Phantom data centers: What they are (or aren’t) and why they’re hampering the true promise of AI

Within the age of AI, public utilities at the moment are going through a brand new, sudden downside: Phantom knowledge facilities. On the floor, it might appear absurd: Why (and the way) would anybody fabricate one thing as complicated as a knowledge middle? However as AI demand skyrockets together with the necessity for extra compute energy, hypothesis round knowledge middle improvement is creating chaos, notably in areas like Northern Virginia, the data center capital of the world. On this evolving panorama, utilities are being bombarded with energy requests from actual property builders who could or could not really construct the infrastructure they declare.

Pretend knowledge facilities symbolize an pressing bottleneck in scaling knowledge infrastructure to maintain up with compute demand. This rising phenomenon is stopping capital from flowing the place it really must. Any enterprise that may assist clear up this downside — maybe leveraging AI to resolve an issue created by AI — could have a major edge.

The mirage of gigawatt calls for

Dominion Power, Northern Virginia’s largest utility, has acquired combination requests for 50 gigawatts of power from data center projects. That’s extra energy than Iceland consumes in a 12 months. 

However many of those requests are both speculative or outright false. Builders are eyeing potential websites and staking their claims to energy capability lengthy earlier than they’ve the capital or any technique round the way to break floor. Actually, estimates counsel that as a lot as 90% of those requests are completely bogus.

Within the early days of the information middle growth, utilities by no means needed to fear about faux demand. Firms like Amazon, Google and Microsoft — dubbed “hyperscalers” as a result of they function knowledge facilities with tons of of 1000’s of servers — submitted easy energy requests, and utilities merely delivered. However now, the frenzy to safe energy capability has led to an inflow of requests from lesser-known builders or speculators with doubtful monitor information. Utilities, which historically cope with solely a handful of power-hungry prospects, are all of a sudden swamped with orders for energy capability that might dwarf their complete grid.

Utilities wrestle to kind truth from fiction

The problem for utilities isn’t simply technical — it’s existential. They’re tasked with figuring out what’s actual and what’s not. They usually’re not well-equipped to deal with this. Traditionally, utilities have been slow-moving, risk-averse establishments. Now they’re being requested to vet speculators, lots of whom are merely taking part in the true property recreation, hoping to flip their energy allotments as soon as the market heats up.

Utilities have teams tasked with financial improvement, however these groups should not used to coping with dozens of speculative requests directly. It’s akin to a land rush, the place solely a fraction of these claiming stakes really plan to construct one thing tangible. The outcome? Paralysis. Utilities hesitate to allocate energy after they don’t know which initiatives will materialize, slowing down the complete improvement cycle.

A wall of capital

There’s no scarcity of capital flowing into the information middle area, however that abundance is a part of the issue. When capital is straightforward to entry, it results in hypothesis. In a method, that is just like the higher mousetrap downside: Too many gamers chasing an oversupplied market. This inflow of speculators creates indecision not simply inside utilities but in addition in native communities, which should resolve whether or not to grant permits for land use and infrastructure improvement.

Including to the complexity is that knowledge facilities aren’t only for AI. Certain, AI is driving a surge in demand, however there’s additionally a persistent want for cloud computing. Builders are constructing knowledge facilities to accommodate each, however differentiating between the 2 is more and more troublesome, particularly when initiatives mix AI hype with conventional cloud infrastructure.

What’s actual?

The respectable gamers — the aforementioned Apples, Googles and Microsofts — are constructing real knowledge facilities, and plenty of are adopting methods like “behind-the-meter” offers with renewable vitality suppliers or developing microgrids to keep away from the bottlenecks of grid interconnection. However as actual initiatives proliferate, so too do the faux ones. Builders with little expertise within the area try to money in, resulting in an more and more chaotic setting for utilities.

The issue isn’t simply monetary danger — though the capital required to construct a single gigawatt-scale campus can simply exceed a number of billion {dollars} — it’s the sheer complexity of creating infrastructure at this scale. A 6-gigawatt campus sounds spectacular, however the monetary and engineering realities make it nearly inconceivable to construct in an affordable timeframe. But, speculators throw these large numbers round, hoping to safe energy capability within the hopes of flipping the venture later.

Why the grid can’t sustain with knowledge middle calls for

As utilities wrestle to kind truth from fiction, the grid itself turns into a bottleneck. McKinsey not too long ago estimated that world knowledge middle demand may attain as much as 152 gigawatts by 2030, including 250 terawatt-hours of recent electrical energy demand. Within the U.S., knowledge facilities alone may account for 8% of total power demand by 2030, a staggering determine contemplating how little demand has grown within the final twenty years.

But, the grid just isn’t prepared for this inflow. Interconnection and transmission points are rampant, with estimates suggesting the U.S. may run out of energy capability by 2027 to 2029 if different options aren’t discovered. Builders are more and more turning to on-site technology like gasoline generators or microgrids to keep away from the interconnection bottleneck, however these stopgaps solely serve to focus on the grid’s limitations.

Conclusion: Utilities as gatekeepers

The true bottleneck isn’t a scarcity of capital (belief me, there’s loads of capital right here) and even know-how — it’s the flexibility of utilities to behave as gatekeepers, figuring out who’s actual and who’s simply taking part in the hypothesis recreation. With no sturdy course of to vet builders, the grid dangers being overwhelmed by initiatives that can by no means materialize. The age of faux knowledge facilities is right here, and till utilities adapt, the complete {industry} could wrestle to maintain tempo with the true demand.

On this chaotic setting, it’s not nearly energy allocation; it’s about utilities studying to navigate a brand new, speculative frontier in order that enterprises (and AI) can thrive.

Sophie Bakalar is a companion at Collaborative Fund.