheapening shelf. What remains is architectural. It belongs to organizations that design the institutions through which decisions flow: the incentives that keep independent actors pulling in one direction, the standards that let separate systems interoperate, the decision rights that let a choice get made without escalating through five layers.

None of this is downloadable; it is built, slowly, which is why two companies with identical technology routinely produce different results. The chain runs from computation to information to decision to coordination to execution, and it is the coordination link, not the computation link, that decides whether collective advantage emerges at the end.

There is a longer pattern worth naming. Every major era of economic development has been defined by the conquest of a single dominant constraint. The industrial age drove down the cost of production, and its defining institutions, including the factory, the assembly line and the integrated corporation, mastered it. The information age drove down the cost of computation, and its winners turned cheap processing into scale. Each era's binding constraint, once loosened, ceased to be where advantage lived, and the frontier moved.

That is what is happening now. Computation has not become unimportant, but for a growing number of organizations, it is abundant enough that it is no longer the primary bottleneck, and the constraint behind it has moved into view. The defining organizations of this century will not be the ones that compute most, because computation will be available to all of them. They will be the ones that master the older, harder discipline of institutional design: building the governance, incentives and standards through which independent actors become capable of acting as one.

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?