Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
There is a high energy cost to generative AI. But even the enormous amount of power needed to train and run large language models pales in comparison to what is required to run the video models behind tools like Sora’s viral app from OpenAIwhich floods our social media feeds with goofy fake clips.
Generative AI models, in general, require a lot of energy to run. The servers running your ChatGPT query are computationally intensive and require a lot of electricity to maintain. Artificial intelligence is The “biggest driver” of electricity use In North America, one report found. This may appear on your energy bill, along with… Artificial intelligence data centers appearing all over the United States, Raising electricity bills From nearby families. Some estimates point to the uses of a single AI query 10 times more energy Just a Google search.
While major AI companies are still reluctant to provide precise details on how much it takes to train and run AI models, there is a growing field of research looking for answers. Sasha Lucioni, AI and Climate Lead at Hugging Face – one of the most well-known AI platforms and think tanks – is a leading researcher studying the energy demands of AI. In a new study, Lucioni and her team examined several of them Open source Artificial intelligence video models. (Popular video tools like Sora, Google I see 3 It was not included in the study because it is not open source.)
The team used the open source Hugging Face database and created AI-powered videos using a variety of models. They measured the amount of electricity needed to create those clips as they varied various factors, including making the videos longer, higher resolution, and higher quality (which was achieved through a process called noise reduction). They ran the test using an Nvidia H100 SXM GPU, a high-powered computer chip that can be used in AI data centers.
“Video creation is definitely a more computationally intensive task — instead of words, you’re creating pixels, and there are multiple frames per second to make videos flow well,” Lucioni said in an email. “It’s complicated.”
Capture AI video at 10 seconds at 240 fps. These are 240 images that the AI needs to create, Lucioni explains. “Especially for high-dimensional content, this really adds in terms of computing power and energy,” she said.
Using AI video power
The study found that publishing video is 30 times more expensive in terms of energy consumption than generating images and 2,000 times more expensive than generating text. Creating a single AI-powered video uses approximately 90 watt-hours, compared to 2.9 watt-hours needed to create images and 0.047 watt-hours needed to create text.
To put these numbers into context, an average energy-efficient LED uses between 8 and 10 watts. LCD TVs Can be used between 50-200 wattswith newer technology such as OLED screens Helping it run more efficiently. Take the 65-inch Samsung S95F, which CNET picked for Best image quality of 2025typically consumes 146 watts, according to Samsung. So creating one AI video would be like turning on this TV for 37 minutes.
The power requirements of generative AI, especially for video, are significant. It sets the stage for a major problem as AI becomes more widely used.
Watch this: The hidden impact of the AI data center boom
Increased demand for artificial intelligence energy
The obstetric video is having a great moment. This is mostly thanks to Google and ChatGPT maker OpenAI. Veo 3 and Sora, the companies’ AI video models, respectively, were launched to much fanfare and have since gone viral. It was the Sora app Over a million downloads Five days after its launch, Google said Gemini users had made More than 40 million videos In the first few months after its debut.
As the use of artificial intelligence grows, the electrical grid in the United States She may not be ready To handle future demand. That’s why AI companies and the US government are supporting a $1 billion campaign to create AI infrastructure. Nvidia recently announced this Investing $100 billion in OpenAI To build artificial intelligence data centers aiming to produce 10 gigawatts on Nvidia systems over the next few years. Microsoft and Constellation Energy are Consider reopening Three Mile Island – the site of America’s worst nuclear power plant disaster – to advance its ambitions in the field of artificial intelligence. But there are other ways in which AI’s energy demands can be mitigated, including using more efficient AI infrastructure.
Individually, we can think critically about whether we need to use an AI tool or not. You don’t always need — or maybe even want — an AI summary every time you search for something, and using alternative browsers can help with that, Lucioni said. But part of the problem is that AI companies aren’t forthcoming about the details of their products’ power requirements.
“AI companies must be transparent about their environmental impacts,” Lucioni said. “It is unacceptable that for the tools we use every day, we don’t have the exact numbers.” “As users, we should have the information we need to make sustainable decisions, and companies have a responsibility to provide us with this information.”