Cerebras CEO Says AI's Doomsday Chatter Is Drowning Out Its Promise

If you've spent any time following AI news lately, you've probably absorbed a steady diet of warnings: models breaking out of their guardrails, researchers quitting in protest, debates over the odds that artificial intelligence could end humanity within a decade. It's the kind of talk that makes for gripping headlines and, apparently, Saturday Night Live sketches. But one chip industry executive thinks all that doom has a cost of its own.

Speaking at TechCrunch Disrupt 2026 this week, Cerebras Systems CEO and co-founder Andrew Feldman didn't mince words about his frustration with the current discourse. "Some of the leaders in the AI world maybe shouldn't be allowed to talk in public," he quipped, referring to the wave of public hand-wringing coming from inside the industry itself.

He's not wrong that the mood has shifted. Back in September, an Anthropic researcher resigned and publicly warned against building self-improving AI, framing the stakes in terms of a rough percentage chance of catastrophe. That same month, Anthropic CEO Dario Amodei became fodder for late-night comedy, with a fictionalized version of him summing up the industry's uneasy conscience in a single line: "We do not condone what we are doing." When your company's existential risk disclosures become punchlines, something has clearly broken through into mainstream consciousness.

Feldman's counterargument is less about denying the risks exist and more about proportion. Think of it like a doctor who spends an entire checkup listing every possible side effect of a medication without ever mentioning that it might also cure what's ailing you. Patients need both halves of that conversation to make good decisions. Feldman thinks AI leaders have been giving the public only one half.

Weighing Promise Against Peril

The example Feldman reaches for is a personal one: a world where, as he put it, your grandkids will "never meet anybody who dies of cancer." It's a big claim, and an emotionally resonant one. Medical research has long been one of the more credible use cases for advanced AI, from accelerating drug discovery to parsing massive genomic datasets faster than any human team could. Whether AI actually delivers that outcome remains to be seen, but the point Feldman is making isn't really about cancer specifically. It's about balance. He wants the public conversation to hold both the risk and the reward in view at the same time, rather than treating them as mutually exclusive.

It's worth being clear-eyed about where Feldman sits in this debate. Cerebras makes the chips that power AI systems, and its customer list includes OpenAI, Block, Figma, pharmaceutical giant GSK, and Lovable. He has a direct financial stake in AI's continued success and expansion. That doesn't automatically invalidate his argument, but it does mean his optimism and his balance sheet are pointing in the same direction.

To his credit, Feldman didn't try to wave away the risks entirely. "I don't think there are powerful technologies that don't come with the challenge of 'what happens when powerful technologies are in the hands of bad people?'" he said. "I think you can't make one and not have the other." That's a fair, almost mundane observation once you think about it. Nuclear power gave us both energy and weapons. The internet gave us both global connection and new avenues for fraud and exploitation. Feldman's prescription is correspondingly modest: acknowledge the power of the technology, stay thoughtful about its deployment, and "regulate it reasonably." It's less a dismissal of AI safety concerns than an argument that fear shouldn't be the only register the industry speaks in.

Feldman also pointed to a more grounded, less apocalyptic source of public anxiety: the way AI data centers have been built. "As an industry, we weren't super good for the past few years at working with local communities, particularly in the U.S.," he said, contrasting the sector's approach unfavorably with how oil and gas companies have historically engaged with the towns and regions where they operate. That failure to build local trust, he argued, has fed the NIMBY, "not in my backyard," pushback now slowing data center construction across parts of the country.

He drew a sharp contrast with other nations actively courting these projects, naming Brazil, Canada, Finland, Iceland, Norway, Mexico, and Sweden as markets rolling out the welcome mat for data center investment. The implication is pointed: some of the resistance to AI infrastructure in the U.S. isn't really about AI's existential risks at all. It's about noise, water use, land, and communities that feel like they're being built around rather than built with.

On the technical side, Feldman also waved off a common worry in his own industry: that efficiency gains in AI models and chips might eventually shrink demand for hardware like his. His answer was essentially that efficiency and demand grow together rather than canceling each other out. "We can do our job as hardware vendors to make our compute more efficient, and then the model makers can do improved work in making their models more efficient to achieve the same goals," he said. "All of this has to come together to make meaningful strides forward."

Why It Matters

Feldman's comments land at a moment when the public is being asked to hold two contradictory ideas at once: that AI might cure diseases we've fought for generations, and that some of the people building it aren't fully sure they can control it. Neither idea is inherently wrong, and pretending one cancels out the other does a disservice to the public trying to make sense of it all. What Feldman is really arguing for isn't less caution. It's a conversation sturdy enough to hold hope and hazard side by side, because right now, most people are only hearing about one.