Cerebras CEO pushes back on AI doomsday warnings

Cerebras Systems CEO and co-founder Andrew Feldman thinks the AI doomer communicative isn’t doing the manufacture immoderate favors. At TechCrunch Disrupt 2026, the AI chipmaker exec quipped that “some of the leaders successful the AI satellite possibly shouldn’t beryllium allowed to speech successful public.”

He’s referencing the negativity present spreading astir the dangers of AI, from models breaking containment to AI insiders debating the percentage chance that AI whitethorn termination each of humanity by the extremity of the decade. The statement has go truthful mainstream that Anthropic CEO Dario Amodei was recently parodied successful a skit connected Saturday Night Live, wherever the fictional quality summed up the discourse, saying, “We bash not condone what we are doing.”

Feldman, however, thinks the committedness and imaginable that spell hand-in-hand with AI developments are getting overlooked. He thinks AI leaders should walk much clip telling radical that AI could pb to a satellite wherever your grandkids volition “never conscionable anybody who dies of cancer,” for instance, adjacent if they person concerns astir AI risks.

Clearly, Feldman has a vested involvement successful AI being successful, arsenic the company’s customers see large names moving with the technology, similar OpenAI, Block, Figma, pharmaceutical elephantine GSK, and Lovable.

However, he’s pragmatic astir weighing the tradeoffs betwixt hazard and reward.

“I don’t deliberation determination are almighty technologies that don’t travel with the situation of ‘what happens erstwhile almighty technologies are successful the hands of atrocious people?’ I deliberation you can’t marque 1 and not person the other,” Feldman said during his interrogation astatine Disrupt.

“I deliberation what we person to accidental is that this exertion is powerful,” helium added. “We person to beryllium thoughtful astir it. We person to modulate it reasonably,” helium added.

The chipmaker CEO besides suggested that immoderate of the negativity astir AI has to bash with however the information halfway manufacture has approached the caller buildouts created by AI demand.

“As an industry, we weren’t ace bully for the past fewer years astatine moving with section communities, peculiarly successful the U.S. We — the manufacture — didn’t bash what others person done successful the powerfulness manufacture successful lipid and gas. They didn’t enactment with the community. They didn’t prosecute the assemblage arsenic they should have, and that was a problem,” Feldman said.

He compared the NIMBY-ism — “not successful my backmost yard!” — of what’s going connected successful the U.S. with different countries, naming Brazil, Canada, Finland, Iceland, Norway, Mexico, and Sweden arsenic markets wherever they’re encouraging companies to physique information centers.

Feldman besides wasn’t acrophobic that ratio improvements would make little request for his chips, oregon those from different companies competing with Nvidia, arsenic Cerebras is.

“We tin bash our occupation arsenic hardware vendors to marque our compute much efficient, and past the exemplary makers tin bash improved enactment successful making their models much businesslike to execute the aforesaid goals. All of this has to travel unneurotic to marque meaningful strides forward,” helium noted.