19th-Century Technology Threatens America's AI Leadership

Despite many claims that America's power is declining, the country is still ahead of China in the race for dominance in artificial intelligence (AI).

Despite many claims that America's power is declining, the country is still ahead of China in the race for dominance in artificial intelligence (AI).

This achievement comes from Silicon Valley’s unique ability to connect scientists, entrepreneurs, and venture capital. But America’s future dominance in 21st-century AI depends on harnessing a 19th-century technology: electricity.

Huge energy demand from AI

Developing advanced AI systems requires enormous amounts of energy. Training AI models like GPT-4 uses a large number of specialized computer chips, consuming as much electricity as thousands of US households in a year.

But this energy demand faces practical limitations. The U.S. power grid, already hobbled by decades of disparate investment and overlapping regulations, is ill-prepared to handle the rapid growth of AI.

Currently, energy investors are waiting to develop 2.6 terawatts of new capacity from wind, solar and battery storage, but grid connection is facing many barriers. Since 2014, the total capacity waiting to be connected to the grid has increased eightfold, while the existing infrastructure can only meet a small fraction of that demand.

In Virginia, home to a large concentration of data centers, grid connection wait times can be as long as seven years. Some counties in the state have imposed limits on new data center construction. Such projects are held up by fragmented policies, lengthy approval processes, and inconsistent investment mechanisms.

China, meanwhile, has shown itself to be a master at rapidly deploying large infrastructure projects. The country can build everything from power plants to transmission lines to data centers at an astonishing rate. By the end of 2023, China had built 34 ultra-high voltage transmission lines, tens of thousands of kilometers in length, while the US has no similar projects.

China is catching up in both energy and technology.

While Chinese AI companies currently have difficulty accessing advanced US-made chips due to export controls, a “gray market” for these chips is emerging.

At the same time, China is also working to develop its own alternative chips. If successful, the ability to rapidly expand its energy infrastructure will allow these companies to operate massive AI training clusters.

Since 2000, China has expanded its power capacity by about sevenfold. In 2023 alone, it added 355 gigawatts of new capacity, compared with just 29 gigawatts in the United States. Over the past decade, China has built more than 30 new nuclear reactors, while the United States has added just three.

Recognizing the energy issue, major US tech companies have taken action themselves. Microsoft signed a deal to use electricity from a decommissioned nuclear reactor at its Three Mile Island facility. Google is investing $20 billion in renewable energy projects near new data centers.

However, to meet the huge demand of AI, many companies have had to use energy from coal and gas, negatively affecting carbon emission reduction goals and neighboring communities.

Solutions for the future

Clearly, the energy problem requires a comprehensive government approach. The U.S. power grid needs to be modernized and expanded with clean technologies like solar and nuclear. Policy and regulation also need to change to speed approvals for critical projects.

The US has already made semiconductors a key component of its AI economy through the CHIPS Act and export controls. Now, the country needs a similar strategy for its power infrastructure. A new agency, such as the Energy Accelerator, could be the solution to streamline processes and spur clean energy projects.

Electricity is not just a utility, it is the foundation of the digital age. If the US wants to maintain its leadership in AI, it must invest heavily in its energy system.

According to VNA

Source: https://doanhnghiepvn.vn/cong-nghe/cong-nghe-tu-the-ky-19-de-doa-vi-the-dan-dau-ai-cua-my/20250102093610638