​​​​

​​

​​​

​

Since the launch of ChatGPT in 2022, Large Language Models have been a must for all big tech companies. With this comes a higher demand for energy. There are significant and growing concern that stems from the increased appetite for more energy. 

 

Why do LLMs require so much energy?

 

It all starts with training. The most energy demanding phase of the process is training the models. Companies need thousands of Graphics Processing Units (GPUs) from companies like Nvidia to run for months at a time to teach LLMs. It is a staggering amount of energy. Not only is the teaching phase energy intensive, so are each prompts asked on a LLM. A query on either ChatGPT or Google’s Gemini consumes around 2.9 watt-hours (Wh) of electricity. This is about 10 times more energy consumption in comparison to one Google search. 

 

Various studies estimated that in 2024 ChatGPT consumed anywhere from 227 - 1,068 Gigawatts (GWh) of energy based on processing 200 million queries per day. You might be asking what does this mean? The energy used by ChatGPT could power up to 100,000 (at 1,068 GWh) US homes for a year. It is estimated that the average data center consumes 100 megawatts per hour. As demand for AI solutions grows, so will the demand for energy. It is estimated that the big tech data center power demand will rise 15% - 20% per year through 2030 (according to Boston Consulting Group). 

​​

​

Big Tech’s Investment In AI

 

Meta, Google, Amazon, and Microsoft spent $125 billion in data centers in 2024. The estimated capital expenditures in order to increase AI capacity is only expected to rise. This expansion of AI capabilities comes with many concerns and questions apart from uncertainty on return on investment. Questions around an aging energy grid and greenhouse gas emissions need to be answered. 

 

Google’s total emissions (measured by metric tons of carbon dioxide) have risen from 9.7m in 2019 to 14.3m in 2023. Microsoft saw emissions of 15m in 2023. In order to supply the amount of energy needed, all sources of energy will be required, including coal and natural gas. There is hope that hydrogen could fuel the new upcoming data centers to a larger degree. But the biggest impact might come from nuclear power. Amazon, Google, and Microsoft are all investing in Small Modular Reactors (SMRs) to supply nuclear power to data centers built in remote areas. SMRs are a type of nuclear reactors designed to be smaller, more flexible, and more scalable than traditional nuclear power plants. Unlike massive conventional reactors, SMRs are manufactured as a self-contained modules and then shipped to a site for instillation. This approach significantly reduces construction time and costs. SMRs have a power capacity of up to 300 megawatts, which is about one third the size of a typical conventional reactor. This makes them ideal for a variety of applications. 

 

There is no question that big tech companies are all in on AI. All of the major players are projected to spend close to $100 billion dollars next year on building their AI infrastructure. The rising adoption of cloud computing, data analytics, and artificial intelligence is driving capital spending like we’ve never seen before. The global data center market is expected to reach $650 billion dollars by 2030 (CAGR of 11%). Essential services provided by data centers like scalability through cloud computing, improved connectivity and security, and computing power for AI advancement will make them indispensable as companies continue to race toward increasing capacity faster. 

 

The challenge with the investment in data centers are high up front costs with an uncertain path toward monetization. Investors loved these companies for their low capital expenditures, small overhead expenses, and large profit margins. Now there is a shift toward capital heavy industry with technology that is rapidly advancing toward obsolescence and ultimately constant replacement. The good news for investors is that these companies are the most profitable and cash rich companies the world has ever seen. So far, with these huge investments, their balance sheets remain strong and profits are still able to support the increased spending. But how many more years are we going to see data center buildouts and what will the ultimate business models look like?

​​

​

​

​

.  .  .​​​

​

​​​​​

​

​

Opinions expressed here are not to be

taken as investment advice. Consult with

your own investment advisor. 

​

​

​

BACK TO INSIGHTS

​​​​​​​
​​​​

​​

​​

​