Applied Digital announced on October 6 that it had entered an agreement giving it access to up to 1 GW of potential power capacity in Finland. The company describes this as its first development opportunity outside the United States. Its primary announcement anticipates initial power availability beginning in 2028.
Those qualifiers matter. Potential power is a development input. It does not establish that a campus is energized, equipped or ready to accept an AI workload.
The announcement says the site is being marketed and hyperscale customer discussions are underway. It also ties future investment and development decisions to demand, commercial milestones and regulatory and development requirements. The dated wire release provides the same company statement, rather than independent confirmation of delivery.
My interpretation is that infrastructure buyers should treat this as an opportunity to investigate future capacity. They should not insert the headline figure into an available-compute inventory.
Track the stage behind each number
A capacity spreadsheet can become misleading when it combines development opportunities with operating facilities in one column. The numbers may look comparable even though the underlying commitments are very different.
For a proposed AI campus, I would track distinct stages: access to a potential power resource, the connection and delivery arrangements, a commissioned facility, installed IT equipment, and capacity allocated to a customer. Each stage needs its own evidence and date.
This is a suggested diligence framework, not a description of additional terms in Applied Digital's agreement. The public release does not provide enough detail to fill every stage.
Ask what an advertised power figure measures. Is it the maximum eventual campus input, an initial phase or an allocation available under a defined delivery arrangement? Ask separately about the IT capacity that can be supported and the compute service being sold. A power figure alone does not answer either question.
Put 2028 into a dependency plan
An anticipated first-power date should become a dependency in the buyer's plan. It is not, by itself, a workload start date.
A customer evaluating future capacity needs to know the sequence from power availability through facility commissioning, equipment installation, acceptance and service handover. The dates should distinguish expectations from contractual commitments and identify which dependencies remain unresolved.
That distinction also helps internal communication. A technical team may understand that a project is still in development, while a commercial forecast quietly treats the same date as the start of revenue-generating compute. A shared milestone record can expose the mismatch before either team makes a commitment.
For an AI deployment that cannot move with the construction schedule, consider how the workload would run if the preferred site became available later than expected. Test the practical requirements for another location, including data transfer, software environment and operating support. This is contingency planning, not a prediction that the Finland project will be delayed.
Verify the network path before choosing the site
A Nordic location may be attractive for several reasons, but geography alone cannot establish application performance.
For training, examine where the source data lives and how it will reach the cluster. For inference, define the users and services that need a response, then test the relevant network paths. Include the routes used during failover rather than relying only on the normal path.
The appropriate evidence is a measurement or a clearly scoped design for the service being purchased. A statement about national connectivity does not replace a test between the endpoints the workload will use.
Storage and checkpoint handling deserve the same treatment. Work out which data must remain near compute, which can move elsewhere and how recovery would work after interruption. A proposed campus gives the buyer time to ask these questions before architecture decisions become expensive to change.
Make customer readiness a separate deliverable
Even an operational facility does not automatically provide the service a particular customer needs. The buyer must establish the equipment, software, access controls, support arrangements and acceptance tests associated with its allocation.
Keep that record separate from the developer's construction milestones. It should explain what the customer receives, who operates each layer and which tests demonstrate readiness for the intended workload.
For technical B2B teams presenting a future site to customers, use the same discipline in sales materials. State the development stage alongside the capacity figure and explain which milestones would make that capacity usable. That gives engineering and purchasing teams something concrete to evaluate.
Applied Digital's announcement establishes a new international development opportunity. The next useful evidence will show how the potential power translates into connection arrangements, project phases and customer-ready infrastructure. Until then, the most useful question is which capacity gate has been passed, and what still needs to happen before a workload can run.
Top comments (0)