The hyperscale data center buildout is falling behind the timelines embedded in nearly every AI forecast. Compute demand keeps accelerating but the grid constraints and politics are making massive new data centers harder to build. Meanwhile, an enormously valuable asset is hiding in plain sight: the existing meter. Arcadia supports nearly 1 in 5 commercial and industrial meters in the U.S. with energy management solutions that save customers time, money, and carbon. And nearly every commercial building has a utility service sized higher than its needs, when the majority of the time its operating far below that capacity. The difference is called 'headroom', and its power capacity that is already permitted, connected, and energized. In a market where speed to power is everything, the fastest path to new compute isn’t always another hyperscale campus. It may be deploying distributed compute behind millions of existing meters without waiting years for a new grid connection. Arcadia has spent nearly a decade aggregating meter, tariff, and interval data across thousands of utilities worldwide and now we’re building tools to identify and monetize available headroom across millions of our existing customer's buildings. The opportunity is significant, not only is it faster compute capacity for the AI economy, but its new NOI and significant bill savings for building owners. Just as Arcadia provides transparency and trust brokering wholesale power contracts, we'll be able to drive the same transparency for our customers into compute costs, lease payments, and token economics. The grid connection is already there, now it’s time to put it to work. Learn more about Arcadia’s distributed AI compute solutions: arcadia.com/distributed-ai
Shan Huang
1w
The harder business is making thousands of buildings with different peak loads and curtailment rules look like one reliable compute pool, which pulls Arcadia much closer to a cloud control plane than an energy marketplace.
It's funny because we were talking about distributed compute being one of the solutions that needs to happen in parallel with the hyperscale AI data center build out to reduce pressure on the grid, as well as potentially maybe reduce community opposition if you have the public buy-in to host, let's say, modular data centers in buildings and homes. It's an interesting build-out strategy that I hope to see scale with also a key focus on optimizing buildings and homes with battery storage systems, window solar panels, etc.
Dlzar Al Kez
1w
Interesting idea. I think it becomes practical when the sites are managed as one aggregated load. Building-meter data can show local headroom, but the utility still needs to set limits at feeder or substation level, and the compute loads may need to ramp or curtail to stay within them. Otherwise, spare capacity at each building is not necessarily spare network capacity.
Uroš Simović
1w
Great point. The next frontier is managing both flexible loads and DERs by orchestrating both demand and generation together. This will unlock existing grid capacity and accelerate AI infrastructure without waiting for new interconnections.
I see this offering having a place for edge workloads, predictable or batch workloads which can be scheduled or rerun and are not latency sensitive.
One of my favorite projects that I've seen regarding AI was a partnership with span to basically install what looked like an air conditioning unit at people's houses to put compute clusters in a distributed fashion. I would love to be the first Arcadia employee to have something like that on my house!
Kiran its a very interesting point you raise. Distributed compute utilising already existing headroom in C&I connections is a great solution to the grid constraints we all know so well. It will be complex to deliver, but there is value in solving hard problems.
Kyle Illgen
1w
Kiran Bhatraju our cooling technology allows deployments of dense compute into areas with existing power but not much else. Facility water is optional, service and support is familiar, i.e. no tanks, just racks. We're aiming for the exact sub-1MW space you're discussing here with some ongoing work with partners on a modular weatherized containment that can land on any property that has the power to run the systems (about 50kW a rack.) Let me know if you have time to discuss more, I think we could be valuable partners in this race to distributed AI!
Read more about how we can help businesses earn new revenue by hosting compute here---> https://www.arcadia.com/blog/grid-connection-capacity-distributed-ai