In our latest episode of Lexicon, we sat down with Diane Watkins, Vice President of the IEEE Power and Energy Society (PES), to explore a side of artificial intelligence that often gets overlooked. 

Throughout our talk, Watkins makes clear that AI is only as powerful as the infrastructure that supports it. She explains how this surge in demand is pushing energy systems to their limits, and why the next phase of AI innovation may depend as much on power infrastructure as it does on code.

Also, subscribe to IE+ for premium insights and exclusive content!

AI’s growth is colliding with grid reality

For decades, electricity demand in many developed countries remained relatively flat. Utilities gradually expanded generation and transmission, responding to steady population growth and incremental industrial development. AI has shattered that pattern.

“It may have taken a utility 100 years to reach a peak load of 1,000 megawatts,” Watkins explained. “And one data center can increase that by 50 percent — or even double it — almost overnight,” she added.

Utilities aren’t seeing one request at a time, she added, “they’re seeing many.”

“We’re seeing utilities receive requests from six to ten data centers at once. That’s a monumental increase in a very short period,” she noted. But the problem is not simply scale; it is also a matter of timing. Infrastructure takes years to plan, permit, and build.

“On average, it takes just over four years to build a thermal generation plant,” Watkins said. “And it can take ten years or more to build a major high-voltage transmission line.”

By contrast, AI’s rise from niche research tool to global economic force has happened in less than a decade. “If you think about where AI was ten years ago versus where it is today, it’s clear those timelines don’t work anymore,” she told IE. 

“We can’t afford to wait ten years to build the infrastructure needed to support AI,” she added.

Why utilities can’t just “move faster”

From the outside, it can look as though utilities are slow to respond. Watkins pushes back on that perception.

“Utilities aren’t dragging their feet because they want to be slow,” she said. “They operate in a highly regulated environment. They can’t just build expensive facilities without demonstrating need and getting approval to pass those costs on to customers,” she added.

That process involves regulatory filings, environmental reviews, interconnection studies, supply-chain constraints, and public scrutiny, all of which add time.

“Every major expansion has to be justified, approved, permitted, and then constructed,” Watkins told us. “That takes years, even when everyone agrees the project is necessary,” she added.

Data centers are uniquely challenging loads

What makes AI data centers different from traditional industrial customers is not just their size, but their concentration. “Nearly every data center that gets built requires new electric infrastructure to support it,” Watkins said.

Historically, growth in electricity demand was dispersed with new homes, new businesses, and gradual increases spread across regions. Data centers flip that model.

“A single data center concentrates enormous demand in one location,” she added. “And even areas of the grid with some excess capacity don’t have nearly enough to support even a modest-sized data center,” she told us.

“A 350-megawatt data center going offline is the equivalent of about 50,000 homes shutting down at once,” Watkins told us. “That kind of sudden change can trigger grid failures and wider outages,” she added.

Reliability isn’t optional

The consequences of grid stress extend far beyond inconvenience. “Electric power isn’t a convenience — it’s required for modern life,” Watkins explained. “Blackouts aren’t just inconvenient. They affect human lives,” she added.

She points to recent large-scale outages as a reminder of what’s at stake.

“When the power goes out, trains don’t run, people can’t access money, emergency services are delayed,” she said. “It has a wide-ranging impact on society.”

Data centers themselves demand extremely high reliability, often far more than homes or small businesses. “They usually meet that requirement by building redundant electric facilities,” Watkins noted. “But that further increases the amount of investment needed from the grid.”

The $700 billion question

IEEE PES estimates that $500–700 billion in grid investment will be needed through 2030. Watkins is clear that this money isn’t just about new power plants.

“It’s all of the above — generation, transmission, distribution, storage, and smarter software,” she said. “We need an unprecedented expansion of our current capabilities.”

Climate pressures compound the challenge.

“Extreme weather is driving massive spending just to repair and harden existing infrastructure,” Watkins explained. “That means fewer dollars available for expansion unless overall investment increases.”

“We need every tool in the toolbox — including tools that haven’t been developed yet,” she added.

Perhaps the most overlooked constraint on the energy transition is talent. “The world may need between 450,000 and 1.5 million more power engineers by 2030,” Watkins said. “That’s nearly double today’s workforce.”

“There are about four and a half times as many computer programmers as power engineers,” she noted.

“The power sector has an image problem,” Watkins says. “It’s often viewed as old or stodgy, while tech is seen as glamorous — even though power engineering is critical, innovative, and impactful,” Watkins told us.

She also argued that reframing the field around climate impact, AI integration, and grid modernization is essential to attracting the next generation.

Misaligned incentives slow everything down

At the heart of the issue is a mismatch between the incentives facing utilities and tech companies. “If a utility moves too fast and causes an outage, they face penalties, hearings, and reputational damage,” Watkins explained. 

“Tech companies don’t face the same downside risk,” she added. “For them, not moving fast enough means falling behind.. That imbalance makes coordination difficult,” she said.

What failure looks like

If action stalls, Watkins sees three outcomes. “Higher prices, more blackouts, and stalled AI growth — potentially all three,” she warned.

Grid investments must be paid for, and consumers often bear the cost. At the same time, many large tech firms can insulate themselves with backup generation.

“Individual homeowners and renters don’t have that option,” Watkins notes. “They end up bearing the risk.”

She likens the moment to a historical inflection point. “AI development today is a bit like the space race of the last century,” she said. “Those who succeed in powering it will gain a massive advantage over those who fall behind.”

Power and AI must rise together

Watkins ends with a reminder of what’s at stake, and what’s possible.“The power grid is one of the greatest human achievements of the last century,” she told us. 

“AI could be one of the greatest achievements of this century — if power and tech work together,” she added.