AI will remake data centers, OCP says

The Open Compute Project (OCP), an industry initiative focused on redesigning hardware for growing infrastructure demands, has turned its focus to the hardware requirements of artificial intelligence, anticipating a massive impact. A key emphasis is  liquid-cooled data centers, with OCP board member and Sun Microsystems co-founder Andy Bechtolsheim among the chief proponents.

At this week’s OCP Global Summit in San Jose, CA, the topic of what AI will mean for computer hardware took center stage. “(AI) is not a trend but a major shift in the way technology is going forward to impact our lives,” said OCP Board Chair Zaid Kahn, general manager of Microsoft’s silicon, cloud hardware, and infrastructure engineering, during a keynote presentation. Kahn predicted that AI will drive tremendous rounds of investment in IT infrastructure and data center buildout in the very near future.

Loi Nguyen, executive vice president and general manager for optical at cloud and data center infrastructure technology company Marvell, agreed. “10 years from now, when we look back, I think most of you will agree with me that 2023 [was] the new beginning of AI. The world will be very different 10 years from now.”

However, AI presents many system design challenges, said Zane Ball, Intel’s vice president and general manager for data center platform engineering and architecture. Ball cited power consumption as the most important problem facing AI users, with AI models growing tenfold per year, driving huge infrastructure builds that consume megawatts of power. Ball said liquid cooling will be pervasive because it enables power savings of 30% at the data center level. Even CPUs will demand liquid cooling, he said, adding that Intel will invest in CPUs to make AI better so it becomes a practical application on a standard server.

Bechtolsheim, chief development officer and co-founder at Arista Networks, chimed in about benefits of water-cooled systems during his own presentation after the keynote, saying the era of cooled data centers has arrived with AI. “People want to deploy very large data center designs that are based on liquid cooling,” Bechtolsheim said, but there are challenges. “The bottom line is liquid cooling is more complex than air cooling, no question about that.”

Recounting recent AI developments such as ChatGPT and Amazon Bedrock, AWS’s new generative AI service, Marvell’s Nguyen predicted that AI would enable innovations ranging from personalized healthcare to the abatement of climate change to communication with whales. But data centers will have to become larger and they will consume more power, he stressed.