What limits AI is not chips but electricity. Listen to what the former head of Google has to say.

Half a month ago, Eric Schmidt, the former CEO of Google, put forward a revolutionary view when he was a guest on a podcast.

"What we are seeing in AI now is just the tip of the iceberg. The capabilities of AI have not been overstated; instead, they have been severely underestimated. Most importantly, what limits AI is not chips but electricity."

This view quickly caught the attention and approval of Elon Musk. He commented on X, saying, "Eric is right. Solving the energy output problem is more difficult than chip production."

What's even more interesting is that his judgment was soon confirmed by the industry.

01

OpenAI recently announced a major cooperation agreement with Oracle. The two parties will expand the Stargate data center cluster in Abilene, Texas, USA. The new round of expansion will connect to a power capacity of 4.5GW, which is equivalent to the peak electricity consumption of a medium - sized province.

What does this mean? 4.5GW not only exceeds the total capacity of many national - level intelligent computing centers but is even equivalent to the output of five nuclear power plants. This is not just an expansion; it is a preview of the "electricity moat" in the AI era.

So what is Stargate? Literally translated, it means "Stargate". It sounds very sci - fi, like a time - travel tunnel in a Hollywood blockbuster. However, OpenAI has given it a real - world meaning - it is the lifeblood built in advance for the next generation of artificial intelligence.

Specifically, Stargate is the "AI power plant + super - brain headquarters" that OpenAI is building. It is not just an ordinary data center but the infrastructure base where future GPT models "eat, sleep, and work".

What does AI rely on to survive? The answer is clear: it relies on a continuous supply of electricity. That is to say, the end - point of AI is energy, and the starting - point of energy is electricity.

Take OpenAI's active layout this time as an example. This cooperation with Oracle is no longer a traditional cloud leasing but a deep - binding of resources: signing a 4.5GW power quota, choosing the ERCOT area of the independent power grid system in Texas, and relying on an energy structure that combines wind power, solar power, and natural gas.

In addition to its multi - point layout in the United States, OpenAI will also build similar bases in energy - rich areas such as Michigan, Wisconsin, and New Mexico in the future, and may even replicate the Stargate concept in Europe (such as "Stargate Norway").

This also means that OpenAI is transforming from an AI model company into a power technology giant. It not only provides "power support" for GPT but may soon become a power generator on the monthly electricity bills of American citizens.

02

On the other hand, against the backdrop of China's "Eastern Data and Western Computing" strategy, as early as 2020, the National Development and Reform Commission included intelligent computing centers in the scope of new infrastructure. Technology giants have also actively started to layout their own intelligent computing centers to prevent possible "energy bottlenecks" in the future.

For example, SenseTime's SenseTime Ark Intelligent Computing Center in Lingang, Shanghai, is Asia's first AI cluster with over ten thousand graphics processing units. Alibaba Cloud has built the Feitian Intelligent Computing Center cluster to serve the training of Alibaba's Qwen series of large models. Huawei, which has always advocated "computing power as a service", has already laid out multiple intelligent computing center nodes across the country. Tencent and Baidu have also been gradually promoting the self - construction or co - construction of intelligent computing resource pools in recent years to form a regional intelligent computing node network.

Does this mean that China is leading the United States in terms of energy supply? At least Eric Schmidt holds this view.

Eric Schmidt led Google for ten years, and his understanding of AI far exceeds that of ordinary people.

Key Point One

How much electrical energy does the development of AI in the United States actually need?

Eric Schmidt believes that AI can be regarded as a learning machine. In some businesses with network effects, this machine will accelerate everything until it reaches its natural limit. However, this limit is not the chips as we thought but electricity.

The expected additional demand for the current AI revolution in the United States is the power equivalent of 92 large - scale nuclear power plants. However, there are basically no new nuclear power plants under construction at present, and only two have been built in the past 30 years. So, nuclear energy, whether fission or fusion, may not be able to meet the needs of AI development at present.

Key Point Two

What is the energy gap between China and the United States?

Eric specifically pointed out that China has a large amount of electricity, and the speed of China's energy expansion is two to three times that of the United States. In contrast, the energy production in the United States has stagnated for a long time. Due to energy - saving and other factors, the electricity demand once declined, which even dragged down the overall energy demand. However, data centers have reignited the energy industry. The prospect brought by a super - brain with a 1 - gigawatt data center is very real.

Regarding this point, some American netizens have expressed their concerns. The phrase "The United States must accelerate" has repeatedly appeared in the comments of many American netizens. "Look at China's efforts in large - scale electrification. It's almost epic."

Actually, the biggest concern revealed by Eric Schmidt in the interview is not just these "technical issues" such as chips, electricity, and energy, but a deeper proposition - how will the path of AI completely change human society? For example, in an environment without proper supervision and full of misinformation, when the system has freedom of speech, and bad people use misinformation for profit, what will democracy become?

In the next issue, we will continue to delve into this topic.

This article is from the WeChat public account "Business Observation" (ID: zgkejisir). The author is Yanhuo. It is published by 36Kr with permission.