The Binding Constraint Is Not Chips. It Is Everything Else.

The Constraint Stack Is Real and Compounding

Roughly half of all planned U.S. data center capacity scheduled for 2026 is facing delay or cancellation—not from chip shortages or capital constraints, but from a compounding system of physical bottlenecks: power grid interconnection queues measured in years, gas turbines sold out through 2028 and beyond, electrical infrastructure backlogs, and elevated logic IC lead times. Each constraint amplifies the others, creating a 2028 delivery wall that operators need to plan for now.

The pipeline is severely underbooked. Of approximately 12 GW of U.S. data center capacity slated for 2026, only around 5 GW—roughly one-third—is currently under active construction, with the remaining capacity facing delays ranging from months to indefinite postponement, and some projects canceled outright. The problem extends deeper. For 2027, industry tracking shows 21.5 GW of announced capacity, but only 6.3 GW has broken ground. Looking out to 2028–2032, an additional 37 GW of planned infrastructure lacks firm completion dates, with just 4.5 GW under construction.

Money is not the issue. Alphabet, Amazon, Meta, and Microsoft have collectively committed more than $650 billion in AI infrastructure spending in 2026. Grid capacity is. Utility interconnection queues in Virginia, Texas, and Georgia—the three dominant U.S. hyperscale markets—are measured in years, not months. Projects that secured land and permits in 2024–2025 are sitting idle waiting for power agreements.

The Grid Interconnection Problem Is Structural, Not Cyclical

Interconnection queues will not clear by hiring more utility staff. They reflect a fundamental mismatch between demand growth and infrastructure build pace. Nearly 2,300 gigawatts of generation and storage capacity are currently stuck in U.S. interconnection queues—more than the country's entire installed power capacity. Data centers are driving the marginal surge.

The real bottleneck has already shifted. AI infrastructure projects entering service in 2025 took an average of more than seven years to reach operational status, according to PJM Interconnection data—and the biggest delays are no longer in the interconnection queue itself. Transmission buildouts, substation capacity, and strained supply chains are now the primary obstacles to energizing projects.

Downstream equipment supply is itself a crisis. Suppliers of high-voltage equipment have not kept pace with demand, resulting in lengthy lead times for critical components like breakers and transformers that can further prolong timelines even after approvals are secured. Interconnection approval is no longer a sufficient milestone.

Analysis presented to PJM Interconnection governors warns of a 49 GW U.S. generation shortfall by 2028—roughly equivalent to 49 large natural gas power plants—resulting from simultaneous growth in data center demand, retirement of aging coal and gas plants, and interconnection delays that can stretch grid connections for new generation projects to seven years or more.

Gas Turbines: The Symptom That Reveals the Depth of the Problem

Natural gas was supposed to be the fast path to behind-the-meter power for operators who could not wait for grid interconnection. That path is largely closed for near-term projects. Large-frame turbines from GE Vernova, Siemens Energy, and Mitsubishi Heavy Industries are booked through 2028 with a roughly 50/50 split between firm orders and reservations. Delivery horizons have extended further. According to CTVC analysis, gas turbine manufacturers are struggling to meet surging demand, with delivery backlogs beginning to stretch past 2029. GE Vernova, Siemens, and Mitsubishi Heavy Industries currently produce the majority of gas turbines globally.

The supply-demand math is stark. Global orders for natural gas turbines reached an estimated 100 GW by end of 2025, yet global manufacturing capacity is estimated at only 60–70 GW annually. New turbine orders are commanding price premiums of 10–20% above existing backlog levels, and near-term available capacity is increasingly spoken for by early-moving hyperscalers.

Even GE Vernova's aggressive ramp lands in future windows. GE Vernova expects to reach 20 GW of annualized output by Q3 2026, with further expansion to 24 GW targeted by 2028. Entering Q1, the company had roughly 10 GW of available 2029 production capacity; after a quarter in which customers pulled aggressively into 2030, it now has approximately 10 GW remaining across 2029 and 2030 combined. Operators placing orders today are looking at 2029–2030 delivery.

Gas turbine prices are projected to reach $600/kW by end of 2027, nearly tripling from 2019 levels, according to Wood Mackenzie. Operators who did not lock in slots in 2023–2024 are paying for that miss.

Logic ICs: The Constraint That Bridges Chip and Facility Timelines

Semiconductor lead times have normalized from the 52-week peaks of 2022, but unevenly. Lead times of 26–40 weeks are the norm for constrained categories, which represent structural tightness rather than temporary disruption. Memory is particularly acute. Lead times for DDR4 and DDR5 now exceed 30 to 40 weeks even as HBM demand consumes a growing share of wafer starts.

For infrastructure planners, the implication is clear: silicon ordered today resolves into late 2026 and 2027 deployment windows at best. When those windows collide with facilities delayed to 2028, operators hold committed silicon with no commissioned rack space. That mismatch has direct consequences for working capital and capex efficiency.

The aggregate picture is blunt. By 2027, 40% of AI data centers will be operationally constrained by electricity deficits, according to Gartner—a center-case estimate, not a fringe projection.

Who Wins, Who Loses, and by How Much

This pain is not distributed evenly. The companies positioned to benefit most are those with existing power contracts and cooling infrastructure—established hyperscalers—not new entrants. Operators who secured power purchase agreements and interconnection rights in 2022–2024 now hold a durable competitive moat that cannot be replicated quickly.

The 2026 capacity gap will push $150–200 billion in infrastructure spending into 2027–2028. That deferred spending will land in a compressed window, creating a secondary surge of equipment and construction labor demand that will itself hit supply constraints. The backlog does not disappear; it concentrates.

For second-tier operators and startups dependent on 2026–2027 capacity, options are limited. Behind-the-meter generation using reciprocating engines—which have shorter lead times than large-frame turbines—offers one path. Geographic diversification toward states with streamlined interconnection processes offers another. Louisiana has implemented a regulatory framework that streamlines approval timelines for power plants serving data centers, and Energy Transfer is expanding its Tiger Pipeline system in northern Louisiana to deliver approximately 250 million MMBtu daily beginning in early 2028 to power a 5 GW data center facility. Relocating the facility is increasingly rational versus waiting for constrained markets to clear.

Forecasters generally agree that the most acute supply-demand imbalance will persist through at least 2028 or 2029, when new generation and transmission capacity currently under development begins reaching commercial operation.

What to Watch

1. Interconnection queue reform. FERC ordered PJM to overhaul its transmission tariff framework. Track whether the new rules accelerate project timelines or simply reclassify the queue. The first quarterly PJM status reports under the new framework will signal which.

2. Hyperscaler guidance corrections. Operators carrying delayed capacity will revise public timeline guidance. Watch for language around "phased commissioning" and "2028 readiness" in earnings calls from second-tier cloud and colocation providers through Q3 and Q4 2026.

3. Turbine and transformer spot pricing. GE Vernova Q3 2026 earnings will show whether the 10–20% new-order price premium is holding or accelerating. A continued premium signals the supply wall is not clearing.

4. Behind-the-meter permitting velocity. Track state-level air permit approval timelines for reciprocating engine installations. This is the fastest workaround available to operators who missed turbine slots. Bottlenecks here close the last short-path option.

5. Power purchase agreement premiums. Operators securing multi-year PPAs in constrained markets are paying above-market rates. When those rates appear in REIT and infrastructure fund disclosures, they reveal the true cost of constraint—and which operators absorbed it early versus who remains exposed.