The demand for artificial intelligence (AI) continues to grow, driving the need for computing power and electricity. As AI becomes more integrated into daily life, concerns about its environmental impact have intensified. However, a breakthrough offers hope for significantly reducing the energy AI systems require.
Researchers at BitEnergy AI have developed a computational method called linear Complexity Multiplication, which could lower AI energy consumption by as much as 95%. While this innovation can potentially transform AI’s energy usage, it may also require overhauling existing hardware systems.
Moving beyond floating-point multiplication
Most current AI systems rely on floating-point multiplication (FPM), an essential technique for handling complex calculations involving large or small numbers. This precision is especially critical for deep learning models, which perform intricate computations to generate accurate results. However, FPM is highly energy-intensive, contributing significantly to the overall power usage of AI applications.
Linear complexity multiplication shifts away from FPM; integer addition is used to perform calculations. According to the researchers, this change drastically reduces the energy demands of AI systems while maintaining performance levels. Despite the lower energy requirements, early tests indicate no computational accuracy or efficiency drop.
The hardware hurdle
Despite its promising benefits, adopting Linear-Complexity Multiplication is challenging. Most AI systems today are built on hardware optimised for FPM, such as GPUs from leading manufacturers like Nvidia. To implement the new method effectively, entirely new hardware will be needed.
The research team claims they have already designed, built, and tested the required hardware. However, making this technology widely available involves licensing and manufacturing processes that could take time to finalise. Until then, the AI industry may face difficulties integrating the energy-saving technique into its systems.
AI’s growing energy problem
The potential of linear Complexity Multiplication comes at a critical moment for the AI industry. AI systems’ energy demands are soaring. OpenAI’s ChatGPT, for instance, reportedly consumes approximately 564 megawatt-hours (MWh) of electricity daily—enough to power 18,000 American homes. Experts predict AI’s annual energy consumption could soon reach 100 terawatt-hours (TWh), rivalling Bitcoin mining, one of the most energy-intensive digital processes.
Rising energy consumption has sparked calls for more sustainable AI technologies. The introduction of linear Complexity Multiplication represents a significant step forward, but its success hinges on widespread adoption and the availability of compatible hardware.
The researchers at BitEnergy AI are optimistic about their method’s potential to reshape AI’s energy footprint. Still, the road ahead involves navigating technological, economic, and manufacturing challenges.
