Before artificial intelligence can change society, technology will first have to learn to live within its means.
Generative AI currently has an “insatiable need” for electricity to power the tens of thousands of compute clusters required to run large language models such as OpenAI’s GPT-4, its chief marketing officer has warned Ami Badani from chip design company Arm Holdings.
If generative AI can ever run on any mobile device, from laptops to tablets to smartphones, it will need to be able to scale without overloading the power grid.
“We cannot continue to develop AI without turning to power,” Badani said at the Fortune Brainstorm AI conference in London on Monday. “ChatGPT requires 15 times more energy than traditional web search.”
Not only are more businesses using generative artificial intelligence, but the tech industry is in a race to develop new and more powerful tools, which will mean demand for computing will only rise, and with it power consumption, if anything will not be done.
The latest breakthrough from OpenAI, the company behind ChatGPT, is Sora. It can create hyper-realistic or stylized videos up to 60 seconds long based solely on user text prompts.
GenAI’s miracle comes at a cost
“Training Sora requires 100,000 artificial intelligence chips running at full processing power and at full power consumption,” Badani said. “It’s a huge amount.”
Data centers, where most AI models are trained, currently account for 2% of global electricity consumption, Badani said. But with generative artificial intelligence expected to go mainstream, she predicts it could end up consuming a quarter of all electricity in the United States in 2030.
The solution to this puzzle is to design semiconductor chips that are optimized to operate with minimal power consumption.
This is where Arm comes to the rescue: this RISC The processors now power 99% of all smartphones, as opposed to the competing x86 architecture developed by Intel. The latter was a standard for desktop PCs, but proved too inefficient to operate portable battery-powered devices such as smartphones and tablets.
Arm applies the same design philosophy to artificial intelligence.
“If you think about AI, it has a price,” Badani said, “and that price, unfortunately, is power.”