Applied Digital adds 75MW of capacity at Ellendale data center campus in North Dakota

Applied Digital just threw another 75 MW of AI‑grade power onto its Polaris Forge 1 campus in Ellendale, North Dakota. That’s the second building they’ve finished, taking the site from a half‑built 150 MW phase to a full‑blown 250 MW campus. The numbers sound impressive until you remember the whole thing is already fully leased to CoreWeave, and the real story is what this means for anyone trying to run a data centre without a billionaire’s war chest.

From crypto‑miner to AI‑factory: a rapid pivot

Founded in 2021 as Applied Blockchain, the firm has been busy re‑branding itself as Applied Digital and shifting its focus from crypto mining to AI and high‑performance computing (HPC). The source material notes that the company’s “crypto and HPC business” kicked off construction on Polaris Forge 1 in September 2022. By the end of 2025 they had the first phase of Building 1 live, and now, in 2026, they’re adding another 75 MW of “critical IT load” in Building 2.

The pivot isn’t just a branding exercise. Applied Digital’s portfolio now spans five AI Factory campuses with a combined 1.4 GW of critical IT load and about 2.15 GW of grid‑connected utility power. That scale‑up is being driven by “large‑scale deals with hyperscale customers,” a phrase that masks the reality that most of the capacity is pre‑leased to the same kind of high‑investment‑grade hyperscalers that dominate the market.

Power is cheap, but the real work is in the stack

Wes Cummins, the CEO, is quick to point out that “securing power is only the starting point.” He’s right. Turning megawatts into usable AI infrastructure requires design, build, commissioning, and ongoing ops – everything that a small‑to‑mid‑size hosting provider has to wrestle with on a daily basis. Applied Digital can afford to lay down three 25 MW data halls and then hand them off to a hyperscaler that brings its own engineering muscle. For an independent operator, the cost of that engineering talent, the risk of under‑utilisation, and the complexity of cooling at that scale are massive hurdles.

In practice, the “real work” includes everything from power distribution units that can handle 25 MW per hall, to custom cooling loops that keep GPUs from throttling under load. Those systems are not off‑the‑shelf. They require a level of engineering depth that most boutique providers simply don’t have, and that’s why many end up either over‑building and sitting on idle capacity, or under‑building and missing out on high‑margin AI workloads.

Leasing to hyperscalers: a double‑edged sword

The source material makes clear that the entire Polaris Forge 1 campus is “fully leased” to CoreWeave, a company that itself is a large AI‑focused cloud provider. That means Applied Digital is collecting rent, but it also means they have handed over control of the infrastructure to a tenant that can dictate the technical requirements. The same pattern repeats at Polaris Forge 2, where an “investment‑grade hyperscaler” will take up most of the site, and at Polaris Forge 3, which is already pre‑leased to an unnamed hyperscaler.

For independent hosting firms, this model is a cautionary tale. Leasing large blocks of capacity to a single hyperscaler can guarantee cash flow, but it also ties you to the whims of a customer that can walk away or renegotiate terms if market conditions shift. The risk is amplified when the hyperscaler’s own pricing model undercuts the smaller player’s margins – a common story when the big players throw down rock‑bottom prices for power and bandwidth.

The hyperscaler pricing trap

When a hyperscaler signs a lease for a 75 MW data hall, they typically negotiate power rates that are far below what an independent provider can secure on the open market. The cost of meeting those guarantees can erode the margin on the lease itself, especially when the provider has to invest in high‑grade UPS, generators, and cooling infrastructure to meet the contract.

Applied Digital’s model sidesteps that by passing the operational burden to CoreWeave. But for anyone else looking to replicate the model, the math doesn’t add up unless you have deep pockets or a guaranteed pipeline of hyperscaler contracts. The reality is that most independent hosting firms will end up subsidising the hyperscaler’s low‑price demand with higher‑margin services elsewhere – a balancing act that can quickly become a financial nightmare.

Risk of over‑building in a volatile market

The announcement that the campus will eventually deliver 400 MW at full build‑out is a classic “build‑it‑and‑they‑will‑come” mantra that many founders have fallen for. The source material tells us that Applied Digital already has a “contracted portfolio” of 1.4 GW across five campuses, but that figure is tied to “large‑scale deals with hyperscale customers.” If those deals evaporate – say, due to a slowdown in AI spending or a shift to edge‑centric workloads – the provider is left with massive, under‑utilised power assets.

From a risk perspective, the prudent move is to align capacity expansion tightly with signed contracts, not speculative demand. The fact that Polaris Forge 1’s first phase only went live in July (year not specified) and the second phase is just now ready for service shows a cautious rollout. Independent operators should emulate that measured approach: build to order, not to hype.

What the spin‑off of ChronoScale tells us

Applied Digital “recently spun off its ChronoScale cloud unit,” according to the source. That move signals a strategic focus on core data‑center assets while shedding the complexities of running a public cloud. For a founder, the lesson is clear: keep your business model lean. Managing a cloud platform adds layers of software, support, and compliance that can distract from the primary revenue driver – the rack space and power you sell.

By off‑loading the cloud layer, Applied Digital can concentrate on the capital‑intensive part of the business: power, cooling, and real‑estate. That’s a playbook that many mid‑size providers can adopt – partner with a cloud or AI workload specialist who brings the software stack, while you provide the bricks‑and‑mortar. The partnership must be structured so that the provider retains control over pricing and capacity, avoiding the trap of becoming a pure‑play landlord with little upside.

Actionable takeaways for independent providers

First, **focus on power contracts that are truly yours**. If you’re leasing to a hyperscaler, negotiate clauses that protect you from sudden price drops or early termination. Second, **stage your capacity build‑out**. Applied Digital’s phased approach – first phase of Building 1 in December 2025, first phase of Building 2 in July (year unspecified), and now the second phase – illustrates disciplined scaling.

Third, **partner with workload‑focused firms** rather than trying to run a full cloud stack yourself. The ChronoScale spin‑off shows that shedding the cloud layer can sharpen your value proposition. Fourth, **guard against over‑reliance on a single customer**. Diversify across multiple hyperscalers or mid‑size AI firms to spread risk.

Finally, **keep a hard eye on operational costs**. The headline numbers – 75 MW, 250 MW, 400 MW – are impressive, but the underlying expense of turning those megawatts into reliable AI capacity is where profit lives or dies. If you can master that conversion at scale, you’ll have a defensible niche. If not, you’ll be another footnote in the hyperscaler‑dominated saga.

— Allan Ali, Founder

This article was produced with AI-assisted research and editorial support. Reporting is based on the source material cited below. Sources: Data Center Dynamics; datacenterdynamics.com; Global1.News (05 October 2026).

By Allan Ali, Global1.News