DeepSeek and the Data Center Energy Crisis That Wasn’t

With a rapid series of sweeping executive orders beginning on Inauguration Day, U.S. President Donald Trump quickly overwhelmed the ability of both the media and the general public to follow the news. Still, a powerful reminder of the limits of presidential authority cut through the clutter earlier this week when the Chinese startup DeepSeek introduced its energy-efficient, artificial intelligence-powered chatbot DeepSeek R1 to the market. The announcement sent U.S. technology stocks into a nosedive. Among other impacts, the arrival of DeepSeek upended the case for building new gas power plants to feed the rising demand for electricity from data centers.

What is DeepSeek?  

The DeepSeek R1 chatbot was developed by the Chinese artificial intelligence (AI) firm of the same name founded by the entrepreneur Liang Wenfeng in 2023. The startup seemed to burst out of nowhere, but reporters traced it to Liang’s other venture, the High-Flyer Quantitative Investment Management hedge fund.

Founded in 2015, High-Flyer deploys machine learning for computerized stock trades. The High-Flyer connection explains why DeepSeek was able to develop the new chatbot with little media attention prior to its release, the Associated Press reports. The hedge fund provided the funding and the computing resources for DeepSeek’s AI research, aided by thousands of AI chips procured from the U.S. firm Nvidia before the Joe Biden administration restricted sales of chips to China.

The global impact of the chatbot was evident within days after its official launch on January 20. “A Chinese-made AI model called DeepSeek has shot to the top of Apple Store's downloads, stunning investors and sinking some tech stocks,” BBC News reported. The Monday sell-off sent the Nasdaq Composite — the stock market index that includes the likes of Apple and Amazon — down by more than 3 percent, topped by a 17 percent loss for the U.S. chip maker Nvidia. 

Investors were rattled by the company’s claim to train its chatbot at a fraction of the cost and energy consumption of leading rivals like ChatGPT. Experts surmised that the company developed a way to pair less powerful chips with advanced chips, such as those manufactured by Nvidia, reducing the number of computing hours needed to achieve similar results.

“DeepSeek's achievements undercut the belief that bigger budgets and top-tier chips are the only ways of advancing AI, a prospect which has created uncertainty about the future of high-performance chips,” BBC News reports.

Implications for data center energy demand

Energy transition professionals were quick to react to the DeepSeek chatbot. University of Texas at Austin assistant professor Arvind Ravikumar wrote on the social media platform Bluesky: “Beyond tech industry implications of DeepSeek, I am wondering what this does to the ridiculous projections of electricity demand growth in the next five years. This is a good time to take stock of unrealized efficiency gains in AI compute before we end up building a bunch of 30-year gas plants."

Sustainable solutions for data center energy demand are already emerging in other parts of the world. Earlier this month two Brazilian firms, WEG and Elea Data Centers, announced a partnership in support of plans for a $300 million AI data center project in São Paulo. The data center will deploy transformers from WEG to improve energy efficiency on a systemwide basis.

Elea also plans to use an energy-efficient liquid cooling system. Liquid cooling systems operate without the need for energy intensive mechanical refrigeration, according to the U.S. National Renewable Energy Laboratory. These systems also allow waste heat to be captured and deployed elsewhere. At the laboratory's campus in Colorado, for example, waste energy from a server cooling system is used to heat office space.

A fossil-free future for data centers

Waste heat from data centers can also be used to offset energy use in whole communities, a movement taking shape over the past 10 years. “In particular, smarter data centers have a greater potential to integrate renewable energy into their operations without sacrificing stability and reliability of supply,” TriplePundit reported in 2013. “That, in turn, can lead to new platforms for interacting with local energy suppliers, engaging with local environmental communities and contributing to local economic growth.”

One such plan is evolving in Finland, where Microsoft is building a data center hub powered exclusively by emission-free electricity. The district heating firm Fortum expects waste heat from the server cooling system will provide for about 40 percent of the district heating used by its 250,000 customers in the city of Espoo and neighboring areas.

In the U.S., the obstruction of new wind and solar projects on privately-owned land has become a common occurrence. The Trump administration’s clampdown on offshore and onshore development in federally-controlled areas presents another obstacle. Still, alternative opportunities for growing the nation’s renewable energy portfolio are emerging alongside new strategies for reducing energy demand by data centers. 

The missing link is supportive federal policies that motivate domestic investor activity and innovation, instead of ceding ground to rivals in the global marketplace of ideas. That link was severed when the Trump administration issued a sudden halt to disbursements from federal agencies, among a flurry of other disruptive orders. Perhaps the U.S. electorate did not sign up for that on Election Day 2024, but that is what they got.