Market · North America
The United States. Deepest market, longest queue.
There is more capacity under construction here than anywhere on earth, and more of it is spoken for than anywhere on earth. Which market you pick decides your in-service date more than any other choice you will make.
Where the capacity actually is
- —Northern Virginia
Still the largest concentration of data center capacity in the world, and the hardest place to add to it. Substation and transmission constraints, plus local political resistance in Loudoun and Prince William, mean new large loads are quoted on timelines that surprise buyers who last transacted here three years ago. Excellent for existing inventory and east coast inference; a poor default for new build.
- —Dallas–Fort Worth and West Texas
The fastest large-load interconnection path in the country, because ERCOT sits inside one state and does not require multi-state coordination. Generation is being built specifically to serve this demand. You accept energy price volatility and, in most structures, a curtailment or demand-response obligation — which for training workloads is frequently an acceptable trade.
- —Phoenix and the Mountain West
Mature, well-served by multiple operators, and constrained mainly by water policy and transmission rather than land. Phoenix remains one of the more reliable places to get real wholesale capacity on a credible date; Salt Lake and Reno increasingly carry the overflow.
- —Atlanta and Columbus
The two markets that absorbed the most displaced demand from Northern Virginia. Both offer sales and use tax treatment that materially changes equipment economics, and both have utilities that have been comparatively constructive with large loads.
- —Chicago and the Midwest
Strong fibre, strong power, and a colder climate that helps at density. Chicago is the most credible tier-one alternative for east–central latency, and the surrounding secondary markets are underused relative to their power position.
- —The Gulf Coast and emerging sites
Where power is stranded, cheap or behind the meter. These are the sites that get a multi-hundred-megawatt requirement to a real date, and they are almost never on a buyer’s initial shortlist.
What actually decides your date
In the United States the building is rarely the problem. Steel and shell can be delivered on a schedule. What cannot be compressed is the utility, and everything downstream of it.
- The interconnection queue. In the multi-state RTOs, large-load and generation queues have become the single longest item on any development schedule. A developer telling you a site is “in the queue” has told you almost nothing — you need the position, the study stage and the expected energisation date in writing.
- Long-lead electrical equipment. Large power transformers, switchgear and generators are ordered years ahead. A site with secured equipment slots is a fundamentally different asset from a site without them, and that distinction almost never appears in a marketing package.
- Local politics. Several of the largest US markets now have active zoning resistance, noise ordinances and moratorium proposals. Entitlement risk is real risk and it belongs in the contract, not in the optimism.
- Water and cooling method. Water availability increasingly determines whether a site can support liquid cooling at scale, and in some jurisdictions it determines whether the site gets permitted at all. This is where cooling method and density stop being an engineering question and become a siting question.
- Tax treatment. Sales and use tax exemption on IT equipment is the incentive that moves the model, because it recurs at every hardware refresh. Property tax abatements usually carry employment or investment clawbacks that erode most of the benefit.
Where the requirement is large enough that no existing building fits, it becomes a powered shell or build-to-suit conversation, and the terms that matter change entirely. Where the constraint is the utility itself, it becomes a power procurement mandate first.
What we source here
Asked in this market.
01Is Northern Virginia still worth pursuing?
For existing inventory and for latency-critical east coast inference, yes. For new multi-megawatt capacity on a near-term date, it is usually the slowest answer on the list, and the honest comparison is against Ohio, Georgia, Texas and the Mountain West — where the same requirement often lands eighteen months earlier.
02Why does Texas come up so often for AI training?
Because ERCOT is a single-state interconnection with a materially faster large-load process than the multi-state RTOs, and because generation is being built to meet it. The trade is that you take on a more volatile energy price and, in most structures, some obligation around curtailment.
03What is realistic for a 20 to 50 MW requirement in the US?
In a tier-one market with an existing shell, twelve to twenty-four months. In a secondary market with a shovel-ready site and utility documentation, faster. Anything that requires new transmission is a multi-year conversation and should be planned as one.
04Do tax incentives actually move the number?
Sales and use tax exemptions on equipment are the ones that matter, because the equipment is the largest line item and the exemption applies at every refresh. Property tax abatements are worth less than they look once you model the clawback conditions.
05Can you source in secondary and emerging US markets?
That is usually where the answer is. Columbus, Atlanta, Salt Lake, Reno, the Gulf Coast and the Mountain West have delivered more usable near-term megawatts than the tier-one markets for several cycles now.
06Are export controls a factor for a US deployment?
Not for domestic deployment by a US entity. They become a factor the moment ownership, end use or downstream access touches a restricted jurisdiction, and that is worth establishing before the allocation conversation rather than after.
Data center capacity sourcing
Retail and wholesale colocation, benchmarked against comparables that are actually comparable.
Open the mandate →EnergyPower capacity procurement
Interconnection position, PPAs and behind-the-meter. In the US this is usually the mandate that matters first.
Open the mandate →ComputeGPU clusters and GPUaaS
Allocation and contract terms for B200, B300 and GB300 NVL72 across US sites.
Open the mandate →Tell us the megawatts and the date.
We will tell you which US markets can actually hit it, which are quoting optimism, and what the contract has to say about being late.