Can the grid handle the AI Boom?

Written By: Nate Ford 

I'm not sure you have heard, but...AI is having a moment. It’s in your phone, it's at the top of your search results, it's in your thermostat, and it may even be helping write my, er, I mean your blog posts. But behind the curtain of all this innovation is a simple yet perhaps overlooked truth: AI needs power. A lot of it. 

And that is not likely to slow down any time soon. 

Why exactly does AI need so much power? 

Imagine you’re in charge of training a language model. It’s not just typing on a laptop; it’s like trying to teach a hyperactive, genius toddler every language, scientific paper, meme, and recipe on the internet simultaneously. The hardware running that show? Rows upon rows of GPU-packed servers humming in massive data centers, with cooling systems, backup generators, and nonstop processing. It all stacks up. 

Experts estimate that AI-related energy consumption could rise tenfold in the next few years. U.S. data centers consumed 176 TWh in 2023 (≈4.4% of U.S. electricity) with projections reaching 12% by 2030.  

As AI becomes embedded in everything from customer service chatbots to high-stakes medical diagnostics, that upward curve could look less like a slope and more like a cliff face. 

The electrical grid isn’t just a background character 

If you’ve never thought a lot about how power gets from dams or solar panels to your laptop, don't worry, you’re not alone. But the grid is suddenly a big part of the story. 

The problem isn’t just the raw amount of power AI needs, it’s the where and the when. Data centers aren’t always near power plants, and a single center might draw as much electricity as a town. Add in demand surges (like when a company trains a new AI model), and you start to see the cracks. 

Our current system was built for predictable use: homes, streetlights, offices turning on and off like clockwork. But the world is no longer shaped like that. AI isn’t like that. And AI isn't the only demanding actor on set. 

Crypto’s still a heavyweight in the ring 

Now, if you’re wondering whether AI is the only digital giant guzzling electricity—spoiler alert—it’s not. Cryptocurrency mining, especially for proof-of-work blockchains like Bitcoin, is still an enormous strain on the grid. 

Mining doesn’t take breaks. It runs 24/7, churning through electricity at a rate that rivals entire countries. In fact, Bitcoin mining alone consumes over 175 terawatt-hours annually. That is more than the usage of the entire country of Poland. Unlike AI, which has peaks and lulls, crypto mining is a relentless, power-hungry machine. 

What’s more? It often sets up shop where electricity is cheapest, meaning it isn’t always where it's greenest or where the grid can best handle it. That can lead to problems. 

So while AI might be the newer, flashier disruptor, crypto's old-school energy draw is still a major player. 

Can the grid catch up? 

There’s no easy fix, but there are some promising moves: 

• Demand flexibility – Getting smarter about when and where we use power. 

• Grid modernization – Investing in tech that makes energy delivery faster and more adaptive. 

• On-site generation – Data centers using solar, batteries, or even microgrids to take the pressure off. 

And here's where energy efficiency folks come in. Because if AI and crypto are going to keep ballooning, we have to make the rest of the system smarter, cleaner, and, well, more efficient. 

So what now? 

You don’t need to shut down your smart assistant or swear off blockchain. But we do need to think bigger. How do we keep pace with innovation without frying the grid, let alone the planet? 

It's been a long time since energy efficiency was all about installing LEDs. It’s about building a future where our digital progress doesn’t come at the cost of environmental regression. 

It’s a tall order, but hey, this is what we do!  To find out more about energy efficiency in your commercial property or our outreach and program management with utility companies, reach out at info@waypoint-energy.com.