Applied Digital finished Phase II of its first 100 MW building at the Polaris Forge 1 AI campus in Ellendale, North Dakota, bringing the building to full critical IT load on schedule. This is the first of three fully contracted buildings for CoreWeave at the 400 MW Polaris Forge 1 site, and sits alongside a separate multibillion‑dollar lease with a US hyperscaler at the nearby Polaris Forge 2 campus.
My Analysis:
This is another clear data point that the AI buildout is shifting into “industrial” mode. A single 100 MW AI building, fully energized and fully leased, is not a pilot. It is a factory line for GPUs.
CoreWeave gets what it needs most: fast, predictable delivery of high‑density megawatts so it can land and expand GPU clusters for customers that cannot wait for the big three clouds. Applied Digital is positioning itself as a neocloud enabler rather than a retail cloud. Long‑term leases and fully contracted 400 MW at Polaris Forge 1 show that GPU aggregators and hyperscalers are happy to outsource the dirt, steel, and power side of the stack.
North Dakota is the real story here. You do not put 400+ MW of AI load in Ellendale for latency. You do it for power availability, cost, regulatory predictability, and local YIMBY economics. That is classic AI factory pattern: go where land is cheap, interconnect is viable, and the community wants the jobs more than it fears the substations.
Their “waterless cooling” positioning is not marketing fluff in this context. AI‑dense sites in water‑stressed regions are already running into resistance. A high‑density campus that advertises no evaporative cooling and “sustainable” design is a deliberate play to get ahead of NIMBY, ESG constraints, and future permitting battles. That makes this campus a template for the next wave of non‑coastal AI builds.
The ~$16 billion contracted revenue across Polaris Forge 1 and 2 tells you how the economic model is evolving. Hyperscalers and neoclouds want dedicated, long‑term, power‑dense shells to drop their own GPUs and networking into. Applied Digital’s job is to derisk schedule, power, and cooling. That matters for enterprises because a lot of their AI capacity will be “indirect” through players like CoreWeave and hyperscalers using these wholesale sites. When you buy GPUs as a service, you are implicitly betting on firms like APLD hitting these RFS dates.
From a supply chain angle, 400 MW+ of fully contracted capacity in one geography sets up a concentration risk pattern we are seeing globally. Power or transmission issues in that region will ripple into GPU availability for any tenant that over‑indexes there. At the same time, campuses like this relieve pressure in constrained metros where enterprises cannot get 10–20 MW, let alone 100.
For CIOs and infra teams, the signal is clear. AI capacity is not just in the big three cloud REITs anymore. It is in specialized AI factories built for tenants like CoreWeave and a “US investment grade hyperscaler” under multiyear leases. When you evaluate “cloud vs on‑prem vs neocloud,” you are also choosing between traditional metro data centers and remote AI factories like Polaris Forge.
The Big Picture:
This ties directly into several macro trends:
- AI data center construction surge: A fully contracted 400 MW AI campus with one 100 MW building already at full critical load is textbook evidence that AI data center growth is now planned and financed at power‑plant scale, not rack scale.
- Neocloud vs public cloud: CoreWeave is a neocloud archetype. It competes on GPU availability and flexibility, not generic cloud services. By leaning on Applied Digital for physical capacity, it can move faster than hyperscalers that must juggle multi‑tenant, multi‑region priorities.
- GPU availability and supply chain: You cannot land tens of thousands of GPUs without somewhere to plug them in. Sites like Polaris Forge are the upstream constraint that sits next to chip supply. Even if Nvidia can ship, operators without power and cooling lose. This campus is essentially reserved GPU runway for CoreWeave and a hyperscaler.
- Energy and water constraints: Building in North Dakota with waterless cooling signals a pivot away from urban, water‑cooled facilities toward inland, energy‑rich, water‑constrained but politically favorable regions. This is how operators will keep scaling without hitting local resistance on water rights and grid constraints.
- NIMBY vs YIMBY: The “economic opportunities in underserved communities” language is important. These AI factories are being sold to local stakeholders as jobs and tax base, not as faceless server farms. That is the playbook for YIMBY AI regions: show up with long‑term leases, strong tenants, and low water use.
- Enterprise AI adoption: Most enterprises will never see Ellendale, but they will feel it in their SLAs. Capacity from campuses like Polaris Forge will underpin the GPU instances they rent from CoreWeave or hyperscalers. As more workloads move to fine‑tuning and inference at scale, they will increasingly land on these dedicated AI factories, not classic multi‑purpose DCs.
Signal Strength: High
