According to Daniel Gregory, quantum computing is coming — probably faster and sooner than anyone expects — and when it arrives, hyperscale data centers could be toast.
“When quantum hits … you’re going to see tremendous shifts away from [artificial intelligence] training at massive data centers into training in a hybrid system, including quantum,” he told attendees at the Earthshot Foundation’s Future of the Grid conference, in Washington, D.C. on June 8. “When that comes in, why are we building classical computer systems for $5 trillion that are fundamentally already leapfrogged?”
Gregory is CEO of Available Infrastructure, a startup that is preparing for that fast-approaching shift with a network of “micro-edge” data centers that will deliver artificial intelligence services close to where customers need them.
“We can distribute a million GPUs across 100 cities and have a fully coherent, less than 5-millisecond data center,” Gregory said. “We can do it without grid upgrades, without using water, without special permits. Probably we won’t have to change transformers, at least not initially, because we’re cherry-picking where there’s a weak spot in the system and locating there,” he said.
The Earthshot conference was the culmination of a year of cross-industry meetings and conversations, all focused on what a future, modern grid might look like, and moving data centers and electricity generation to the grid edge is part of that bigger picture.
Available Infrastructure is one of a handful of startups rolling out new and very cool grid-edge technologies that could turn today’s hyperscale data centers into so many white elephants.
Likewise all of the fossil-fueled power plants President Donald Trump and Energy Secretary Chris Wright keep throwing millions in federal funds at, claiming they will provide the electricity needed to win the AI race with China. Their latest boondoggle, announced June 4, will channel $700 million in taxpayer dollars to expand 12 existing coal plants, bring one shuttered plant back online and build two new plants.
Grid-edge data centers, on the other hand, can be powered by grid-edge solar, storage and other clean technologies — distributed energy resources in industry-speak — all of which can be built quickly and cost-effectively, where they are needed most.
Available Infrastructure already has its micro-edge data centers installed in 30 cities and “will be firing up our first sort of super nodes in New York City and Philadelphia by the end of the month,” Gregory announced at the conference.
Behind-the-meter ‘compute nodes’
We need a jargon break here so I can explain all the industry- and wonk-speak I’ve thrown around thus far.
To begin with, I make no claims to completely understanding what quantum computing is. Most online explanations – like this one from IBM – say it is a whole different level of processing based on quantum physics (which I really don’t understand, but here is an explainer from Caltech).
The main things you need to know about quantum computing are that it can solve incredibly complex problems that current AI can’t, and it works at absolutely astonishing speeds – hence Gregory’s caution about how quantum will leapfrog over hyperscale data centers.
These massive, warehouse-like structures where AI processing takes place can require up to a gigawatt of power, which is the equivalent of putting a city the size of San Francisco on the grid in one place, all at once. GPUs are the graphic processing units, or chips, that make AI so smart and powerful.
But what’s starting to happen is a shift in focus from centralized, hyperscale data centers to the “grid edge,” an industry term with different shades of meaning depending on context. Defined literally, the grid edge is where power lines end, where electricity is delivered and used.
But in the current context,, what we’re talking about is the grid’s last-mile delivery system, the poles and wires sending power to homes and businesses where DERs can be aggregated or managed to relieve stress on the grid or power a small “compute node.”
Such behind-the-meter micro data centers are being developed by XFRA, a spinoff of SPAN, a DER developer producing super cool electric panels that monitor and manage home electricity use (which I wrote about here).
(Behind-the-meter means the node — a boxy piece of equipment that looks like a heat pump — is wired into a home’s electric system, rather than the poles and wires the utility manages.)
At the Future of the Grid conference: Richard Caperton of SPAN (left) and Daniel Gregory of Available Infrastructure. Credit: K Kaufmann
Speaking on the same panel as Gregory, Richard Caperton, SPAN’s vice president for public policy, said the SPAN/XFRA system takes advantage of a home’s regular “load curve” – how electricity use changes at different times of the day.
Residential load curves have peaks and valleys, Caperton said. “And so we are managing homes’ load to then fill those valleys with computing power. We’re deploying distributed compute nodes in residential settings, behind a residential meter, behind a residential panel and then operating those so they can take advantage of the existing capacity on the distribution system.” (See image at top of online version of article.)
Paired with a SPAN panel – and where possible, rooftop solar – the node boxes combine 16 advanced AI processors with backup storage and a cooling system, according to the XFRA website.
A company white paper cautions that the compute nodes will not replace the hyperscale data centers needed to train AI models – that is, feed them all the information they need to operate. But they can provide fast, local processing for gaming, content streaming, cloud computing and “other geographically distributed workloads,” the white paper says.
The company now has “revenue-generating test units” in the field and is planning a 100-home demonstration project later this year, the white paper says.
(Thanks to AI, “compute” has done one of those wonky transitions from verb to noun, now defined as both computing power and the associated hardware. As ever, jargon happens; get used to it.)
Taking it to the streets
Another argument for pushing AI to the grid edge is speed, Gregory said. Grid-edge stuff happens in analog, real time versus a slight time lag for digital, which can become significant if a data center and its customers are on opposite sides of the country.
“Most people don’t realize analog is fast; it’s real, it’s instant. Everything else just adds delay,” he said. “The real world operates pretty much in less than 5 milliseconds. You don’t want to add more [delays] than that. If you do, all your automation stuff starts unraveling. …
“If you stack [delays] all the way back to a data center in Wyoming from Boston, you can imagine you’re going to have unpredictable amounts of delay.”
In other words, as with electricity, the closer AI is to the people using it, the faster and more efficient it’s going to be.
Conflow Power Group, a UK startup, is taking that idea and going hyperlocal with streetlamps that combine 600 watts of wrap-around solar panels with energy storage batteries and an AI chip (see below). The goal, CEO Edward Fitzgerald told pv magazine, is to stand up a virtual power plant network, aggregating hundreds or thousands of streetlamps to offset the electricity demand from data centers.
Curbside AI? Conflow’s iLamp street lights combine solar, storage and AI. Credit: Conflow Power Group.
“A lot of people say we can’t compete with a large-scale data center by putting GPUs into streetlights. That’s true,” Fitzgerald said. “But we don’t have to cool ours; so, we’re already more efficient, and the compute is more efficient and it costs us and the environment less.”
Similar to XFRA’s compute nodes, ConFlow’s iLamps bring “the data center closer to your phone for lower-demand tasks like asking ChatGPT a question,” he said.
Another goodie, curbside electric vehicle chargers can be added to the streetlights.
Conflow is deploying the streetlights through regional licenses. For example, in the U.S., licensees in Florida, Texas, and California are planning and rolling out the technology, and more states are in negotiations for their own regional licenses, according to the company website.
Looking for mega shifts
The grid-edge technologies I’ve been talking about are set to disrupt the electric power industry and likely will continue to do so at what I suspect will be increasing speed and scale.
The challenge for the U.S. – and other industrialized nations – is the extent to which developing a cohesive vision or plan for the grid’s future has been politicized and polarized, said Mark Paterson, an industry consultant with Energy Catalyst, an Australian firm.
Speaking at the Earthshot conference, he called for a “grand strategic transformation” of the grid and the electric power industry, which is not likely to come from “traditional entities,” like utilities, regulators and policymakers.
“We’re looking for mega shifts,” Paterson said. “Reverting to a centralized, top-down grid predominantly is about as likely as yellow taxis, traditional hotel chains and record stores regaining their monopoly.”
Ditto, coal plants becoming a major source of clean, affordable electricity in the U.S.
Rather, the question now is “how do we build processes to reimagine and reengineer the grid for its second century of operation in a very different future? That takes very different processes from the incremental, keep the lights on, … Band-aid kind of initiatives,” he said.
The way forward will likely include “two parallel paths of activity” – one traditional and incremental, to keep the lights on and buy time, and the other unconstrained “by all the things you can’t say, all the things you can’t do … the tyranny of the urgent,” Paterson said.
The future grid must be able to coordinate “a power system, moving from hundreds to tens of thousands to hundreds of millions of resources,” he said.
But at what point will we cross over from our incremental legacy electric system to the grid and clean, distributed data centers we are only beginning to imagine?
The Earthshot Foundation’s report on its yearlong Future of the Grid initiative says it this way:
“Our ability to embrace industry disruptors as the biggest innovation opportunity in decades will determine the might of tomorrow’s energy systems.”
I want a pile of toast and a whole herd of white elephants.