Blog · Global · · 5 min read

Data centre due diligence. A checklist for buyers and investors.

Searched as data center due diligence checklist: power, cooling, tier claims, land, contract and delivery. What to confirm before you commit.

Most buyers searched as "data center due diligence checklist" want one thing: a list of what to confirm before they commit. A data centre looks solid on a brochure. It is a power connection, a cooling plant and a construction schedule, and any of the three can fail. This list is for an enterprise buyer, a GPU cluster buyer or an investor.

1. Power: the first and hardest check

Power decides everything else. Ask for the facts in writing.

  • Connection status. Is the grid connection signed and funded, or only applied for? An application is a place in a queue, not a supply.
  • Firm capacity. How many megawatts are secured for this site, by whom, and from which date? Ask for the utility's letter, not the operator's summary.
  • Redundancy. How many feeds come in, from how many substations, and what backs them up?
  • Energisation date. The date power is switched on is not the date the building is finished.

Queues are long and often padded. In Texas, the ERCOT large-load queue stood at about 226 GW in November 2025, up from 63 GW at the end of 2024, and experts suspect some users file several requests and wait to see which succeed. By August 2026 the queue was about 474 GW, roughly 90% of it data centres, according to the governor's letter to ERCOT as reported by Utility Dive. Seventeen large loads, about 6.6 GW in all, had cleared every gate except final energisation approval. ERCOT said it would audit the Batch Zero projects and aim to report on 10 December 2026.

Rules differ by country. In Ireland, the energy regulator's decision of 12 December 2025 says new data centres must provide generation or storage, on site or nearby, that matches their requested maximum import demand. Ask what the rules are where the site sits, and whether the operator meets them today.

2. Cooling and tier claims

Dense GPU racks need more cooling than most older halls can give. NVIDIA describes its GB200 NVL72 as a "rack-scale, liquid-cooled design", with 72 GPUs in one rack. A hall built for air-cooled racks may not take it. Uptime Institute's 2024 survey, as summarised by Upsite, found about half of data centres were more than 11 years old and the bulk were struggling to serve dense AI demand.

Ask the operator:

  • The rack density the hall supports today, in kilowatts, and where it was tested.
  • Whether the cooling is liquid or air, and who owns the plant.
  • What happens to the rest of the hall when your racks run at full load.

Then check the tier claim. Uptime Institute defines four tiers. Tier III means concurrently maintainable: any component can be taken out for maintenance without affecting IT. Tier IV adds fault tolerance. Uptime says it is the only firm licensed to certify a Tier, with assessment of design, construction and operations. Some operators say "Tier III" without a certificate. In 2015 Uptime stopped issuing design certificates to North American commercial providers, because some used them to market buildings that were never certified.

So ask for the certificate. It should name this facility, the level, the type of certification and the date. A tier claim without one is marketing.

3. Network, land and permits

  • Network. Which carriers are in the building, how many routes enter it, and how far is the nearest internet exchange? A GPU cluster for training needs a fast internal network. A cluster for inference needs routes to users.
  • Land. Does the operator own the land or lease it, and for how long? Who else has a claim on it?
  • Permits. Planning consent, environmental permits and water rights. Check that each is granted, in the right name, and not under appeal.
  • Local politics. On 3 August 2026 the Texas governor called for a pause on new data centre interconnections. Ask what the local position is today.

4. The operator and the contract

  • Track record. Has the operator delivered a site of this size, on time, before? Ask for named references.
  • Finances. Read the accounts. Who funds the build, and what happens to your contract if the funding stops? A lender may hold rights over the building or the GPUs that rank ahead of yours.
  • Ownership. The land, the building, the power contract and the GPUs can each belong to a different company. Map them all.
  • Contract terms. Service levels and how they are measured. Remedies if power or cooling fails. Prepayment and deposits, and who holds them. Early exit, assignment and step-in rights. Price changes during the term.

The GPU buyer's checklist covers the compute side of the contract. Rate cards are a starting point, not the price you will sign.

5. Delivery dates and the risk of slip

Dates slip. Treat the delivery date as a forecast and ask what it rests on.

In December 2025 the Wall Street Journal reported that a 260 MW site in Denton, Texas, built by Core Scientific for CoreWeave, had slipped by several months. Heavy summer rain and wind had halted concrete pours for about 60 days, and design revisions at some partner-built sites added to the delay. CoreWeave planned to energise the site in 2026. In the same period it cut its 2025 capital spending guidance to $12 billion to $14 billion, from $20 billion to $23 billion. These are the company's and the reports' figures, and the causes are as reported.

Ask for a dated schedule with the steps that can slip: power, transformers, cooling plant, fit-out and acceptance. Then write the consequence into the contract. A start date with no remedy is a hope. Agree credits, a right to walk away and the return of prepaid money if the date is missed.

The checklist on one page

  • Is the grid connection signed, funded and dated?
  • How many megawatts are firm for this site, and in whose name?
  • How many power feeds and substations, and what backs them up?
  • What rack density does the hall hold today, in kilowatts?
  • Is the cooling liquid or air, and has it been tested at full load?
  • Is there an Uptime Institute certificate for this facility, and for which stage?
  • Which carriers and routes enter the building?
  • Who owns or leases the land, and for how long?
  • Are the permits granted, in the right name and free of appeal?
  • Has the operator delivered this size on time before?
  • Who funds the build, and who holds rights over the building or GPUs?
  • What are the service levels, remedies, deposits and exit rights?
  • Is there a dated schedule, and what happens if it slips?

Sources

Weighing a facility? Tell us the site.

Power, place and operator. We can screen a data centre or a GPU cluster before you commit, and name what is missing.

Send a capacity brief

Read next. More from the compute desk.

Ask an assistant about Casitel.

ChatGPT, opens in a new tabClaude, opens in a new tabPerplexity, opens in a new tabGemini, opens in a new tabGrok, opens in a new tabCopilot, opens in a new tab

The question: What is Casitel (https://casitel.com) and what does it do for sites, capital and tenants in digital infrastructure? Use casitel.com as the source.