As large data centers scramble for power and face years-long delays in grid queues and local resistance, a wave of companies is testing out a different strategy: bringing computing to locations where extra capacity already exists.
Katie Clasen, a partner at Prelude Ventures, told Latitude Media in a recent Latitude Dispatch that she’s tracking 35 different players in the burgeoning edge computing space — from early-stage startups to publicly traded companies looking to tap into a new line of revenue. These include Rune, which is placing small servers at utility-scale renewable energy projects that often curtail electricity during periods of low demand, as well as Tesla. In June, Tesla filed a patent for modular AI hardware called “Megapod,” which Clasen said is a signal that the company’s electric vehicle charging stations might host small data centers someday, too.
Edge computing is best suited for AI inference, or the day-to-day use of chatbots or self-driving cars that require less electricity than training large language models. That’s where the market is already shifting; analysts predict that inference will account for 60% of AI demand by 2030. Meanwhile, load growth forecasts from data centers continue to rise.
This is partly why Clasen is critical of developers building multi-gigawatt-scale data centers powered by gas. Some of that computing power could potentially interconnect faster in locations that already have extra headroom on the grid, avoiding the need for costly new fossil fuel infrastructure that worsens both climate change and local air pollution.
Clasen noted that she’s talked to some neocloud developers, which lease space for computing hubs, and found that some of their enterprise customers are already requesting that large projects be broken up into smaller sites for AI inference. The developer Crusoe, for one, last month told the Wall Street Journal it’s pursuing that path, after making a name for itself by building the massive Stargate campus in Abilene, Texas.
Clasen specifically raised concerns about Nscale’s Monarch Compute Campus in West Virginia, an off-grid project with two gigawatts of gas generators lined up from Caterpillar. Microsoft intends to buy 1.4 GW of computing power from the campus, which aims to be “a cornerstone facility for next-generation AI training and inference capacity,” according to a May press release.
“That really freaks me out,” Clasen said. “There’s no reason why we should be servicing inference demand via two GWs of Caterpillar natural gas gen sets. That just doesn’t make sense to me.”
That said, the edge computing market is still in its early days, so it’s challenging to predict how much it can scale. Clasen said some of the more promising business models plan to aggregate sites with between one and five megawatts of computing power — an amount that office buildings or other commercial properties, industrial sites, and EV charging hubs can probably handle, but residential sites likely cannot.
“Anything under one MW per site, and I start to get a bit worried,” Clasen said. Getting as many MWs per location is key, but at this stage each and every edge computing site is “hard-won,” she added.
She’s also concerned with theft, security, and insurance. GPU chips are really expensive, so companies need to consider how secure their chosen site is. An office building may be a better choice than someone’s garage, for example.
Another challenge facing the broader industry: how to structure contracts between all the parties involved. They might include property owners, the companies that own the computing hardware, third-party customers like hyperscalers or AI labs, and utilities that need to make sure there really is extra headroom on their grids. Meanwhile, startups are raising debt and equity to finance the purchase of expensive GPUs and build out their software platforms, which brings new financiers into the mix.
“There’s not much consistency in the market on how these deals are getting done. It’s playing out in real time. Everyone’s in the pilot phase,” Clasen said, adding that she thinks the teams who’ll be successful will be the ones with the expertise to deliver all those contracts.
Editor’s note: Prelude Ventures is an investor in Latitude Media. This Dispatch and resulting article were produced using the editorial team’s standard process; neither were subject to review by Prelude.
Watch the whole Latitude Dispatch featuring Katie Clasen here:
