The artificial intelligence boom is usually framed as a race for faster processors, larger models and increasingly sophisticated software.
But another constraint is becoming impossible to ignore: electricity infrastructure.
Data centers are expanding at a pace that existing power systems were not designed to accommodate. The result is a growing mismatch between how quickly technology companies can construct AI facilities and how quickly utilities can generate, transmit and deliver the electricity those facilities require.
For the next phase of AI infrastructure, access to power could become just as important as access to chips.
The Grid Was Built for a Different Era
The modern electricity grid was largely designed around gradual and relatively predictable changes in demand.
AI data centers are disrupting that model.
These facilities can require enormous amounts of electricity concentrated in a single location, creating sudden load growth that can overwhelm local transmission and distribution capacity.
U.S. data centers consumed approximately 176 terawatt-hours of electricity in 2023, according to industry estimates. Demand is expected to continue increasing as AI workloads expand, adding pressure to infrastructure that already faces long planning and construction cycles.
The fundamental problem is straightforward:
Data centers can be built in years. Power infrastructure can take much longer.
That difference is becoming a major constraint on AI deployment.
Interconnection Queues Are Becoming a Major Bottleneck
Before new generation or storage projects can connect to the electricity grid, they typically enter an interconnection process.
Projects are studied to determine whether they can safely connect to existing infrastructure and what upgrades may be required.
The system works reasonably well when demand and generation change gradually.
AI infrastructure is creating a very different environment.
Large data centers can suddenly request substantial amounts of electricity, forcing utilities to evaluate new substations, transmission capacity and other infrastructure upgrades.
The resulting backlog can affect entire regions.
More than 1,300 GW of generation capacity has been reported as sitting in U.S. interconnection queues, illustrating the scale of the challenge.
The problem isn’t simply the amount of electricity available.
It is whether that electricity can reach the location where it is needed, when it is needed.
The Timeline Mismatch Is Getting Worse
The biggest challenge may be the difference between technology development cycles and infrastructure development cycles.
AI companies can announce a new data center, secure financing and begin construction within a relatively short period.
Transmission projects operate on a completely different timetable.
New high-voltage lines can require years of engineering, environmental assessments, land acquisition, permitting and construction.
Environmental reviews alone can take several years, while some transmission projects require approvals from multiple government agencies and jurisdictions.
This creates a structural mismatch:
AI infrastructure moves at software speed. The electricity grid moves at infrastructure speed.
Unless that gap narrows, available power could increasingly determine where AI companies can build.
Permitting Creates Another Layer of Friction
Building transmission infrastructure isn’t simply an engineering challenge.
It is also a regulatory one.
A major transmission project can require approvals from local governments, state regulators and federal agencies. Projects crossing state boundaries can face additional regulatory complexity because multiple jurisdictions may have different requirements.
A single objection can delay an entire project.
This becomes particularly problematic when a data center is ready for construction but the transmission infrastructure required to power it remains years away.
Federal reforms, including long-term regional transmission planning initiatives, are intended to address some of these problems. But regulatory changes take time to translate into physical infrastructure.
For AI developers working on aggressive deployment schedules, that delay can be costly.
Three Approaches to Solving the Power Constraint
The power bottleneck is creating opportunities for companies that can help utilities and data center developers extract more capacity from existing infrastructure, accelerate permitting or integrate energy planning into the earliest stages of development.
1. TRC: Making Transmission Permitting More Efficient
One solution is improving how transmission projects are planned and permitted.
Large infrastructure projects often require environmental assessments, route analysis, community engagement and multiple regulatory filings.
TRC’s work on the Poseidon Project demonstrates how coordinated program management can help navigate these requirements.
The project involves an approximately 80-mile high-voltage direct-current transmission line requiring extensive environmental analysis and regulatory approvals.
Work involving wetlands, protected species, cultural resources and alternative routes must be coordinated alongside engineering and permitting requirements.
The lesson is important for the AI infrastructure market:
Transmission planning cannot be treated as an issue that comes after a data center has already been designed.
It needs to be incorporated into the project from the beginning.
2. GridCARE and Portland General Electric: Finding Capacity That Already Exists
Another approach is to make better use of infrastructure that already exists.
Instead of immediately building new transmission capacity, utilities can use advanced modeling and forecasting to identify periods and locations where additional electricity demand can be accommodated.
Portland General Electric worked with GridCARE to identify more than 80 MW of incremental capacity on the existing grid for data centers in Hillsboro, Oregon.
This approach illustrates an important alternative to traditional infrastructure expansion.
Rather than asking:
“How quickly can we build more capacity?”
Utilities can also ask:
“How much capacity are we failing to use efficiently today?”
Advanced forecasting, real-time grid information and AI-driven demand modeling could help utilities identify that unused capacity faster.
3. Jacobs: Making Power Part of Data Center Site Selection
A third approach is to rethink how data centers are planned from the outset.
Traditionally, companies may select a site based on factors such as land availability, connectivity, tax incentives and proximity to customers before addressing energy infrastructure in detail.
For AI data centers, that sequence may no longer work.
A location with inexpensive land is of limited value if connecting it to the grid takes five years.
Companies therefore increasingly need to evaluate:
- Available generation capacity
- Transmission infrastructure
- Interconnection timelines
- Energy procurement options
- Regulatory requirements
- Future power demand
- Expansion potential
Jacobs promotes an integrated approach that combines site selection, energy strategy and engineering.
The advantage is straightforward: energy constraints can be identified before they become construction delays.
AI Infrastructure Is Becoming an Energy Infrastructure Problem
The AI industry has spent enormous amounts of capital expanding computing capacity.
Chips are becoming faster.
Models are becoming larger.
Data centers are becoming more sophisticated.
But none of those investments matter if the electricity required to operate them cannot be delivered.
That changes the competitive landscape.
The next generation of AI infrastructure may increasingly be built around access to reliable electricity rather than simply access to available land or cloud infrastructure.
Regions with abundant generation, strong transmission networks and faster permitting processes could become disproportionately attractive to AI companies.
Meanwhile, locations with constrained grids may struggle to compete regardless of how attractive they are on other dimensions.
The Grid Could Become AI’s New Strategic Bottleneck
The AI boom has created an unusual infrastructure problem.
Technology companies can innovate rapidly, while the infrastructure supporting them operates on much longer cycles.
That gap cannot be solved by a single technology.
It will require a combination of:
Grid expansion, regulatory reform, smarter forecasting, better transmission planning, energy procurement and more intelligent data center site selection.
The companies capable of connecting those pieces could become just as important to the AI economy as the companies building processors and models.
The biggest AI constraint of the next decade may therefore not be computational power.
It may be electrical power delivered at the right place, at the right time, and at the right scale.
AI can move at software speed. Its energy infrastructure cannot.
The winners of the next phase of the AI boom will increasingly be the companies that understand how to bridge that gap.
