America’s Data Centers May Soon Swallow One‑Fifth of the Country’s Power

When you picture the next wave of artificial intelligence, you probably imagine flashy robots or clever chatbots, not a sea of humming servers. Yet a fresh BloombergNEF report makes it clear that the real engine behind the AI boom is a massive, growing fleet of data centres, and they’re about to become huge power‑hungry neighbours.

Today, data centres account for roughly 5.9 % of U.S. electricity consumption. That figure looks modest, but the report projects it will leap to about 20 % by 2035 – roughly one‑fifth of the entire nation’s power needs. In plain terms, every coal plant, every natural‑gas turbine, every solar farm in the United States could see a fifth of its output redirected to keep those racks running.

What’s driving that surge? AI, plain and simple. The analysts estimate that around 200 GW of new data‑center capacity will be built over the next decade, and about half of that will be dedicated to AI training and inference. Think of the massive tensor‑core GPUs chewing through petabytes of data to teach a model how to write prose or diagnose disease – they’re power‑guzzlers.

It isn’t just a national story; the impact is already visible in regional grids. In PJM – the 13‑state interconnection that stretches from Virginia to Illinois – data centres could gobble up roughly 34 % of total electricity by 2035. In Texas’s ERCOT grid, the share could climb to about 22 %. Those are eye‑popping numbers that have utilities on edge.

“The grid is already strained,” warns Lloyd Arnold, the BNEF analyst behind the study. He points out that utilities like American Electric Power have hinted they might pull out of PJM if demand keeps spiralling. The concern isn’t just about higher bills; it’s about whether the existing grid can stay stable when a quarter of its power is being sucked into server farms.

There’s also a geographical twist. Data‑center siting decisions tend to follow cheap, reliable power – often in regions rich in natural‑gas pipelines or renewable farms. That means some areas could see a sudden, dramatic shift in their energy mix, while others remain relatively untouched.

What does this mean for policymakers and the tech industry? For one, it underscores the urgency of making AI workloads more energy‑efficient. Companies are already exploring custom chips that do more work per watt, and the push for greener cooling methods – like using outside air or even “liquid‑on‑silicon” immersion – is gaining steam.

Second, regulators may need to rethink how they allocate capacity on the grid. If an AI‑driven data centre can consume as much power as a mid‑size city, then it makes sense to treat it like any other major load, with grid‑impact studies and, perhaps, incentives for locating in low‑stress zones.

Finally, consumers should be aware that the invisible cost of AI is real. Every time you ask a chatbot a question, a tiny slice of that future power demand is already being booked.

In short, the AI boom is not just a software story; it’s quickly becoming an electricity story. If the projections hold, by 2035 the United States will need to reckon with data centres that together draw the same amount of power as the whole nation’s banking, healthcare, and manufacturing sectors combined. Whether the grid can keep up remains to be seen, but the writing is already on the wall – and it’s glowing in a very bright, very electricity‑hungry way.