Arm's CEO says AI could cure cancer, but data centers are creating a bottleneck

Arm's CEO says AI could thief humanity cure crab wrong our lifetimes. First, the manufacture needs to fig retired really to make capable chips to powerfulness it.

"I've ever thought that the slayer app for AI is health," Rene Haas said successful an question and reply pinch the BBC released connected Tuesday. "AI is going to not only shorten the magnitude of clip that those narcotics tin beryllium invented, it's going to shorten the magnitude of clip that you trial them."

Haas has led Arm since February 2022, aft joining the British spot designer successful 2013. Before Arm, he spent 7 years astatine Nvidia, wherever he was vice president of its computing products business.

In the interview, the Arm CEO said that AI could ace problems that humans unsocial mightiness not.

"I judge successful our lifetime, AI will thief cure cancer," he said.

Getting there, however, will require tremendous amounts of computing power, Haas said. And the chips needed for crab investigation aliases humanoid improvement are being gobbled up by AI companies readying information centers measured successful aggregate gigawatts.

"Right now, it's rather constrained. We request much fabs earlier we tin put a information halfway successful space, I'll show you that much," he said. "We'll astir apt only put data centers successful abstraction erstwhile the biggest impediment to information centers is the costs of the information center."

Haas said AI's appetite for chips and representation is astatine a standard the semiconductor manufacture hasn't dealt pinch before. Building caller factories isn't a speedy fix: fabs tin costs tens of billions of dollars and return 2 aliases 3 years to construct.

He said the manufacture was successful an "absolutely supply-constrained environment" and expected proviso to stay tight for immoderate time.

Even without the spot bottleneck, AI curing crab successful the adjacent fewer years is acold from guaranteed. This is because aesculapian researchers opportunity that crab is not a azygous illness but a postulation of complete 100 diseases that alteration successful patients.

While AI tin thief researchers place targets and foretell really narcotics mightiness behave, grounds of its clinically applicable effect remains limited. The wide scope of variables successful each assemblage besides makes it difficult for AI models to reliably foretell really a imaginable supplier will behave successful humans.

Data halfway boom

Other spot executives are sounding akin alarms.

In a January interview, Micron's main operating officer, Manish Bhatia, called the representation shortage "really unprecedented" and said high-bandwidth representation for AI accelerators was consuming truthful overmuch manufacturing capacity that it was leaving a "tremendous shortage" for accepted products specified arsenic phones and individual computers.

In a February net call, Qualcomm CEO Cristiano Amon said "industry-wide representation shortage and value increases" would find the size of the handset marketplace done the year, adding that "the full assemblage is impacted by memory."

Data halfway construction successful the US has skyrocketed successful caller years. By the extremity of past year, much than 1,400 had been built aliases approved for build-outs. They person go progressively divisive owed to their effect connected aerial and h2o value successful the neighborhoods successful which they are being built.

A Gallup study of 1,000 adults successful the US, published successful May, recovered that 7 successful 10 respondents opposed constructing information centers for AI successful their section area. Nearly half indicated they were "strongly opposed" to construction.