10 criteria for choosing an AI data center colocation provider for the long term
The technical evaluation of an AI data center colocation provider always ends up converging. The difficult part comes afterward, when it is time to secure internal approval for a multi-year commitment from stakeholders who do not think in terms of PUE or kW per rack and who primarily view a long-term contract as a financial risk to be managed.
From that point on, success depends on the ability to translate an infrastructure decision into the language of each decision-maker. Here is how to build that business case and the ten criteria on which to base it.
TL;DR Choosing an AI data center colocation provider is ultimately won internally, not just on a technical specification sheet. To secure approval for a 6- to 9-year commitment, the infrastructure must be translated into the language of three key stakeholders. Finance requires budget predictability (OPEX vs. CAPEX), procurement demands objective and measurable selection criteria (SLAs, exit provisions), and governance requires absolute confidence in the operator’s long-term viability.
Why a Technical Evaluation Alone Is No Longer Enough When Choosing an AI Data Center Colocation Provider
The neocloud market could approach $400 billion by 2031, growing at nearly 58% annually according to analyses by Synergy Research Group. This trajectory is attracting capital, but it also serves as a reminder that the sector remains highly capital-intensive and that not every player will endure. Any committee approving a multi-megawatt IT commitment over nearly a decade understands this reality, making the decision particularly sensitive.
The technical case has already been made. The challenge shifts to three internal objections that have nothing to do with technology:
- a financial objection regarding the nature of the commitment,
- a process-related objection regarding the selection criteria,
- and a risk-related objection concerning the operator’s long-term viability. Each has its own stakeholder and its own language.
Finance: Turning a Long-Term Commitment into Budget Predictability
The first—and entirely legitimate—source of resistance almost always comes from the finance department. From a distance, reserving capacity with an AI data center colocation provider for six to nine years can look like an implicit liability.
The argument that changes this perception comes down to one key point: colocation is predictable OPEX. Reserving a dedicated block of 6 to 7 MW of IT capacity means budgeting for capacity over time, with a clearly defined annual cost, whereas building a facility in-house would require substantial upfront capital investment in infrastructure that must then be operated internally.
The long-term commitment therefore stops being a constraint and becomes a planning tool, as it locks in pricing conditions for the duration of the agreement and protects the budget from the volatility that on-demand models can impose on margins.
This is also where total cost of ownership comes into play. A native DLC infrastructure designed for a PUE of 1.2 and an optimized AI inference TCO directly translates energy efficiency into cost per token—a metric that finance teams immediately understand. The commitment becomes a controlled variable within the economic model.
Procurement: Objective and Verifiable Selection Criteria
Procurement wants to understand the basis on which a partner was selected. The decision must therefore be supported by objective, measurable, and enforceable criteria before any contract is signed.
Three elements are enough to structure the discussion.
- First, a Tier III to Tier III+ certification that can be independently verified through the Uptime Institute.
- Second, an SLA with availability commitments defined component by component, backed by meaningful financial penalties that also cover direct liquid cooling (DLC). This topic is explored in detail in our article on what an AI colocation SLA should guarantee.
- Finally, a clearly identified industrial operator named in the contract, with explicit operational accountability.
One additional criterion is particularly appreciated by procurement because it limits risk: reversibility and exit provisions. A serious multi-year agreement clearly specifies how capacity can be increased, the required notice period, and the applicable conditions if strategic priorities change. A provider willing to put these clauses in writing is a provider prepared to stand behind its commitment in both directions.
Governance: Assessing the Long-Term Viability of an AI Data Center Colocation Provider
The final objection is the most direct: “What if the operator no longer exists in five years?”
The right answer lies in a risk structure that can be objectively assessed. At Voltekko, operations are supported by EQUANS, a subsidiary of the Bouygues Group, which has been designing, building, and operating data centers for around twenty years. Financial backing comes from REED, the infrastructure fund of the Société Générale Group, which finances the network rollout and validates its business model. These are two institutional counterparties whose strength does not depend on the next funding round.
The value of this three-tier structure (operator, industrial operator, and institutional investor) is that it reduces concentration risk around a single entity while providing reassuring operational continuity over the lifetime of a long-term commitment.
Checklist: 10 Criteria for Evaluating Your AI Data Center Colocation Provider Over a 9-Year Horizon
Below are the ten technical and contractual criteria to review before committing, presented in a format that can be incorporated directly into your internal evaluation frameworks:
| No. | Contractual Selection Criterion | Target Indicator |
|---|---|---|
| 1 | Infrastructure certification | Tier III to Tier III+ certification verifiable through the Uptime Institute. |
| 2 | Availability granularity | Availability defined for each critical component (not a global SLA). |
| 3 | Cooling redundancy | N+1 architecture explicitly covering DLC systems. |
| 4 | Financial penalties | Service credits applicable to DLC liquid-cooling infrastructure. |
| 5 | Service restoration | Contractual MTTR guaranteed by incident type. |
| 6 | Resource allocation | Dedicated capacity block expressed as guaranteed IT power (kW). |
| 7 | Flexibility and scaling | Defined expansion terms and notice period (e.g., 90 days). |
| 8 | Risk management | Explicit reversibility and exit provisions. |
| 9 | Ecosystem strength | Named industrial operator and top-tier institutional backing. |
| 10 | Sovereignty and compliance | Infrastructure location and compliance with European law. |
These ten criteria are the condensed version of the business case. The full version, including the contractual language that should be required for each point, is available for download from our team.
From Technical Selection to a Defensible Decision
Ultimately, what you defend is not a technology but a strategic decision supported by quantified guarantees, verifiable selection criteria, and a transparent risk structure. Finance, procurement, and governance each require different forms of evidence. Bringing their answers together in a single business case, supported by data and contractual commitments, is what secures approval.