With the midterm elections weeks away, big data centers are doing something unexpected: uniting Democrats and Republicans. From California to Texas, politicians from both parties are pressing pause on data center development. With demand for compute still red hot, developers are looking for new places to put that load and bypass the political stalemate.

An obvious place to look

Enter the distributed data center coming to the underutilized commercial real estate near you.

It turns out a lot of downtown office space is still underutilized relative to where it was before the COVID pandemic. Many buildings were designed and interconnected for much larger electrical loads than they use today, leaving potentially meaningful capacity in place.

That is why a crop of companies, including new entrants like Perimeter and Gravity, established tech companies like Arcadia and Span, and seasoned developers like Generate Capital, are pursuing versions of the same basic idea: rather than wait years for a new purpose-built data center and all of the power infrastructure that comes with it, find buildings where some of that infrastructure already exists and put compute there instead. There is already a lot of activity and deal-making around the concept, even if it’s still too early to know exactly how large the market will become.

On paper, there’s a lot to like. AI companies get power faster, property owners get a new use for struggling buildings, and developers may avoid some of the political fights around giant data centers. It starts to look like a two-for-one solution to two very real problems.

So, naturally, we’re wondering: What could go wrong?

The part that should make utilities nervous

The complication is that commercial buildings don’t use electricity the way data centers do.

Utilities have decades of experience understanding what a commercial load curve looks like. Demand rises and falls throughout the day, then generally drops overnight and on weekends. Utilities plan their grids around these patterns and count on them to keep the lights on.

Compute loads can look very different. If a building starts filling some of its unused electrical capacity with servers, it could begin using much more power, much more consistently — potentially 24/7.

That matters because a building having enough capacity on paper doesn’t necessarily mean the surrounding distribution system — the transformers, feeders, and substations that manage the flow of electricity across blocks and neighborhoods — was designed for that building to use close to its maximum capacity all the time.

One building may be manageable. But what happens when there are 10 buildings on the same feeder, or dozens across a downtown? At some point, the stress moves beyond the building and starts showing up on the distribution system. That could mean brownouts or other reliability problems.

The blind spot

This is where distributed compute starts to look very different from a traditional data center.

A giant new data center has to announce itself. It needs a major interconnection, there’s a utility process, there are studies, and everyone knows a very large load is coming. An existing commercial building is already interconnected.

If the owner or tenant stays within the building’s existing electrical envelope, there may not be the same obvious trigger that says, “This building is about to start behaving like a data center.” The utility may eventually see the change in meter data, but by then the compute is already there.

If this market grows quickly, the risk is that utilities don’t recognize the scale of the shift until they are already dealing with congestion, accelerated distribution upgrades, higher costs, or, in the worst case, reliability problems.

There’s an interesting analogy here to cannabis grow houses. When grow houses started popping up, utilities suddenly found very ordinary buildings using a very unordinary amount of power. Distributed compute could create a similar dynamic, except the potential loads are much larger.

It’s a bit like whack-a-mole. We may solve some of the political problems associated with giant, highly visible data centers by creating a much more distributed class of load that is harder for utilities and regulators to see coming.

Let’s not repeat the same cycle

Distributed compute could be a useful tool for scaling compute demand. But we should learn from what’s happening with large data centers today, where policy is scrambling to catch up with a market that moved first.

Utilities should be thinking now about how to identify sustained changes to the load shape of large commercial customers.

For example:

• Develop early warning systems: Utilities could monitor large commercial buildings for major changes in load and report results to their regulator. Regulators and legislators could consider creating reporting requirements when a building is being converted for compute, even if the project remains within the building’s existing service capacity.
• Create flex tariffs: Utilities, with regulatory approval, could create flexibility tariffs that reward distributed data centers for switching to batteries or otherwise reducing demand when the grid is constrained.
• Make upgrades conditional: Regulators could make building upgrades for distributed data centers conditional on installing battery storage and enrolling in a flexibility tariff, or require developers to pay for the upgrade.
• Invest in batteries up front: Developers and building owners should also invest in batteries that allow them to temporarily go off-grid or throttle back demand if/when they start creating distribution system problems.

The opportunity is real. The question is whether utilities and regulators can put basic guardrails and an early-warning system in place — and whether developers and building owners will take a longer view — before the market reaches another boiling point and everyone reaches for the pause button yet again.