The conversation about AI infrastructure has shifted. The binding constraint is no longer silicon. It is electrons, water, and licensed electricians.
What Happened
Three converging signals this week make the case emphatically.
First, Denmark has halted new grid connection agreements entirely as AI-driven demand overwhelms Nordic transmission capacity, per Data Center Knowledge. This is not a temporary queue. It is a hard stop, and it mirrors a broader European pattern where country-by-country grid and permitting divergence is concentrating hyperscale builds in whichever markets can most quickly deliver energized space. As Data Center Knowledge reports, France, Spain, and Poland are emerging as beneficiaries while markets like the Netherlands and now Denmark effectively close to new large-scale builds.
Second, in the US, The Next Platform reports that skilled electrical trades, not GPU allocation, are now throttling data center commissioning timelines. Facilities can be permitted and partially built with GPUs staged and waiting. Without licensed electricians to terminate high-voltage switchgear and commission the power distribution units (PDUs, the in-rack power strips that feed servers), nothing goes live. Chip availability is moot if the facility cannot be energized.
Third, Applied Digital signed a 430MW lease with an unnamed hyperscaler, one of the largest single colocation transactions on record. Simultaneously, TeraWulf is converting a 500MW former coal plant in Buffalo into an AI campus, bypassing the multi-year greenfield grid interconnection queues that plague new builds. Hyperscalers (the largest cloud providers, specifically AWS, Azure, GCP, and Oracle) are locking in decade-scale capacity commitments at unprecedented power densities. The window for smaller clients to secure favorable colocation terms in the same markets is narrowing fast.
Why It Matters
The structural pattern here is important. Power scarcity is not evenly distributed. It is highly localized, and that localization is becoming the dominant variable in site selection, ahead of latency, labor cost, or tax incentives.
In the US, ERCOT in Texas, with its existing high-voltage CREZ (Competitive Renewable Energy Zone) transmission corridors, is attracting a disproportionate share of hyperscale AI builds. Northern Virginia (NoVa), the largest US data center market, faces increasing power queue congestion. Phoenix and Dallas remain viable but are tightening. Markets with repurposed industrial power infrastructure, like Buffalo, are suddenly more competitive than greenfield alternatives.
Water is compounding the constraint. Municipal water and wastewater limits are now functioning as siting gatekeepers, pushing operators toward closed-loop cooling and air-cooling architectures. For frontier labs running dense GPU clusters, the cooling spec of a colocation facility is no longer a secondary consideration. It is a qualification criterion.
For Fortune 500 enterprises entering AI infrastructure for the first time, this environment is disorienting. Snowflake's $6 billion AWS commitment signals that enterprise AI workloads are graduating from pilots to always-on production deployments. That shift structurally increases demand for both hyperscaler reserved capacity and Tier III colocation space (Tier III is a data center reliability classification guaranteeing 99.982% uptime). As hyperscalers absorb gigawatt-scale blocks at the wholesale level, available retail and mid-market colocation inventory at operators like Equinix, Digital Realty, QTS, CyrusOne, Aligned, and Iron Mountain compresses. The enterprises arriving late to this market will find fewer options at worse terms.
For sovereign AI programs in the EU, the Denmark situation is a warning. Governments that assumed national grid infrastructure would accommodate national AI ambitions on a flexible timeline need to reassess. The queue dynamics are identical to the commercial market and in some cases worse due to procurement cycle friction.
What Clients Should Do
If you are a frontier lab or a large sovereign AI program planning a 5,000 to 20,000 GPU training cluster, the colocation conversation needs to happen before the GPU procurement conversation. A facility that cannot support 100kW-plus per rack with adequate power redundancy and liquid-cooling infrastructure is not a viable host, regardless of price. Engage colocation brokers now to identify which Tier III operators in Texas, NoVa, Chicago, Atlanta, Phoenix, or key EU markets can actually support your density requirements and have capacity available within your deployment window.
If you are a Fortune 500 enterprise or system integrator sourcing AI infrastructure for a client in the 1MW to 20MW range, do not assume hyperscaler managed services are your only path. Neocloud operators (specialized GPU cloud providers, distinct from AWS, Azure, and GCP) often deliver the same H100 or H200 capacity at 30 to 50 percent lower cost, with ramp times measured in weeks rather than quarters, and with more flexible contract structures. Running a portfolio across one hyperscaler, one to two neocloud operators, and a direct colocation footprint is the approach sophisticated infrastructure teams are executing today.
If you are an AI scaleup with inference workloads that have graduated to production, neocloud operators are structurally better suited to your needs than hyperscaler on-demand pricing. The integrated inference-plus-training platforms emerging in the neocloud market are purpose-built for this workload profile in ways that general-purpose hyperscaler GPUs are not.
In every case: earlier conversations get better terms. The Applied Digital 430MW deal and the TeraWulf conversion are not anomalies. They are the rate at which quality capacity is being absorbed. What remains for mid-market clients is contracting now, not after the next planning cycle.
How XIRR Advisors Can Help
XIRR Advisors brokers reserved GPU capacity from neocloud operators and Tier III colocation space across the USA. We represent the client. Providers pay our fee. Clients pay nothing.
Share your requirements, region, GPU type (H100, H200, B200, GB200), cluster size, and timing, or megawatt requirement for colocation, and we will canvas the neocloud and colocation markets on your behalf and return a shortlist within 48 hours. Many clients need both GPU capacity and colocation infrastructure simultaneously. We source both. The teams that engage us earliest consistently secure better pricing and faster ramp timelines than those who wait. Reach us at contact@xirradvisors.com or DM @XIRRAdvisors.
References
[1] Data Center Knowledge: Denmark Halts New Grid Agreements as AI Demand Overwhelms Capacity
[2] Data Center Knowledge: Power and Permitting Redraw Europe's Data Center Expansion Map
[3] The Next Platform: Electricians, Not GPUs, Are Now AI Infrastructure's Binding Constraint
[4] Data Center Dynamics: Applied Digital Signs 430MW Lease with Unnamed Hyperscaler
[5] Data Center Knowledge: TeraWulf Converts 500MW Buffalo Coal Site Into AI Campus With Schneider
[6] Data Center Knowledge: Texas CREZ Wind Corridors Attract Hyperscale AI Build-Out
[7] Data Center Knowledge: Water Emerges as Next Hard Constraint for Hyperscale AI Siting
Share your requirements. We'll canvas the market.
Tell us your needs (region, GPU type, capacity, timing — or MW for colocation) and we'll canvas the neocloud and colocation markets on your behalf. Shortlist in 48 hours.
Earlier conversations get better terms. When you engage early, we have time to negotiate with vendors before you need to commit. You pay nothing. Provider-paid model.