The Grid Is the New AI Supply Chain

Why substations, transformers and transmission - not electricity generation alone are determining where AI infrastructure gets built.

The AI power conversation usually starts with a very large number.

Three hundred megawatts.

One gigawatt.

Five gigawatts.

The numbers make the headlines because they communicate the extraordinary scale of the data centers being planned. But they leave out the more difficult part of the story.

Where will those megawatts come from and how will they reach the servers?

A region may generate plenty of electricity while lacking the transmission capacity, substations, transformers and local distribution infrastructure required to deliver it to a particular site.

A utility may have enough energy in its long-term plan but no way to serve a new campus by the developer’s target date. New generation may be proposed, financed and sitting in an interconnection queue without permission to connect to the grid.

The electricity exists in the forecast.

It does not exist at the data-center fence.

That distinction is becoming one of the most important competitive factors in the Intelligence Economy.

AI does not need theoretical power.

It needs deliverable power.

The Short Answer

The power grid is becoming a critical part of the AI supply chain because electricity must move through generation facilities, transmission lines, substations, transformers, switchgear and campus distribution systems before it can reach an AI processor.

Having sufficient electricity in a region does not guarantee that a data center can access it.

AI infrastructure requires power in the right amount, at the right location, with the required reliability and on a timeline that matches construction. If any part of the delivery system is unavailable, the project can be delayed even when land, capital and computing equipment are ready.

That is why substations, transformers and transmission capacity are becoming as strategically important as chips.

A power plant can produce the electricity.

The grid still has to deliver it.

The Signal

AI electricity demand is moving faster than the physical systems supporting it.

The International Energy Agency reported that global data-center electricity demand grew 17% in 2025. Electricity consumption from AI-focused data centers increased 50% during the same year.

The IEA now expects total data-center electricity consumption to roughly double from 485 terawatt-hours in 2025 to 950 terawatt-hours in 2030. Electricity use from AI-focused facilities is expected to triple. International Energy Agency

This is not simply more demand.

It is more demand arriving in larger blocks, at fewer locations and on compressed development timelines.

A typical utility is accustomed to adding customers gradually. A new neighborhood, manufacturing plant or commercial district may increase demand over several years.

An AI campus can request hundreds of megawatts at once.

Some proposed developments are now measured in gigawatts—the scale of an entire power plant or midsized city.

The grid was not designed around customers appearing at that speed or scale.

What Is the Difference Between Available and Deliverable Power?

Available power describes electricity that may exist within a utility territory, regional market or future generation plan.

Deliverable power describes electricity that can physically reach a specific location under defined operating conditions.

That Difference Depends On Several Questions:

• Is enough generation operating when the data center needs it?

• Can the transmission system move the electricity into the region?

• Does the local substation have available capacity?

• Are transformers and switchgear installed?

• Can the grid continue serving the facility if another component fails?

• What upgrades are required?

• Who will pay for those upgrades?

• When can the service actually begin?

A market may appear power-rich when viewed at the state level and power-constrained when examined at the property level.

This is the Deliverable Power Principle:

Generation capacity does not create AI readiness unless the grid can deliver it to the site on the required timeline.

How Does Electricity Reach an AI Data-Center?

The journey begins long before power enters the building.

1. Generation

Electricity is produced by power plants, including natural-gas facilities, nuclear plants, solar farms, wind farms, hydroelectric systems and other energy resources.

The amount of installed generation receives most of the attention.

But installed capacity is not always available capacity. Power plants experience maintenance, fuel constraints and operating limits. Solar and wind production varies with conditions. Some resources can produce continuously, while others require storage or backup generation.

The grid must balance all of these resources against demand in real time.

2. High-Voltage Transmission

Electricity travels long distances through high-voltage transmission lines.

These lines connect power plants to population centers, industrial markets and regional substations. They also allow electricity to move between utility territories and organized power markets.

Transmission capacity determines whether power available in one location can reach demand somewhere else.

A region may have new generation under development, but if the transmission system cannot carry the additional electricity, that generation does not solve the data center’s problem.

3. Utility Substations

Substations connect different parts of the grid and transform electricity from one voltage level to another.

For a large AI campus, the local substation may need major expansion—or the project may require an entirely new dedicated substation.

That can involve new land, engineering studies, high-voltage equipment, transformers, protection systems and utility approvals.

The substation is where regional power availability becomes local power delivery.

4. Campus Electrical Infrastructure

Once electricity reaches the property, it must move through the data center’s internal power systems.

These Can Include:

• High-voltage transformers

• Medium- and low-voltage switchgear

• Busways and power-distribution units

• Uninterruptible power supplies

• Battery systems

• Backup generators

• Control and monitoring equipment

• Redundant electrical pathways

Each component must be engineered for the density and reliability of the computing environment.

A data center does not simply connect to electricity.

It builds a private grid inside the campus.

The Substation Bottleneck

Substations rarely appear in AI product announcements.

They may matter more than the product.

A utility substation has a limited amount of capacity. When that capacity has already been allocated to existing customers or previously approved projects, a new data center may require additional transformers, new transmission connections or a separate facility.

Those upgrades take time.

A substation project may require site control, environmental review, utility engineering, equipment procurement, construction and testing. The work must also be coordinated with the larger transmission system.

This creates a hidden competition among development projects.

Two data centers may be evaluating sites within the same utility territory. Both may see the same regional generation resources. But the project positioned near a substation with expandable capacity may have a significant advantage over the project requiring several years of grid construction.

The valuable asset is not simply land near electricity.

It is land near expandable electrical infrastructure.

Why Have Transformers Become AI Equipment?

Transformers change electricity from one voltage level to another.

They are essential at power plants, substations, industrial facilities and data-center campuses. Without them, electricity cannot move safely and efficiently through the different layers of the system.

They are also difficult to manufacture.

Large power transformers can weigh hundreds of tons, cost millions of dollars and be designed for the specifications of a particular project. They are not products that can always be pulled from a warehouse when demand suddenly increases.

A U.S. Department of Energy supply-chain assessment reported average lead times ranging from 51 weeks for distribution transformers to 137 weeks for power transformers. It also found transformer prices had increased between 37% and 80%, with manufacturing capacity, labor shortages, electrical steel and copper among the supply-chain constraints. U.S. Department of Energy

That changes how AI projects must be planned.

A developer can secure a data-center tenant and still wait years for the electrical equipment required to serve it.

Transformers are therefore no longer ordinary utility components.

They are schedule-defining AI infrastructure.

Everyone is tracking GPU availability.

Infrastructure leaders should also be tracking transformer production slots.

Interconnection Queues Are Changing the Map

New power plants cannot simply begin sending electricity onto the grid.

They must apply for interconnection, undergo technical studies and determine which transmission upgrades are required before connecting. The process is designed to protect reliability, but the volume of proposed generation has created a large national backlog.

At the end of 2025, more than 2,060 gigawatts of generation and storage capacity were actively seeking connection to the U.S. grid, according to Lawrence Berkeley National Laboratory.

That included approximately 1,312 gigawatts of generation and 749 gigawatts of storage spread across roughly 8,200 projects. The median timeline from an interconnection request to commercial operation exceeded five years for projects completed in 2025. Lawrence Berkeley National Laboratory

Most of that proposed capacity will never become operational. But even the projects that do move forward can face lengthy studies, transmission upgrades and equipment requirements.

Data centers follow a different large-load connection process. They are customers rather than generators.

But the systems are connected.

If a utility needs new generation to serve data-center demand, the generation may be delayed in an interconnection queue. If the transmission network requires upgrades, both the power plant and the data center can be affected.

A data center may be ready to consume electricity before the grid is ready to produce and deliver it.

Capacity Has a Clock

Power availability is usually discussed as a quantity.

For AI infrastructure, it must also be discussed as a date.

A utility might be able to serve 300 megawatts in 2031. That does not help a developer planning to open the first data hall in 2028.

The timeline can determine whether a project moves forward, changes locations or develops its own power solution.

This Is Creating a New Hierarchy of Site Value:

• Power available today

• Power available after limited upgrades

• Power available after a new substation

• Power available after transmission expansion

• Power dependent on future generation

• Power that exists only in a long-term planning document

Each category carries a different level of cost, risk and credibility.

The further the power sits from the present, the more assumptions must hold before the data center can operate.

A megawatt without a delivery date is not a development asset.

It is a possibility.

Why Are Data Centers Moving Behind the Meter?

Behind-the-meter power is electricity generated on or near a customer’s property and delivered without relying entirely on the traditional utility delivery system.

For Data Centers, this Can Include:

• On-site natural-gas generation

• Fuel cells

• Solar and battery systems

• Microgrids

• Dedicated energy campuses

• Co-located generation

• Long-duration energy storage

The appeal is straightforward.

If the grid cannot provide enough power on the required timeline, the developer attempts to bring the power directly to the project.

Behind-the-meter systems can also improve resilience, provide backup capacity and allow a campus to operate more independently during grid disruptions.

But they are not a universal shortcut.

The IEA estimates that approximately 15 to 27 gigawatts of on-site natural-gas generation could be serving data centers by 2030, primarily in the United States. It also found that providing reliable on-site electricity for variable data-center loads may require developers to build 30% to 70% more generating capacity than the facility’s expected demand. International Energy Agency

On-site generation still requires turbines, fuel delivery, permits, air-quality approvals, financing, maintenance and backup capacity. Gas-turbine supply is constrained as well.

Leaving the grid does not eliminate the infrastructure problem.

It changes the infrastructure that must be built.

Hyperscalers Are Becoming Energy Developers

The largest technology companies are no longer waiting for existing electricity markets to solve the problem.

They are signing long-term power agreements, supporting new generation technologies and providing the commercial commitments needed to finance energy projects.

In January 2026, Meta announced nuclear agreements intended to unlock up to 6.6 gigawatts of energy capacity. The portfolio includes agreements supporting new projects from TerraPower and Oklo, as well as existing and expanded generation from Vistra facilities.

Meta’s TerraPower agreement covers two initial units capable of providing up to 690 megawatts, with rights connected to additional units. Its Oklo partnership could support up to 1.2 gigawatts of new baseload power entering the PJM market. Meta

These agreements illustrate a larger shift.

Hyperscalers are not simply purchasing electricity after it is generated. They are helping create the financial certainty required to develop the generation itself.

The customer is becoming part of the power supply chain.

The Market Is Responding

When demand moves toward electrical infrastructure, the companies producing that infrastructure feel it first.

Power-management company Eaton reported that data centers remained a key growth driver during the second quarter of 2026. Its Electrical sector backlog increased 43% year over year, while sales across the company reached a record $8.5 billion. Eaton

Eaton’s quarterly investor materials reported that data-center orders across its Electrical businesses were approximately 85% higher than a year earlier, while related revenue increased approximately 65%. Eaton Q2 2026 presentation

Those numbers tell us something important.

AI demand is moving beyond chips and appearing in orders for switchgear, power distribution, cooling and electrical systems.

The same activity is increasing the strategic importance of companies such as GE Vernova, Siemens Energy, Hitachi Energy, ABB, Schneider Electric, Quanta Services, Powell Industries, Vertiv and Hubbell.

The AI supply chain is expanding.

It now includes the companies that build the grid.

Company Watch: Eaton

Everyone is watching the processor rack.

I’m watching what connects it to the electricity.

Eaton operates across several layers of the AI power system. Its products help utilities manage electricity, help data centers distribute power inside facilities and help operators protect equipment from interruptions.

That position gives Eaton exposure to both sides of the power challenge.

Utilities need to expand and modernize the grid. Data-center developers need more sophisticated power architecture inside the campus. Higher rack densities also increase demand for cooling, electrical protection and power-management technology.

Eaton’s growth does not depend on one hyperscaler or one generation technology.

It depends on the broader need to move, control and protect electricity.

That makes it an important signal company for the Power pillar.

If Eaton’s data-center orders, electrical backlog and manufacturing investment continue rising, the physical AI buildout is still accelerating beneath the software headlines.

Ticker: ETN

Power Readiness Is a Supply-Chain Question

A community cannot establish AI readiness by pointing to a nearby power plant.

It must understand the entire delivery chain.

That Includes:

• Available generation

• Transmission import capacity

• Substation capacity

• Transformer availability

• Utility construction timelines

• Redundancy requirements

• Equipment lead times

• Interconnection studies

• Rate structures

• Backup and on-site generation options

• Skilled electrical labor

• The capital required for upgrades

Any missing component can delay the project.

This is why utility engagement must begin early in the site-selection process. Developers need more than an estimate of future electricity supply. They need a credible path from the generation resource to the campus.

The winning sites will not always offer the cheapest electricity.

They may offer the clearest path to energization.

Power Across the Eight Pillars

Power does not operate independently from the other seven pillars.

Transmission lines and substations require land. New facilities need permits and supportive policy. Grid construction depends on engineers, electricians and skilled trades. Cooling systems connect power decisions to water and sustainability.

Capital determines whether generation and grid upgrades can be financed. Connectivity influences where compute can be distributed when power availability differs between regions.

The relationship also works in reverse.

A site with scalable power becomes more valuable. A region with available electricity attracts fiber investment, construction activity and infrastructure capital.

Power creates economic gravity.

But only when it can be delivered.

The Hot Take

The AI power race may not be won by the region with the most electricity.

It may be won by the region with the fewest steps between electricity and the server.

Generation announcements sound impressive. A new five-gigawatt project pipeline can signal long-term opportunity.

But developers make decisions around timelines.

A smaller market with expandable substations, clear permitting and equipment already ordered may outperform a larger market where power remains trapped behind transmission constraints and interconnection studies.

The next competitive advantage is not simply energy abundance.

It is delivery certainty.

What Leaders Should Start Asking

Communities, Utilities, Developers and Investors Should Ask:

• How much power is available at the specific site?

• Is that capacity firm or dependent on future projects?

• What is the earliest realistic energization date?

• Which substation would serve the campus?

• How much expandable capacity does that substation have?

• Is a new substation required?

• Are transmission upgrades necessary?

• Who will pay for those upgrades?

• Have transformer and switchgear production slots been secured?

• What equipment has the longest lead time?

• Can the facility be energized in phases?

• What generation resources will serve the load?

• Are those resources operating, under construction or still proposed?

• What redundancy standard can the utility provide?

• Could behind-the-meter generation accelerate the project?

• What permits would on-site generation require?

• Can the campus support the grid during periods of stress?

• How will costs be allocated without shifting excessive risk to existing customers?

These questions move the conversation beyond power marketing.

They reveal power readiness.

The Golden Nugget

Power on a map is not the same as power a data center can use.

The real asset is deliverable capacity - electricity that can reach the site at the required scale, reliability and date.

That is what transforms land into Powered Land.

Frequently Asked Questions

Why Do AI Data-Centers Require So Much Electricity?

AI data centers use large clusters of processors to train models and deliver AI services. Those processors require electricity directly and also produce heat, creating additional demand for cooling, networking and power-distribution systems.

What Part of the Grid Connects a Data-Center to Electricity?

Large data centers may connect to the high-voltage transmission or utility distribution system through dedicated substations. Transformers reduce voltage before electricity moves through the campus’s switchgear, backup systems and internal distribution equipment.

Why Can’t a Data-Center Connect When a Region Has Sufficient Generation?

Generation may exist without enough transmission, substation or transformer capacity to deliver the electricity to the proposed site. Grid reliability studies and infrastructure upgrades may also be required before service begins.

How Long Does It Take to Build a Data-Center Substation?

The timeline varies by project, utility, equipment availability and permitting requirements. Large substations can take several years when new transmission connections, transformers and site development are required.

What Is Behind-the-Meter Power?

Behind-the-meter power is electricity produced on or near the customer’s property and delivered without relying entirely on the traditional utility grid. Examples include on-site natural-gas generation, fuel cells, solar, batteries and microgrids.

Are Transformer Shortages Affecting Data Centers?

Yes. Long manufacturing lead times for power transformers and other electrical equipment can delay utility upgrades and data-center construction. Developers increasingly need to reserve equipment and manufacturing capacity much earlier in the planning process.

Which Companies Benefit From AI Grid Investment?

Potential beneficiaries include electrical-equipment manufacturers, turbine producers, power-management companies, engineering firms and grid-construction contractors. Examples include Eaton, GE Vernova, Siemens Energy, Hitachi Energy, ABB, Schneider Electric, Quanta Services, Vertiv, Hubbell and Powell Industries.

The IOI Take

The market continues to describe AI power as a generation problem.

It is also a delivery problem.

The Intelligence Economy requires a new power supply chain extending from fuel and generation through transmission, substations, transformers, switchgear and campus distribution.

Every layer matters.

This changes how regions should evaluate their competitiveness. Statewide generation numbers are useful, but they do not reveal which properties can be energized. Utility territories must be examined at the substation level. Equipment availability must be treated as part of site readiness. Power timelines must be verified before development announcements are treated as capacity.

It also changes what investors should watch.

The AI opportunity is not confined to utilities and power producers. It extends to the companies manufacturing transformers, switchgear, turbines, cooling systems and grid components—and to the engineering and construction firms installing them.

The processor may be the most visible part of the AI factory.

The grid is the supply chain that makes the factory possible.

One Last Thought

The AI industry is learning a lesson every physical economy eventually learns.

Production depends on delivery.

A power plant can generate the electricity. A utility can forecast the demand. A developer can control the land. A hyperscaler can order the processors.

But none of it matters until the electricity reaches the building.

The chip may create intelligence.

The grid determines when that intelligence can come online.

Workforce

Policy

Capital

Sustainability