AI engineers report a new algorithm cutting AI power consumption by 95% through replacing floating-point multiplication with integer addition.

Engineers from BitEnergy AI, a firm specializing in AI inference technology, have developed a method that replaces floating-point multiplication (FPM) with integer addition. This new approach, called Linear-Complexity Multiplication (L-Mul), comes close to the results of FPM while using the simpler algorithm, and despite this, it maintains the high accuracy and precision that FPM is known for. According to TechXplore, this method reduces the power consumption of AI systems by potentially up to 95%, marking a crucial development for the future of artificial intelligence.

However, BitEnergy AI's innovation faces a significant hurdle: popular and readily available hardware on the market, like Nvidia's upcoming Blackwell GPUs, isn't designed to handle this algorithm. Even if L-Mul's performance is confirmed to match FPM, the technology needs compatible systems to function. This reality might give several AI companies pause, particularly those who have recently invested millions or even billions of dollars in AI hardware. Nevertheless, the potential 95% reduction in power consumption could motivate major tech companies and AI chip makers to shift their strategies, potentially developing application-specific integrated circuits (ASICs) that leverage the algorithm.

Power has become the primary constraint on AI development, with data center GPUs sold last year alone consuming more power than one million homes in a year. This energy demand has become so significant that even Google has deprioritized its climate target, with the company's greenhouse gas emissions increasing by 48% from 2019, contrary to its expected year-on-year decline. The company's former CEO even suggested opening the floodgates for power production by dropping climate goals and using more advanced AI to solve the global warming problem.

If AI processing can achieve greater power efficiency through L-Mul, advanced AI technologies could advance without sacrificing environmental sustainability. The 95% reduction in energy use would also ease the burden that massive data centers place on the national grid, reducing the need to rapidly construct additional energy plants. True advancement in AI processing comes not only from increased computational power but also from enhanced efficiency, and if L-Mul works as advertised, humanity could achieve both simultaneously.