The solution to the AI power boom is a better grid, not a bigger one

The looming surge in power demand for EVs and data centers makes utilities nervous not just because of the cost, copper, and labor required to supply it, but because meeting the challenge will require a deep cultural shift in the stodgy, lumbering power sector toward Silicon Valley-style speed and innovation.

For a century, the basic rule for utilities has been when you need more power, you build more generators and wires. In the energy transition era, when the grid needs to be both much cleaner and much greater in capacity, that approach can’t work fast enough or at the necessary scale. So while utilities race to build substations and transmission lines, many are also increasing their investments in tech-forward solutions, often utilizing AI, that don’t require new large-scale infrastructure to expand the grid’s efficiency and carrying capacity.

Although most people are fixated on the problem of an under-built grid, an even more serious risk, some in the power business say, is the over-built grid, one where financial, natural, and bureaucratic resources are wasted, and backyards and landscapes are needlessly encroached on, to achieve the same carbon footprint outcome that could have been achieved faster through less expensive and invasive means.

“There’s a lot of additional headroom sitting around most of the time” on the grid, Alexina Jackson, vice president of strategic development at the Virginia-based multinational power company AES, told Semafor. “We need to move from the mindset of ‘the world will always be this way and never change’ to realizing that we need to be increasingly dynamic in the way we consume and transmit energy.”

Dynamic line rating systems like LineVision’s are one approach. Startups like AutoGrid, which took early investment from NGP and SE Ventures before it was acquired by the grid software company Uplight in December, use machine learning to predict power production and demand curves in a local area. That information can be used to smooth down the peaks — perhaps incentivize a power-hungry factory to ramp down for a few hours on a hot afternoon with a lot of air conditioners running — and lower the utilities’ peak capacity requirements. Another NGP-backed startup, AiDash, uses satellite imagery and AI to “apply a risk-based approach to utility vegetation management,” such as informing decisions about where and when to trim trees, which are the top cause of damage to power lines.

“Most utilities really struggle to answer the question of what’s happening on their grid right now,” said Astrid Atkinson, co-founder of the startup Camus Energy, which uses AI to analyze real-time demand and supply data from smart meters and other sources. “That makes it really difficult for utilities to make changes. What exactly would happen if they connect a solar farm? They don’t know. That data deficit is a really big reason for the interconnection backlogs.”

There are also some relatively low-tech and low-cost upgrades to existing grid hardware that can make a big difference. Reconductoring, or replacing old wires in transmission lines with more advanced ones, can unlock huge capacity additions at a far lower cost than building new lines, and shave years off the process by avoiding the minefield of new-build permitting bureaucracy. Siemens Energy is removing greenhouse gasses directly from grid infrastructure itself, by upgrading substation equipment to remove sulfur hexafluoride, which is commonly used as an insulator but leaks at high rates and is extremely potent at trapping heat in the atmosphere.

There’s more work to do, Atkinson said, to get utilities comfortable with the idea that they can proactively manage the grid in a way that has been impossible before, rather than just overbuild for the worst-case scenario and hope for the best.

“There’s a process of trust-building,” she said. “But utilities are really starting to understand where they have gaps.”

The reason for the relatively slow adoption of cutting-edge grid tech isn’t because utilities are averse to well-reasoned risk, said Lucy Yu, CEO of the Centre for Net Zero, a research group on AI-driven grid strategies, founded by London-based Octopus Energy. It’s because government targets are misaligned, with too much focus on — and therefore policy and financial support for — renewable energy adoption (hitting a certain share by a certain date), and no clear mandate for grid flexibility and performance: “That’s something we’re not recognizing the full potential of at the moment.”