DeepSeek's Chatbot: A Low-Cost Challenge to the AI Industry

Chinese artificial intelligence startup DeepSeek has surprised markets and AI experts by claiming it developed its highly popular chatbot at a fraction of the cost incurred by American tech giants.

This assertion raises important questions about the billions of dollars that U.S. technology companies are investing in expansive energy-intensive data centers, which they argue are necessary to drive the next wave of AI advancement.

Could this newly developed AI signify that the global demand for electricity to support such technologies is significantly lower than previously thought? The implications of this could be immense, especially in the context of climate change.

The AI industry is known for consuming vast amounts of energy, much of which is derived from fossil fuels, contributing to climate change. Tech companies have reported an increase in their electricity consumption, contrary to earlier predictions of a gradual decline, undermining their plans to combat climate change.

“There has been a very gung ho, go ahead at all costs mentality in this space, with a push towards investment in fossil fuels,” stated Eric Gimon, senior fellow at Energy Innovation. “This presents an opportunity to slow down some of these developments.”

Experts argue that making AI technologies more efficient could lessen their impact on the environment, even if the substantial energy demands persist.

DeepSeek's Rising Popularity

DeepSeek’s claim of developing their standout chatbot on a budget has sparked considerable interest, propelling its AI assistant to become the most downloaded free app on Apple’s iPhone in a week, surpassing competitors like ChatGPT and Google’s Gemini.

“Suddenly, we woke up Monday morning to see a new player at the top of the App Store, and it could overnight change the game,” noted Jay Woods, chief global strategist at Freedom Capital Markets. “This caused quite a stir among some of the hottest stocks globally.”

DeepSeek's app successfully competes with other prominent AI models, displaying capabilities such as composing software code, solving complex math problems, and answering multi-step queries. It has also piqued interest due to its ability to articulate its reasoning in the answering process.

Analysts have been analyzing DeepSeek's publicly available research papers concerning its new model, R1, and its predecessors. A standout detail is DeepSeek’s claim that the cost to train its flagship v3 model was only $5.6 million, which is remarkably low compared to the billions invested to develop ChatGPT and other well-known AI systems. However, DeepSeek has not yet commented on this claim.

It's important to note that the $5.6 million figure represents only the actual training costs of the chatbot; earlier research and development expenses were not included. Moreover, the startup faced limitations due to U.S. export restrictions on the most powerful AI chips, using a less capable chip from Nvidia that is still permitted for sale in China.

Projected Energy Consumption

Energy consumption by data centers in the United States is anticipated to double or triple by 2028. According to the Lawrence Berkeley National Laboratory, data centers accounted for about 4.4% of all U.S. electricity consumption in 2023, with predictions suggesting this could rise to between 6.7% and 12% by 2028.

The prevailing assumption has been that U.S. tech giants must invest significantly more in data center infrastructure to train and operate their AI systems. For instance, Meta Platforms plans to spend up to $65 billion this year, including a major new data center complex in Louisiana.

Meanwhile, Microsoft has indicated plans to allocate about $80 billion for this year. Additionally, a collaborative venture involving Trump, CEOs from OpenAI, Oracle, and SoftBank recently announced aspirations to invest up to $500 billion in data centers and related electricity generation for AI development, starting with an ongoing project in Texas.

Future of AI and Energy Efficiency

Experts believe that if AI technology becomes more efficient, its usage will likely increase, resulting in continued growth in energy consumption. Vic Shao, founder of DC Grid, which provides off-grid direct current power to data centers and EV charging stations, stated that when innovative technology becomes accessible and affordable, it tends to be widely adopted.

As a result, while more efficient data centers may be built, the demand for electricity will still rise, according to Travis Miller, an energy and utilities strategist at Morningstar Securities Research. He stated, “We think the growth in electricity demand will end up at the lower end of most scenarios.”

If DeepSeek's claims are valid, some common AI queries may not require data center operations and could be processed on mobile devices, as suggested by Rahul Sandil, vice president and general manager for global marketing at MediaTek. This shift could alleviate computing demands and provide a timeframe to enhance renewable energy sources for data centers.

Bloom Energy, one of the AI-related stocks affected by this news, saw its share prices decline on Monday. Founder and CEO KR Sridhar emphasized the importance of U.S. leadership in AI, pointing out its potential to power data centers using clean energy compared to other countries that still rely heavily on coal.

“We can continue to improve our efficiency and will keep striving to do better,” he added.

Rick Villars, an analyst from IDC, indicated that the developments from DeepSeek might influence AI research directions in the future, but the need for substantial data center capacity and electricity will remain. “We believe this could accelerate the timeline for when AI becomes significantly integrated into various aspects of our work and daily lives,” he remarked. “Thus, the requisite capacity will still be necessary.

AI, Energy, Climate