Cerebras CEO pushes back on AI doomsday warnings

Andrew Feldman, co-founder and CEO of AI chipmaker Cerebras Systems, used his appearance at TechCrunch Disrupt 2026 to argue that the industry's loudest voices are spending too much time on worst-case scenarios. He suggested that some prominent figures in AI perhaps should not be allowed to speak publicly, a remark aimed at the doomer discourse that has moved into the mainstream.
That discourse, as Feldman characterized the landscape, runs from models breaking containment to insiders openly debating the odds that AI could kill all of humanity by the end of the decade. The debate has become familiar enough to be parodied on Saturday Night Live, where a fictional AI executive summed up the mood by saying the company does not condone what it is doing.
Feldman's counterargument is that the upside is getting lost. He wants AI leaders to spend more time describing a world in which, as he put it, your grandkids will never meet anybody who dies of cancer, even if those leaders still hold concerns about risk. That framing is notable coming from a hardware CEO whose business depends on continued AI investment. Cerebras counts OpenAI, Block, Figma, pharmaceutical company GSK, and Lovable among its customers, so a more optimistic public narrative is commercially useful to him. He acknowledged the tension directly rather than dodging it, saying he does not believe powerful technologies exist without the accompanying problem of what happens when bad actors get hold of them. In his view you cannot have one without the other. The position he outlined was not deregulation: he said the technology is powerful, that the industry has to be thoughtful, and that it should be regulated reasonably.
Part of the negativity, Feldman argued, traces back to how data center operators handled the wave of new construction driven by AI demand. He said the industry was not particularly good over the past few years at working with local communities, especially in the United States, and contrasted that with what he said the power, oil and gas sectors have done. In his account, data center builders did not engage communities as they should have, and that became a problem.
He tied the resulting not-in-my-backyard resistance in the U.S. to a set of countries he named as actively courting data center investment: Brazil, Canada, Finland, Iceland, Norway, Mexico, and Sweden. The implication is that siting fights at home can push capacity toward jurisdictions offering a friendlier reception. For developers and infrastructure planners, that is a practical variable, since where compute physically lands affects latency, power costs, and regulatory exposure.
Feldman also addressed a question that hangs over every chip vendor benefiting from the current boom: whether making compute more efficient would shrink demand for the hardware itself. He was not worried. His reasoning is that hardware vendors can improve efficiency while model makers separately improve the efficiency of their models toward the same goals, and that all of these efforts have to combine to produce meaningful progress. This is essentially an argument that efficiency gains and demand growth are not zero-sum. It is also a contested position in the broader market, where cheaper inference per token has historically tended to expand usage rather than shrink it. Whether that pattern holds is something developers watching inference pricing will be able to judge for themselves.
Cerebras occupies an unusual spot in the accelerator market. It builds wafer-scale chips, which keep an entire silicon wafer intact as one processor instead of slicing it into many small dies, a design aimed at moving data quickly for training and inference work. That puts it in the same conversation as Nvidia, whose GPUs dominate AI compute, along with a growing field of alternatives. Feldman's remarks position him as a counterweight to the safety-focused messaging that has become common among frontier model labs, and as an advocate for the infrastructure buildout his company supplies.
The remarks came during a conference interview rather than as a formal policy announcement, so they signal a posture rather than a product or regulatory commitment. Cerebras has not announced new chips, pricing or availability in connection with the comments, and none of the countries Feldman named were described as receiving new Cerebras commitments.
Why it matters: The public argument over AI risk shapes how regulators, local governments and customers treat the buildout, and a chip vendor pushing a benefits-first message adds a commercial voice to a debate dominated by model labs. For developers and product teams, the practical read is that siting, power and community relations are becoming as strategically important as model quality, and that efficiency work on both hardware and models is expected to compound rather than cancel out. Feldman's optimism should be weighed against his obvious commercial stake in continued AI spending.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
Cerebras CEO Andrew Feldman has a message for AI leaders spreading doomsday warnings: Maybe stop talking? At TechCrunch Disrupt, he argued that the industry's focus on AI's risks is drowning out its potential benefits.