Rene Haas, the CEO of chip design company ARM Holdings, stated in an interview on Wednesday that moving some artificial intelligence functionalities away from the cloud will aid in reducing energy consumption. He believes that the vast gigawatt-scale data centers will be unsustainable over time.

“I think there are two vehicles,” Haas remarked. “One is low power; you can achieve the lowest power solutions in the cloud. ARM indeed contributes to that. But I think more specifically, it's about shifting these AI workloads from the cloud to local applications.”

While he acknowledged that AI training may always occur in the cloud, he noted that operating AI, known as inference, can take place locally, meaning on chips within people's phones, computers, and glasses. Haas pointed out that history shows, “We always adopt a hybrid computing model.”

He believes that in the realm of AI, hybrid approaches will be effective, which will help reduce the immense power investment. ARM's technology supports devices produced by many major tech companies, including Microsoft and Amazon. NVIDIA holds the majority stake in ARM and attempted to acquire the company in 2020.

On Wednesday, ARM and Meta announced that they would expand their partnership “to enhance AI efficiency across every layer of AI software and data center infrastructure.” Following the announcement, ARM's stock price rose, closing up 1.49%.

Haas mentioned in the interview that the collaboration with Meta “mainly revolves around data centers, but more broadly… around software and the associated software stack.” He also discussed ARM’s involvement in Meta’s new Ray-Ban Wayfarer glasses, stating that the AI technology runs simultaneously in the cloud and locally.

He said, “For instance, when you say ‘Hey, Meta’ to the glasses, that’s not happening in the cloud; it’s happening on your glasses and running on ARM.”