Cerebras CEO Andrew Feldman: AI Leaders "Maybe Shouldn't Be Allowed to Talk in Public"

Cerebras Systems CEO and co-founder Andrew Feldman told the TechCrunch Disrupt 2026 audience that "some of the leaders in the AI world maybe shouldn't be allowed to talk in public," directly attacking the AI doomer narrative that has moved from research labs into mainstream entertainment and policy debate. Feldman, whose company builds AI chips that compete with Nvidia, argued that the industry's loudest voices are drowning out the technology's measurable promise — and he put his own alternatives on the record in verbatim detail.

The remarks matter beyond one executive's frustration. Cerebras counts OpenAI, Block, Figma, pharmaceutical giant GSK, and Lovable among its customers, which gives Feldman both a commercial stake in AI's public image and a platform from which to shape it. His comments land at a moment when AI risk discourse has become mainstream enough to be parodied on Saturday Night Live, where a skit depicted Anthropic CEO Dario Amodei and summed up the discourse with the line: "We do not condone what we are doing."

What did Feldman actually say about AI doomers?

Feldman's core claim is straightforward: the constant public discussion of existential AI risk — from models "breaking containment" to insiders debating the percentage chance that AI kills all of humanity by the end of the decade — is damaging the industry without making anyone safer. He did not name specific executives in his quip about who should stop talking, but the target was clear: the cohort of AI leaders who foreground catastrophe scenarios in public appearances.

Instead, Feldman wants AI leaders to spend more time on the upside. He offered a concrete example during his Disrupt interview: a world in which your grandkids will "never meet anybody who dies of cancer." The framing is deliberate. Feldman concedes that AI risk is real, but he insists that leaders can hold concerns about risk while still telling the public what the technology could deliver.

He addressed the tradeoff directly.

"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?' I think you can't make one and not have the other," Feldman said.

He then laid out his prescription in three parts: "I think what we have to say is that this technology is powerful. We have to be thoughtful about it. We have to regulate it reasonably."

That position — powerful, thoughtful, reasonably regulated — places Feldman closer to the pragmatic center of the AI debate than to either accelerationist libertarians or the doom-focused faction his joke targeted. It also aligns with his commercial position: a chip vendor whose revenue depends on sustained AI infrastructure buildout has little incentive to amplify stories about models escaping human control.

Why does Feldman blame the data center industry for AI's image problem?

Feldman offered a second, less obvious explanation for public negativity toward AI: the data center industry's handling of the buildout boom that AI demand created. In his account, the industry failed at community relations in the United States, and the backlash now attaching itself to AI is partly a delayed consequence of that failure.

"As an industry, we weren't super good for the past few years at working with local communities, particularly in the U.S. We — the industry — didn't do what others have done in the power industry in oil and gas. They didn't work with the community. They didn't engage the community as they should have, and that was a problem," Feldman said.

The comparison is pointed. Feldman effectively argued that the AI infrastructure sector repeated a mistake the oil and gas industry already paid for, and that "not in my back yard" opposition to data centers in the U.S. is the predictable result.

He contrasted American NIMBY-ism with a list of countries actively courting data center construction. Feldman named Brazil, Canada, Finland, Iceland, Norway, Mexico, and Sweden as markets where governments encourage companies to build. The list spans three continents and includes both Nordic countries with abundant clean power and North and South American neighbors of the United States, suggesting Cerebras sees the global data center map shifting rather than contracting.

The community-engagement argument also serves as a rebuttal to one strand of AI criticism: that data centers impose costs on local residents — power, water, noise, land — while delivering benefits elsewhere. Feldman's answer is not that the criticism is fabricated, but that the industry created the hostile reception through its own conduct and can fix it by engaging communities the way mature infrastructure industries learned to.

Is efficiency a threat to Cerebras' chip business?

Feldman also dismissed a persistent investor concern: that as AI models become more efficient, demand for his chips — and for chips from Nvidia and other competitors — will fall. The worry has a name in industry circles, the Jevons dynamic, though Feldman did not use it. The logic is simple: if software does more with less compute, hardware vendors sell less silicon.

Feldman rejected the premise by splitting the efficiency gains across the stack.

"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. All of this has to come together to make meaningful strides forward," he said.

In other words, Feldman treats hardware efficiency and software efficiency as complementary rather than substitutes. Compute that costs less per unit of work expands what model makers attempt; model makers then push capabilities further, which consumes the headroom the hardware created. He offered no demand projections to support the claim, but his framing positions efficiency improvements as an industry-wide obligation rather than a threat to any single vendor's order book.

For Cerebras specifically, the stakes are concrete. The company sells AI inference and training systems built around its wafer-scale chips, and its customer list — OpenAI, Block, Figma, GSK, Lovable — spans frontier labs, fintech, design software, pharmaceuticals, and consumer app development. If efficiency-driven demand collapse were real, Cerebras would feel it as sharply as Nvidia. Feldman's public answer signals that he does not see that scenario in his order pipeline.

What does the dispute over AI messaging mean for the industry?

The context for Feldman's comments is a genuine split among AI leaders over how to talk about risk. On one side sit executives and researchers who publicly debate extinction-level scenarios and containment failures, a discourse now visible enough that Saturday Night Live built a sketch around Anthropic's Dario Amodei and the line "We do not condone what we are doing." On the other side sit executives like Feldman who argue that the doom framing taxes the entire industry — affecting hiring, regulation, community acceptance of data centers, and public trust — without producing corresponding safety gains.

Feldman's position is not that risk talk should stop. His own words concede that powerful technology in the hands of bad people is an unavoidable challenge, and his policy ask is for reasonable regulation. The dispute is over emphasis: whether leaders should lead with catastrophe probabilities or with outcomes like a generation that never loses anyone to cancer.

The commercial dimension is impossible to separate from the rhetorical one. Feldman runs a company whose fortunes depend on AI infrastructure spending continuing at scale, on communities accepting data centers, and on governments regulating without restricting. Every element of his Disrupt appearance — the joke about talkative AI leaders, the cancer-grandchildren framing, the community-engagement mea culpa, the efficiency optimism — supports that agenda. That does not make the arguments wrong, but it does locate them within a specific business model under pressure from public skepticism.

The data center argument is the most actionable part of Feldman's case. If he is right that U.S. NIMBY opposition stems from industry conduct rather than opposition to AI itself, then the fix — engaging local communities early, the way oil and gas companies eventually learned to — is within the industry's control. If he is wrong, and the opposition reflects deeper concerns about energy use and environmental cost, then improved community relations will slow but not stop the friction. Countries on his welcome list, from Finland to Brazil, will test the proposition: where governments court data centers, the industry will find out how much of the backlash was about process and how much was about the technology itself.

Feldman's closing message to his fellow executives is the one most likely to travel: the industry's biggest near-term risk may not be the models, but the people talking about them.