Some reasons to be skeptical of Chinese AI hype

Over the weekend, a Chinese artificial intelligence company announced that its open-source AI model, DeepSeek-R1, required only a small fraction of the investment that U.S. tech companies are pouring into their AI models. The stock market plunged, but there are a few reasons why it would be premature to revise U.S. power demand forecasts downward.

AI models use electricity in three main ways: training, inference, and infrastructure support. Training is the most energy-intensive, where the model learns from vast datasets. Inference occurs when an AI model processes user inputs and generates responses. Electricity is also used in running data center infrastructure, both in direct computation and indirect needs like cooling, storage, and networking.

DeepSeek claims its model required less than $6 million in computer chips and electricity to train. For context, OpenAI’s ChatGPT-4 model cost $74.8 million to train. The news plunged stocks of computer chip makers like Nvidia (down 17 percent on January 27) as well as electricity generation companies like Constellation Energy (down 20 percent), which will be restarting the Three Mile Island Nuclear plant, and Siemens Energy (down 20 percent) and GE Vernova (down 21 percent).

However, it’s unlikely that this figure is the whole truth. Some analysts note that it seems that DeepSeek is excluding the “costs associated with prior research,” which would lower cost claims compared with U.S. companies. In addition, OpenAI is actively investigating whether data output from its technology was used to train DeepSeek in late 2024.

Having U.S. companies do the heavy lifting would obviously cut costs for the Chinese AI model. The Verge reports:

President Donald Trump’s artificial intelligence czar David Sacks said “it is possible” that IP theft had occurred. “There’s substantial evidence that what DeepSeek did here is they distilled knowledge out of OpenAI models and I don’t think OpenAI is very happy about this,” Sacks told Fox News on Tuesday.

“We know PRC (China) based companies — and others — are constantly trying to distill the models of leading US AI companies,” OpenAI said in a statement to Bloomberg. “As the leading builder of AI, we engage in countermeasures to protect our IP, including a careful process for which frontier capabilities to include in released models, and believe as we go forward that it is critically important that we are working closely with the US government to best protect the most capable models from efforts by adversaries and competitors to take US technology.”

There is also privacy and security concerns about using Chinese AI, including DeepSeek. DeepSeek records keystroke data and stores it in Chinese servers indefinitely, which means the model “might be significantly cheaper to run, but your privacy and security are the actual cost.” It also seems to censor itself when it comes to criticizing the Chinese government.

The U.S. Navy directed its members not to use DeepSeek “in any capacity” for work-related or personal tasks due to “potential security and ethical concerns associated with the model’s origin and usage.” Privacy and security are two reasons why U.S.-made AI might continue to be preferred, especially with its higher electricity usage.

Even assuming that AI does become more efficient in terms of computing power and electricity use, it’s likely that it will simply enable a Jevons-style paradox, where the cost of using AI drops but demand increases enough that total resource consumption rises.

Chinese advances in AI technology should be taken seriously as a national security issue, but U.S. utilities ought to prepare for increasing electricity demand either way.