[Structural Analysis] AI is breaking the power grid (2026.8.6)

When "emergencies" become the new normal—the reality of the US power grid that AI has begun to change

In news surrounding AI, GPU shortages, semiconductor investments, and the construction of massive data centers are attracting attention.

However, what is truly constraining that competition?

I believe the answer is electricity.

Moreover, the problem is not as simple as "not enough power plants."

The emergency order issued by the US Department of Energy (DOE) this time revealed the reality that as we enter the AI era, the very systems supporting power infrastructure are beginning to change in nature.

This is also a sequel to the "AI and Power Demand" and "AI Power Geopolitics" topics I covered previously.

Last time, I organized the structure of how AI is driving up power demand.

This time, I will focus on how that increase in demand is beginning to change systems and markets.

—

The irony of 1935 "emergency systems" supporting the daily life of 2026

On July 26, 2026, the US Department of Energy (DOE) declared a statutory emergency for the Southwest Power Pool (SPP) wide-area power grid under Section 202(c) of the Federal Power Act.

The scope covers 17 states, a region supplying power to approximately 20 million people.

The reasons cited for the order included,

* A rapid surge in power demand
* Insufficient generation capacity
* Constraints on transmission facilities
* Maintaining grid stability

and so on.

At first glance, it looks like a measure against a heatwave.

However, if you view this order as a one-off response to abnormal weather, you will miss the essence.

What is truly important is the fact that

an emergency system created in 1935, assuming wars or major disasters, is being used repeatedly in 2026 to maintain the power grid during peacetime.

—

It is not that "emergencies" have increased

Section 202(c) is originally an emergency power intended to respond to "unforeseeable crises."

The operating rules established by the DOE in 1981 also clearly state that

it should not be used as a substitute for appropriate capital investment or power grid maintenance.

However, in 2026, it has been issued one after another for major power grids, such as:



* January: ERCOT* February: PJM
* March: Multiple times related to PJM
* June: PJM
* July: SPP
In other words,

"Abnormal weather has increased"

is less accurate than viewing it as

a state where normal operations cannot be maintained without emergency systems.

The system has not changed.

The role of the system is beginning to change.

—

Why the power grid cannot keep up

What is important here is the meaning of the term "power shortage."

It is not that power plants have suddenly disappeared.

It is not that the power grid has broken.

The problem is that

the design philosophy of the power grid did not anticipate the AI era.

For many years, the US power grid has been planned based on conventional demand such as:



* Population growth* Factory demand
* Economic cycles
Then, suddenly, AI data centers appeared, consuming massive amounts of power 24 hours a day, 365 days a year.

Moreover, the speed of that increase is outpacing the speed of power grid and power generation facility maintenance.

In other words, this crisis is

the time lag between design philosophy and reality,

hidden beneath the "trigger" of a heatwave.

—

The era where AI moves the "power grid," not the "city"

Virginia is symbolic.

It is one of the world's largest data center hubs, with over 600 facilities in operation.

It is said that data centers account for more than a quarter of the entire state's power consumption.

In the past, companies built factories in places where roads and ports were well-developed.

However, in the AI era,

companies are beginning to influence the design of power infrastructure itself.

It is no exaggeration to say that we are transitioning from an era where cities nurture companies to an era where companies influence the infrastructure investment of cities.

—

The market is already sending danger signals

These changes are also appearing in financial markets.

In the PJM capacity market, the cost of securing reserve capacity has risen significantly.

While there are differences in figures depending on the source,

the direction that "the market has begun to price in the risk of future supply shortages" is consistent.

This is an important point.

Grid strain is not a future story.

It has already begun to be a factor moving market prices.

—

Regulators have also admitted that "conventional methods cannot cope"

Even more interesting is the stance of the DOE itself.

In its Notice of Proposed Rulemaking (ANOPR) to the FERC, the DOE admits that

existing power grid plans cannot cope with future demand increases.

In other words,

the regulators themselves have officially recognized that

"the old ways cannot cope with the AI era."

That is why

Section 202(c), which is originally an exceptional measure, is being used repeatedly.

It is more appropriate to view this not as an abuse of emergency systems, but as a result of normal systems failing to keep up with reality.

—

AI investment cannot be stopped

So, should we curb data center construction?

The reality is not that simple.

The US currently positions its AI competition with China as a national strategy.

If investment is stopped here,

it could lead to the loss of competitiveness itself.

As a result,

the policy has become to maintain demand and supplement only the insufficient power with emergency measures.

In the short term, it is rational.

However, in the long term,

dependence on the power grid will continue to increase.

—

The "power security" competition happening around the world

This problem is not limited to the US.

Each country is facing the same challenge in different ways.

For Japan, the challenge is how to compensate for its dependence on energy imports.

China is securing power sources to support its AI industry while expanding renewable energy and maintaining coal-fired power.

In Europe, the shift away from dependence on Russian natural gas and the stabilization of electricity prices continue.

And in the US,

AI demand itself is straining the domestic power grid.

On the surface, these are separate pieces of news, but

there is a common theme.

That is,

it has become an era where "countries that can stably supply sufficient power" have a new competitive advantage.

—

The connection to the AI bubble

I have previously covered topics such as

* The SOX Index
* BIS AI risk warnings
* AI investment concentration
* Data center construction

continuously.

I believe this DOE order is an event that connects them all together.

If you organize the flow,

AI demand expansion

↓

GPU demand increase

↓

Data center construction

↓

Power demand surge

↓

Power grid strain

↓

Normalization of emergency systems

↓

Capacity market price increase

↓

Electricity rate increase

↓

AI company operating cost increase

↓

Re-evaluation of AI-related stocks

a chain becomes visible.

In other words, there is a possibility that the constraint on the AI market is beginning to shift from GPU shortages to power infrastructure shortages.

—

Conclusion: The Next AI Race Will Be a Competition of "Institutions" and "Power"

This DOE emergency order is not just news about "not having enough power during a heatwave."

Its essence lies in the fact that a system created for emergencies in 1935 is beginning to transform into a mechanism that supports peacetime power supply in the AI era.

This is not merely a matter of energy policy, but a shift in national strategy itself.

Moving forward, we may see countries proceed with the redesign of infrastructure to support the AI race, including the re-evaluation of nuclear power, the proliferation of Small Modular Reactors (SMRs), the extension of natural gas-fired power plants, large-scale investment in power grids, and even the reorganization of data center locations based on electricity prices.

In a previous article, I wrote that "in the AI era, the country that controls power will have the advantage."

This DOE emergency order can be said to be an event that takes that hypothesis one step further.

The AI race is entering a phase of competition in comprehensive strength, encompassing not just semiconductors, but also institutions, power grids, and power source development.

And that change has begun quietly, but surely.