Insight · AI Infrastructure
Insight No. 01 · AI Infrastructure

Where AI compute is going — and how it gets powered.

AI compute is splitting two ways: giant centralized campuses chasing the lowest cost per token, and distributed sites placed closer to load, users and available power. Different shapes — one shared constraint: firm power, and the years-long wait to get it from the grid.

The one-line answer

Centralized = a few enormous gigawatt campuses optimized for training at scale. Distributed = many smaller sites near metros, load and users, increasingly for inference. Both are growing fast — and both hit the same wall: the grid can’t energize them in time. The fix is the same too — firm power, built on site.

Closer to load

Distributed AI

Compute placed near demand.
  • Scale: tens to ~100+ MW per metro or regional campus — many sites, not one
  • Where: near metros, users, fiber and available power — spread across markets
  • Mission: low-latency inference, regional capacity, resilience, proximity to data
  • Why it’s growing: inference is exploding, latency matters, and giant single sites keep getting blocked
  • The constraint: firm power per site — metro grid queues stall these builds for years
Scale & cost

Centralized AI

One enormous campus.
  • Scale: 500 MW to multi-gigawatt single mega-campuses
  • Where: wherever land, water and power can be assembled cheaply — often remote
  • Mission: large-scale model training — the lowest possible cost per token
  • Why it’s growing: training at frontier scale rewards concentrating compute and power
  • The constraint: gigawatts in one place — transmission and generation take 5–7+ years
DimensionDistributed AICentralized AI
What it isMany smaller sites placed near load and usersA few enormous single campuses
Typical scaleTens to ~100+ MW per site, across many markets500 MW to multi-gigawatt in one location
Where it sitsNear metros, fiber and available powerWhere cheap land, water and power can be assembled
Primary workloadsInference, regional capacity, low-latency servicesFrontier-scale model training
Why it’s growingInference demand, latency, and blocked mega-sitesTraining scale rewards concentration
The power problemFirm power per site — metro queues stall buildsGigawatts in one place — 5–7+ yr transmission
RelationshipBoth, not either/or — the same operators build both, and both run into the same wall: the grid can’t deliver firm power fast enough. On-site generation is the unlock for either shape.
Where AnchorPower Fits

Distributed or centralized, the missing layer is the same: firm power, on site.

Whichever shape AI compute takes, the bottleneck is power — and metro grid connections take 5–7+ years. AnchorPower closes that gap with one modular architecture that scales from a single distributed campus to a centralized gigawatt factory.

AnchorFirm → any scale

Fully off-grid, behind-the-meter firm power under a 15-year fixed-price PPA. One modular platform scales from a 50–100 MW campus to 500 MW+ and gigawatt-scale — no interconnection queue, first power in 18–20 months.

Built like the compute → modular

AnchorPower systems are modular, containerized and mostly pre-fabricated — the same factory-built approach that makes modern compute fast to deploy. Power arrives the way the racks do: in modules, ready to energize.

Both shapes are also converging on native-DC distribution at 800V DC — the architecture NVIDIA, Google and the OCP ecosystem are standardizing on. AnchorFirm is DC-coupled at the source, with solar, BESS and firming on a common 800V DC bus, recovering roughly 6–7 points of conversion loss versus an AC-coupled on-site plant (modeled per stage).
Pragmatic · Firm on site · Built on an 800V DC backbone
© 2026 AnchorPower Energy. All rights reserved.