The Grid Queue Arbitrage

In partnership with

Following last week's analysis of circular water economics in semiconductor-nuclear integration, this week we examine a different kind of arbitrage: the temporal advantage of bypassing interconnection queues entirely.

The Queue That Never Moves

The numbers tell a sobering story. Whilst Stargate commits £395bn ($500bn, €463bn) to AI infrastructure and the UAE launches its 4GW nuclear-powered AI campus, 2,600 GW of generation capacity sits stranded in US interconnection queues. For context, that's more than double America's entire electricity generation capacity, waiting sometimes for a decade to connect to the grid. Last month, a Fortune 500 tech company abandoned a £1.6bn ($2bn, €1.9bn) data centre project after three years in PJM's queue. They're not alone.

Here's the disconnect: hyperscalers need AI compute online by 2027. Traditional grid interconnection won't deliver until 2035 at earliest. The maths doesn't work. Yet solutions exist, if you know where to look.

The PJM Bottleneck Nobody's Solving

PJM Interconnection, the world's largest wholesale power market serving 65 million people across 13 US states, reports 2,600 GW trapped in its interconnection queue as of October 2025. The Federal Energy Regulatory Commission's Order 2023 promised to streamline the process. Industry reaction? NextEra Energy called it "insufficient to address the fundamental capacity constraints." Grid operators push for "cluster studies" and "fast-track pathways," acknowledging the system's fundamental breakdown.

The Lawrence Berkeley National Laboratory data reveals the acceleration: queue volumes grew 47% in 2024 alone. Not because more projects exist, but because nothing moves through the system efficiently. Average wait times now exceed 5.2 years. For solar projects, 7.3 years. For offshore wind connecting to coastal substations, a decade or more.

Meanwhile, EirGrid in Ireland reports 15 GW queued (triple installed capacity), National Grid UK faces 350 GW waiting, and Korea's KEPCO manages 400 GW in limbo. This isn't an American problem. It's a grid architecture problem.

Why Traditional Interconnection Fails for AI Data Centers

Scale Mismatch

A 500 MW hyperscale AI data centre powering 100,000 GPUs requires instantaneous, uninterruptible power delivery at 99.999% uptime. Grid interconnection assumes variable renewable generation with storage buffering. The models literally don't compute. PJM's interconnection studies evaluate impact on transmission stability, not AI workload reliability. The wrong question produces unusable answers.

Reliability Requirements

AI training demands six-nines uptime (99.9999%). Grid-connected facilities introduce multiple points of failure: substation transformers, transmission line exposure, regional grid instability, NERC compliance curtailments. Each adds complexity and reduces availability. Google's 2024 infrastructure report confirms: behind-the-meter nuclear configurations deliver 99.997% uptime versus 99.92% for grid-sourced power. That 0.077% difference translates to £63m ($80m, €74m) annually in lost GPU utilization for a £3.9bn ($5bn, €4.6bn) AI training cluster.

Economic Penalties

Grid-connected facilities pay interconnection fees, transmission charges, capacity payments, and congestion costs. For a 500 MW data centre in PJM, these can exceed £27m ($34m, €32m) annually. Pure overhead that behind-the-meter configurations avoid. The Pennsylvania Public Utility Commission's 2025 analysis confirms: co-located facilities pay 18-23% less for electricity on a lifecycle basis.

The Constellation-Microsoft Precedent

Microsoft's £1.6bn ($2bn, €1.9bn) agreement with Constellation Energy to restart Three Mile Island Unit 1 demonstrates the approach. 835 MW of dedicated nuclear capacity flowing directly to Microsoft's mid-Atlantic data centre cluster. No grid interconnection, no PJM queue, no waiting. Power purchase agreement signed in September 2024, Unit 1 operational target 2028. Four years from deal to electrons versus 7-10 years for traditional grid-connected projects.

The Nuclear Regulatory Commission's approval process took 18 months, focused exclusively on reactor safety. No transmission impact studies. No grid stability modelling. No capacity market complications. The behind-the-meter configuration eliminated eight years of regulatory overhead.

France's EDF replicates this at Bugey (320 MW data centre), Belgium explores Doel co-location, and China's CNNC operates 1.2 GW across Qinshan, Yangjiang, and Tianwan facilities. The model scales globally.

The Strategic Disconnect: Temporal Arbitrage

Here's what market observers miss: the temporal disconnect between AI infrastructure needs and grid development timelines creates a structural advantage for behind-the-meter nuclear configurations.

Projects requiring grid interconnection face:

• 2-3 years for interconnection application and queue position

• 3-4 years for system impact and facilities studies

• 2-3 years for network upgrades and construction

• Total: 7-10 years before operation

Behind-the-meter nuclear projects bypass most delays:

• 6-12 months for site selection and PPA negotiation

• 12-18 months for NRC license amendments (restart cases)

• 0 years for transmission interconnection

• Total: 2-3 years to first power

The arbitrage opportunity is temporal, not just financial. First movers capture seven years of AI compute advantage whilst competitors wait in interconnection queues. At current GPU cluster economics (£79,000 per rack annually, $100,000, €93,000), a 500 MW facility generates £2bn ($2.5bn, €2.3bn) in cumulative compute value during that seven-year head start. Compound that across multiple facilities, and the strategic advantage becomes insurmountable.

Regulatory Evolution: Behind-the-Meter as Critical Infrastructure

The US designation of AI training facilities as 'critical national security infrastructure' in 2024 reshaped regulatory treatment. When data centres become critical infrastructure, behind-the-meter configurations gain national security justification. Grid vulnerability becomes a feature to avoid, not accept.

FERC's traditional "open access" principle assumes generation serves broad public benefit. For AI national security infrastructure, engineering reality suggests otherwise. Regulators acknowledge what engineers have known: reliability trumps market efficiency when national competitiveness depends on uninterrupted compute. The UK's Department for Energy Security and Net Zero issued similar guidance in August 2025, with Japan's METI following in September.

The Path Forward

The solution isn't fixing interconnection queues. It's recognising when grid connection isn't the answer. For nuclear-AI infrastructure, three principles emerge:

1. Proximity Trumps Markets: Every metre of transmission adds failure points, latency, and cost. Behind-the-meter configurations eliminate these entirely. The most reliable electron is the one that never touches the grid.

2. Reliability Through Simplification: The most resilient power system is the one that minimizes components. Direct nuclear-to-data centre configurations reduce complexity by orders of magnitude compared to grid-mediated delivery.

3. Speed Through Bypass: Whilst others wait in interconnection queues, behind-the-meter projects begin generating returns. First-mover advantages compound in AI markets where weeks of compute lead translates to product dominance.

Investment Implications

For stakeholders evaluating nuclear-AI opportunities, interconnection constraints reshape investment criteria:

Existing Nuclear Assets: Prioritize facilities with available capacity near network hubs. Three Mile Island's £59 per MWh ($75, €69) represents 40% premium to PJM wholesale, justified by zero interconnection wait and guaranteed off-take.

Geographic Arbitrage: The differential between 10-year queue wait and 2-year behind-the-meter deployment may prove more valuable than any technology advancement. Pennsylvania, Ontario, and northern France offer optimal nuclear capacity and hyperscaler demand combinations.

Temporal Advantage: Early movers capture seven years' AI compute advantage whilst competitors navigate interconnection bureaucracy. Training data scale translates to model superiority and market dominance.

The Bottom Line

The 2,600 GW stuck in interconnection queues represents trapped value and opportunity. Whilst conventional wisdom focuses on fixing queue processes, engineering reality points elsewhere: bypass the grid entirely.

The winners in nuclear-AI infrastructure won't navigate interconnection most efficiently. They'll recognize when grid connection itself became the problem and engineer accordingly.

As one former FERC commissioner noted: "We spent three years and £79m ($100m, €93m) improving our process before realizing: for dedicated large loads, direct connection isn't a loophole, it's the correct engineering solution."

The question isn't how to accelerate interconnection. It's whether grid connection makes sense at all.

Next week: We examine data center developer-nuclear operator partnerships: why the most obvious collaboration model remains the least implemented.

The AI Insights Every Decision Maker Needs

You control budgets, manage pipelines, and make decisions, but you still have trouble keeping up with everything going on in AI. If that sounds like you, don’t worry, you’re not alone – and The Deep View is here to help.

This free, 5-minute-long daily newsletter covers everything you need to know about AI. The biggest developments, the most pressing issues, and how companies from Google and Meta to the hottest startups are using it to reshape their businesses… it’s all broken down for you each and every morning into easy-to-digest snippets.

If you want to up your AI knowledge and stay on the forefront of the industry, you can subscribe to The Deep View right here (it’s free!).