AI cycle to last over 20 years: ex-Intel CEO

• By Lisa Wang / Staff reporter

• • Add TT as Preferred Source

The artificial intelligence (AI) supercycle should last for more than two decades, former Intel Corp chief executive officer Pat Gelsinger told an event in Taipei yesterday, dismissing fears of an AI bubble amid mounting skepticism over massive infrastructure investments.

“We do see that there will be periods of disruption, correction, etc. So, we do not think of it as an AI bubble,” Gelsinger, who is a general partner at Playground Global LLC, said at the US venture capital firm’s second annual Taiwan Deep Tech Summit.

However, “we’ll have some speed bumps as different areas and different technology disruptions occur along it,” Gelsinger said.

Playground Global LLC general partner and former Intel CEO Pat Gelsinger speaks at the Taiwan Deep Tech Summit in Taipei yesterday.

Photo: CNA

“Fundamentally, we see AI as a two-decade-plus supercycle of investments and expansion,” he said.

Regarding recent market turbulence, Gelsinger dismissed the panic, saying that “some modest corrections along the way [are] healthy” and serve to keep the industry “in bounds for a long cycle to the future.”

Peter Barrett, another general partner of Playground Global, said there is a massive value migration taking place from the digital realm to the physical world.

Regardless of whether open-source or proprietary models dominate, the one certainty is their reliance on hardware, Barrett said.

This is why Playground Global is heavily invested in its partnership with Taiwan to build the physical foundations of AI, he said.

“Nowhere else in the world do ideas move to reality faster, more significantly than Taiwan,” Gelsinger said. “You and I can have an idea at breakfast, and by lunchtime, Taiwan has produced the prototype. And before dinner, we have early production.”

With Taiwan playing a central role in the global AI infrastructure buildout, accounting for about 90 percent of global AI server production, Playground Global is bringing six start-ups under its portfolio to seek Taiwanese partners, Gelsinger said.

The start-ups aim to tackle critical challenges in AI infrastructure, with focal points ranging from next-generation memory and power management to optical technologies and advanced packaging, he said.

Among them is Atlanta, Georgia-based Nagare Labs, an AI memory start-up that officially emerged from stealth on Monday.

The company is developing Z2RAM — a novel, oxide-semiconductor-based memory architecture designed to shatter the AI memory wall by unifying the speed of SRAM, the density of DRAM and the scalability of flash memory into a single tier.

Gelsinger called Nagare Labs’ work a breakthrough technology that the memory industry has not witnessed in three decades.

“For the AI age, AI is a memory workload,” he said. “I think people have sort of viewed the evolution of AI as being dominated by GPU performance, but it is tipping toward memory at an extraordinary rate.”

AI inference places heavy demands on memory capacity and bandwidth, increasing interest in technologies that combine SRAM-like speed with greater density.

Nagare Labs is in Taiwan to secure foundry partners and build strategic alliances across the ecosystem, cofounder and interim CEO Suman Datta told the summit.

While Datta declined to disclose a specific commercialization timeline, he said a novel memory technology typically requires six to seven years to achieve high-volume production.