How Can We Solve AI’s Energy Problem?

By: Keith Wagstaff · April 22, 2024 · Reading Time: 3 minutes

Power Hungry

Asking ChatGPT to write you a business plan or sonnet might feel like magic. But what if we told you that no matter how simple your ask, it expands your CO2 footprint quite a bit?

In reality, in response to a prompt, graphical processing units run complex calculations in large data centers, which require a ton of electricity. That’s in addition to the massive amounts of energy needed to train AI models such as Anthropic’s Claude 3 and OpenAI’s ChatGPT 4.

This high energy consumption has raised concerns about how AI might contribute to climate change. A single ChatGPT query requires 10 times the energy of a Google search, according to the International Energy Agency.

With the additional AI load, the world’s data centers are on pace to consume more power than the entire country of India, the world’s most populous nation with 1.417 billion people, by 2030, said Arm Holdings plc (ARM) CEO Rene Haas last week to Bloomberg.

Power Savings

There are several ways companies might use and improve AI to lower carbon emissions.

First, the technology could be used to make systems and products more efficient. Over the next four years, AI software could reduce the energy consumption of buildings by 15% to 25%, said Schneider Electric CEO Peter Herweck at the CERAWeek by S&P Global conference last month.

AI could also help make power grids more efficient and resilient, given they have come under stress with the rise of renewable energy sources and EVs. Companies, including Samsung, have already touted AI-enabled appliances meant to cut power consumption.

Second, AI models and infrastructure could be designed to lower their carbon footprint. Google (GOOGL) has increasingly relied on wind and solar energy to power its data centers. It also favors giant hyperscale data centers over smaller, less efficient alternatives.

Cutting-edge chips and servers can reduce the energy demands of artificial intelligence. So can training AI models in novel ways. With CO2 levels on the rise, there could be pressure on companies to make AI as green as possible.

Looking for more stories like this? Check out On the Money — SoFi’s one-stop-shop for news, trends, and tips!

Check it out

Please understand that this information provided is general in nature and shouldn’t be construed as a recommendation or solicitation of any products offered by SoFi’s affiliates and subsidiaries. In addition, this information is by no means meant to provide investment or financial advice, nor is it intended to serve as the basis for any investment decision or recommendation to buy or sell any asset. Keep in mind that investing involves risk, and past performance of an asset never guarantees future results or returns. It’s important for investors to consider their specific financial needs, goals, and risk profile before making an investment decision.

The information and analysis provided through hyperlinks to third party websites, while believed to be accurate, cannot be guaranteed by SoFi. These links are provided for informational purposes and should not be viewed as an endorsement. No brands or products mentioned are affiliated with SoFi, nor do they endorse or sponsor this content.

Communication of SoFi Wealth LLC an SEC Registered Investment Advisor

SoFi isn’t recommending and is not affiliated with the brands or companies displayed. Brands displayed neither endorse or sponsor this article. Third party trademarks and service marks referenced are property of their respective owners.

TLS 1.2 Encrypted
Equal Housing Lender