AI’s Real Constraint . . . it is not what your think.

Todd Colpron
Dec 14, 2025
3 min read

There is a habit, common in fast-moving industries, of mistaking the most visible problem for the most important one. In artificial intelligence, that habit has taken the form of an obsession with chips. We speak of shortages, races, allocations, and breakthroughs, as if the future of AI were decided solely by silicon. It is not.

Artificial intelligence does not ultimately run on chips. It runs on power, land, and permission. And those are in far shorter supply.

A recent analysis from Goldman Sachs asks whether there is enough data-center capacity to meet the demands of AI. Their answer, cautiously optimistic, suggests that on paper the world can build fast enough. That conclusion is not wrong. But it is incomplete. Capacity that exists only in forecasts, filings, and announcements is not capacity at all. It is intention.

What matters is whether AI infrastructure can be delivered—reliably, continuously, and on time.

When the Old Model Stops Working

For many years, data centers were treated as a form of specialized real estate. They were places—important places, to be sure—where servers lived. Power was assumed. Permits were routine. Time was manageable. AI has undone this arrangement.

Modern AI workloads consume extraordinary amounts of electricity, without pause and without tolerance for interruption. They do not wait politely for grid upgrades or regulatory alignment. They demand constancy. This shifts the center of gravity. The data center ceases to be a building first and becomes an energy system that happens to house compute.

In this new order, square footage recedes in importance. What rises instead is access to electrons, the certainty of delivery, and the legal right to operate at scale. Compute follows power, because it must.

Why So Much Exists, and So Little Is Available

From a distance, the industry looks busy enough. Announcements pile up. Pipelines swell. Megawatts multiply in presentations. And yet, on the ground, AI developers encounter the same problem again and again: delay.

The reasons are not mysterious.

Electric grids are strained and aging. Transmission takes years to approve and longer to build. Interconnection queues lengthen while demand accelerates. In many regions, land can be purchased quickly and capital can be raised readily, but electricity remains a promise rather than a fact.

Permitting, once a footnote, has become decisive. Environmental review, zoning, community consent, political will—each adds weight. Together, they determine whether a project can proceed or merely persist as a plan.

Time, meanwhile, exacts its quiet toll. AI economics are unforgiving of delay. A model trained too late, a platform launched behind schedule, an inference cluster that comes online after the market has moved on—these are not inconveniences. They are failures of timing, which is another word for failure.

The Return of Practical Energy

Against this backdrop, a certain pragmatism is returning to the energy conversation. It is not loud, and it is not ideological. It is driven by arithmetic.

AI requires power that is steady, abundant, and acceptable to regulators. That combination narrows the field. Nuclear energy, advanced gas generation, and hybrid systems—once dismissed or deferred—are reentering the discussion not as preferences, but as necessities. The future of AI will be powered less by slogans and more by what works.

Capital Learns to Think Like Infrastructure

As the shape of the problem changes, so does the shape of capital.

The winners in AI will not be determined by software excellence alone. They will be shaped by those who understood early that this was an infrastructure story, not a product cycle. That land matters. That permits matter. That power is destiny.

This favors investors with patience, operators with foresight, and structures built for endurance rather than speed alone. It rewards those who understand that long-term value is created not by abundance of ambition, but by certainty of delivery.

Where Eliakim Capital Stands

At Eliakim Capital, we approach AI as what it has become: a foundational layer of modern civilization, akin to electrification or broadband. Our work sits at the junction where compute meets power, where capital meets execution, and where plans become operating assets.

We focus on what makes AI real—energy systems that function, sites that are entitled, capital that is aligned with duration rather than fashion. In a landscape crowded with projections, we concern ourselves with delivery.