Even if the AI Bubble Bursts, the Competition for Electricity Won’t End

The AI bubble may burst. However, electricity costs will not decrease.

This is the essence of what is happening now. Regardless of stock prices, the construction rush for AI data centers continues unabated. Microsoft, Google, and Amazon have collectively invested over $200 billion in data centers just in 2024. The investment is expected to increase further in 2025. This investment is not because “AI is profitable,” but rather because it is a power struggle where “those who control the infrastructure win.”

So, who will lose out on this electricity? Local small and medium enterprises.

Understanding the “AI’s Electricity Consumption”

First, let’s grasp the scale of the situation.

According to the IEA (International Energy Agency) report for 2024, the global electricity consumption of data centers was about 460 TWh as of 2022. This is equivalent to half of Japan’s total annual electricity consumption (approximately 900 TWh). With the spread of AI, it is projected that data center electricity consumption could exceed 1,000 TWh by 2026. This represents more than a doubling in just four years.

It is said that a single query to ChatGPT requires about ten times the electricity of a Google search. While a Google search consumes about 0.3 Wh, ChatGPT consumes about 3 Wh. This may seem like a small difference, but when used billions of times a day worldwide, the implications change significantly.

NVIDIA’s AI training GPU, the “H100,” consumes up to 700W per unit. When thousands of these are used to train AI models, the electricity cost for a single training session can reach millions of dollars. The estimated electricity cost for training a GPT-4 class model is said to be between $5 million and $10 million.

This electricity must come from somewhere.

The Competition for Electricity Has Begun

In the United States, this is already a reality. In Virginia’s data center hub, electricity demand has outstripped supply, leading power companies to issue “connection waitlists” for new data center constructions. Dominion Energy, a major power company in the state, has reported that the waitlist for new connections could extend for several years by 2024.

The impact of this situation is also affecting general businesses. When electricity supply and demand become tight, electricity prices rise. In Japan, many small and medium enterprises have experienced electricity costs increasing 1.5 to 2 times since the invasion of Ukraine in 2022. Many business owners remember that pain. The increased electricity demand from AI data centers will gradually exert similar upward pressure on prices.

Japan is not exempt from this issue. There are ongoing plans to build data centers in Hokkaido and Chiba, raising concerns about the impact on regional power infrastructure. According to estimates from the Ministry of Economy, Trade and Industry, domestic data center electricity consumption could increase 2 to 3 times by 2030.

For a small factory with a monthly electricity bill of 500,000 yen, an increase to 600,000 or 700,000 yen would have a significant impact. This translates to an annual cost increase of 1.2 to 2.4 million yen, equivalent to the salary of one employee.

Data Centers Are Creating a “Battery Market”

This electricity issue is explosively growing another market: batteries.

Data centers cannot afford to stop even for a second. However, as the proportion of renewable energy increases, supply becomes unstable. To resolve this contradiction, the demand for large-scale battery energy storage systems (BESS) is surging.

According to BloombergNEF’s forecasts, the global stationary battery market is expected to expand its annual installation volume to more than five times its current level by 2030. One of the growth drivers for this market is AI data centers. Google has begun to implement large-scale batteries in its data centers, and Microsoft is showing similar trends.

Herein lies a paradoxical opportunity for small and medium enterprises. As mass production of batteries progresses, the unit cost of batteries will decrease. The cost of lithium-ion batteries per kWh has dropped from about $1,200 in 2010 to around $140 in 2023. The demand from data centers could further accelerate this price decline.

In other words, “electricity prices will rise” and “battery prices will decrease” can happen simultaneously. For small and medium enterprises that self-consume energy with solar panels and batteries, this presents an opportunity to reduce their purchases from power companies. As electricity prices rise, the ROI for self-generation and storage improves. Whether they can seize this structural change will be crucial.

Putting Servers in Space?—Grand, but Not Relevant Now

Elon Musk has spoken about the concept of “orbital data centers.” The rationale is that in space, cooling costs are nearly zero, and solar power can be generated 24 hours a day.

To be honest, this is not relevant for today’s small and medium enterprises.

The cost to launch 1 kg into orbit with SpaceX’s Falcon 9 is about $2,700. A server rack weighs about 30 to 50 kg. Just the launch would cost between $80,000 and $130,000 per unit. If something breaks, it cannot be repaired. There are also latency issues; while ground data centers respond in 1 millisecond, satellite connections take 20 to 40 milliseconds, making them unsuitable for real-time processing.

The situation may change in 10 or 20 years. However, this is not something to factor into current business decisions. “Servers in space” is an interesting sci-fi concept, but what small and medium enterprise owners should be concerned about is “next month’s electricity bill.”

What Should Small and Medium Enterprises Do?

So, what should be done? Here are three key points.

1. Assume “structural increases in electricity costs”

The increased electricity demand from AI data centers is not a temporary boom. It is a structural trend. Business plans should be built on the premise that electricity costs will rise in the medium to long term. If you calculate “current electricity costs” as fixed expenses, profits could disappear in a few years.

2. Consider securing self-consumption power

The installation costs for solar panels and batteries are decreasing year by year. The cost of installing a 50 kW solar power system is now less than half of what it was ten years ago. When combined with batteries, it is possible to cover 50 to 70% of daytime electricity consumption through self-consumption. While the initial investment may be several million yen, in many regions, it can be recouped within 5 to 7 years considering the rise in electricity costs. Subsidies can also be utilized.

3. Shift to being a “user” of AI. Do not participate in the infrastructure competition

Small and medium enterprises do not need to gather GPUs and train AI models themselves. That is the job of large corporations. What small and medium enterprises should do is to “use” the AI models created at great cost by large companies at a lower price. The API usage fee for OpenAI is just a few cents per 1,000 tokens. For a few thousand to tens of thousands of yen per month, businesses that once cost hundreds of thousands of yen can be automated.

As the costs for those who own the infrastructure rise, the cost-performance for users stands out. This asymmetry is the weapon of small and medium enterprises.

The Rampant AI Infrastructure Will Affect Everyone in the Form of “Electricity Costs”

Whether the AI bubble will burst is uncertain. However, one thing is certain: the expansion of AI infrastructure will fundamentally change the electricity market, and its impact will reach all businesses in the most straightforward form—electricity costs.

Large corporations have the capacity to own their power plants, enter into long-term contracts with power companies, and implement battery storage. Small and medium enterprises do not have that capacity. Therefore, recognizing the reality that “electricity costs will structurally rise” and taking action quickly will be a matter of survival.

When discussing AI, some business owners say, “It doesn’t concern us.” They can choose whether or not to adopt AI. However, they cannot choose to avoid rising electricity costs. Even without using AI, the costs associated with AI will come around.

How will they respond to this reality?