Talk about guzzling energy: ENIAC, the world’s first electronic digital computer, weighed 30 tons and consisted of more than 18K vacuum tubes and 1.5K relays, as well as hundreds of thousands of resistors, capacitors, and inductors. Its sole source of cooling was two 20-horsepower blowers, and the room temperature regularly reached 120 degrees Fahrenheit. All of this took 160 kilowatts of energy, and when ENIAC was switched on, there were rumors that the lights in Philadelphia dimmed. When ENIAC was fully humming, blackouts occurred in the area.Today, the major IT energy guzzlers are the large language models of artificial intelligence (AI). Indeed, training a single model consumes more electricity than about 130 US homes use in an entire year, some researchers estimate. However, it’s not just about carbon emissions — vast quantities of water are needed to cool the servers used to power advanced AI models.Right now, financial institutions (FIs), like companies in other industries, probably aren’t aware of the amount of energy needed to process advanced AI-powered algorithms, as they are just beginning to leverage this technology.However, as the use of large language models becomes mainstream, energy consumption will likely become a factor in how FIs deploy AI solutions, as vendors may start to pass on the costs. Moreover, as customers, shareholders, and the general public increasingly demand FIs to be more sustainable, the need to find ways to reduce AI’s energy consumption will become paramount.Strategies To Minimize AI’s Energy Consumption“The commercial realities will probably drive banks to evaluate how they leverage this technology,” says Ben Wallace, a partner at Summit Technology Consulting Group.Here are several ways to mitigate energy use of these advanced AI models:
- Leverage models that have already been trained. Less-expensive, faster, optimized models could cost about one-tenth as much as a large language model that needs to be trained, experts contend.
- Shift to cloud computing. Cloud infrastructure like Microsoft Azure, Amazon Web Services, and Google Cloud have been found to be more efficient for companies than performing workloads in on-premises data centers. Moreover, Microsoft Azure claims to be carbon neutral, offsetting its energy use with large investments in wind and solar energy.
- Perform workloads during off-peak hours and leverage renewable energy sources. FIs like Swiss-based UBS run heavy data workloads at night when there’s low demand and there’s a lot of wind to leverage wind energy sources, or in the morning when the sun is coming out to leverage solar energy.
- Offset energy use. FIs like the $540MM-asset Climate First Bank in St. Petersburg, Florida, actively find ways to go greener. In 2022, Climate First offset 100% of its carbon emissions and gave $71K to environmental nonprofits.
“What I have seen with banks regarding sustainability, is that they see it as an opportunity,” says Falk Rieker, global vice president and head of banking at SAP. “Banks have an important role to play in the economy and in society. Therefore, a lot of the institutions I talk to see it as an obligation to do something around sustainability.”As your institution finds ways to leverage the power of AI’s large language models, look for ways to decrease the amount of power needed to process those advanced algorithms — or find ways to offset energy consumption with other sustainable practices. Your institution will not only reduce costs but also gain the appreciation of your various stakeholders.
