Everyone talks about chips. Nobody talks about what's actually blocking AI from getting built.
The U.S. grid interconnection queue: 2,600 GW backlog. Median wait approaching five years. Some data center projects facing delays of up to 12 years. Of all capacity that applied for interconnection between 2000 and 2019, only 13% reached commercial operation by end of 2024. 77% was withdrawn.
For reference — the startup survival rate from pre-seed to seed is roughly 10-15%. We accept that as the brutal nature of early-stage investing. The grid interconnection queue has the same odds. Except here, the projects that don't make it aren't bad ideas. They're just stuck in line.
Working on data center projects in the mid-Atlantic right now, the picture is clear. Utilities like PPL are already receiving more GWs of data center applications than their historical maximum peak demand. Grid power isn't guaranteed. It has to be engineered around.
Which is why behind-the-meter generation has become the critical near-term workstream — not because it's the cleanest option, but because it's the one that moves.
BTM now accounts for over 25% of all new data center capacity. 56% of developers are exploring on-site or co-located generation. This isn't a workaround anymore. It's the strategy.
The business models are evolving too. Some developers are building out a dedicated on-site powerhouse and bringing in a specialist operator to run it — keeping capital control while outsourcing the operational complexity. Others are structuring straightforward PPAs with third-party energy providers who install, own, and operate the generation assets on-site, selling power directly to the data center at a contracted rate. Each model has different risk profiles, balance sheet implications, and timelines. The right one depends on the developer's appetite for owning infrastructure versus just consuming electrons.
That said — nobody is financing projects that run exclusively on BTM with no grid connection. And for good reason. It's still high-risk, operationally and commercially. Lenders want a grid backstop. Offtakers want redundancy. The math doesn't close without it.
But watch this space over the next three years. It only takes one or two projects — even at 20-30 MW — to demonstrate that BTM-primary can operate reliably and bankably. Once those case studies exist, the conversation shifts fast. Capital follows proof. And regulators tend to build frameworks around what's already working in the field, which could itself accelerate permitting for the next wave.
The unlock for truly off-grid data centers won't be a financing structure or a policy change. It will be an energy product — something that can be designed, permitted, and deployed fast enough to match the pace of compute buildout. That product doesn't fully exist yet at scale. The team that cracks it is building one of the most important infrastructure companies of the decade.
The sustainability tension is real and worth being honest about. Natural gas BTM conflicts with 24/7 carbon-free commitments. Most operators are treating it as a bridge — gas now, SMRs or clean alternatives later. That's the honest framing.
On permitting: AI tools can accelerate document review, environmental cross-referencing, public comment analysis. Virginia's permit digitization platform already cut approval times by 70%. But AI speeds up the paperwork. It doesn't fix the physics — and it doesn't fix the organizational problem underneath it.
The real permitting bottleneck isn't just regulatory. It's internal. Product teams, legal, deployment, and project development are rarely aligned early enough. By the time a jurisdiction's specific requirements surface — local emissions standards, grid interaction rules, structural mandates — the project is already in motion. Some of those requirements aren't paper fixes. They're product changes. Retrofitting compliance into a design that's already locked is expensive and slow. The teams that are winning are treating jurisdictional requirements as an input to product development, not a checklist at the end.
The projects winning right now aren't the ones with the best ESG story. They're the ones that show up with power already figured out — and compliance baked in from day one.
REFERENCES
[1] Lawrence Berkeley National Laboratory / interconnection.fyi — U.S. Interconnection Queue Data Through 2025
Key stats: 2,060+ GW actively seeking grid connection as of end of 2025; only 13% of projects applying between 2000–2019 reached commercial operation by end of 2024; 77% withdrawn.
[2] Enki AI — Grid Interconnection Delays 2026: A Threat to US Energy
https://enkiai.com/ai-market-intelligence/grid-interconnection-delays-2026-a-threat-to-us-energy/
Key stats: U.S. queue swelled to 2,600 GW backlog; median wait approaching five years; data center projects facing delays up to 12 years; ~80% withdrawal rate.
[3] Camus Energy — Why Does It Take So Long to Connect a Data Center to the Grid?
https://www.camus.energy/blog/why-does-it-take-so-long-to-connect-a-data-center-to-the-grid
Key stats: CenterPoint Energy reported 700% increase in large load interconnection requests (1 GW to 8 GW, late 2023 to late 2024); PPL and other utilities receiving more GWs of applications than historical peak demand; transmission and generation buildout takes 5–10 years.
[4] Landgate — Behind-the-Meter Report: 2026 Outlook
https://www.landgate.com/news/behind-the-meter-report-2026-outlook
Key stats: BTM solutions now account for over 25% of all new data center capacity.
[5] datacenterHawk — Behind-the-Meter Power Solutions: The Data Center Industry's New Reality
Key stats: Foley 2026 Data Center Survey — 56% of developers exploring co-located or on-site power generation; tension between natural gas BTM and 24/7 carbon-free commitments.
[6] Data Center Knowledge — How Data Centers Redefined Energy and Power in 2025
Key stats: Projected shortfall of over 45 GW; operators deploying natural gas now while evaluating SMRs for future baseload.
[7] Global Data Center Hub — The Interconnection Bottleneck
https://www.globaldatacenterhub.com/p/the-interconnection-bottleneck-how
Key stats: Virginia permit digitization platform cut approval times by 70%; over $22 billion in renewable projects canceled in H1 2025; PJM capacity auction ballooned from $2.2B to $14.7B in one year.
[8] Data Center Frontier — Project Stalled: Grid Bottlenecks Threaten the Fifth Industrial Revolution
Key stats: PJM rules create faster pathway for combined data center and power-generation projects; President Trump's 2026 State of the Union "ratepayer protection pledge" — tech companies expected to build or finance their own power plants.
