Andrew Feldman wants AI leaders to stop scaring the public. Speaking at TechCrunch Disrupt 2026, the Cerebras Systems CEO and co-founder said that “some of the leaders in the AI world maybe shouldn’t be allowed to talk in public,” according to TechCrunch AI. He’s pushing back on a doom narrative that has spread from niche forums into mainstream culture.
That narrative has gone a long way. TechCrunch AI notes that Anthropic CEO Dario Amodei was recently parodied on “Saturday Night Live,” where the fictional version of him summed up the whole debate in one line: “We do not condone what we are doing.” When a sketch comedy show can get laughs out of AI safety talk, it’s no longer just an insider argument.
Why Feldman Is Speaking Up Now
Feldman’s case is simple. The industry talks so much about risk that it’s drowning out the upside. He wants leaders to tell people that AI could lead to a world where your grandkids “never meet anybody who dies of cancer,” even if those same leaders have real concerns.
He isn’t brushing off risk, though. His view is more balanced than the headline suggests:
- Power comes with misuse. “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.”
- Regulation is fine, within reason. “We have to be thoughtful about it. We have to regulate it reasonably.”
- The message matters. You can admit a technology is powerful and still make a hopeful case for it.
Keep his incentives in mind. Cerebras sells AI compute to OpenAI, Block, Figma, GSK, and Lovable. A CEO whose business depends on AI demand has every reason to want the public mood to stay positive. That doesn’t make him wrong, but you should weigh his take with that in view.
The Real Problem May Be Data Centers
What stands out here is Feldman’s second point. He thinks part of the backlash has less to do with existential risk and more to do with concrete and power lines.
“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. He argued that AI builders didn’t engage communities the way oil, gas, and power companies have learned to.
That’s an unusually frank admission. U.S. towns are increasingly fighting data center projects over electricity prices, water use, and land. Feldman contrasted that NIMBY pushback with Brazil, Canada, Finland, Iceland, Norway, Mexico, and Sweden, which he says are actively inviting these builds.
This connects the two halves of his argument. Abstract fear about AI ending humanity and local anger about a giant facility next door feed the same distrust. Fixing one without the other won’t work.
Efficiency Won’t Kill Chip Demand
Feldman also dismissed the worry that more efficient models will cut demand for hardware from Nvidia rivals like Cerebras. In his view, chipmakers and model makers both need to push efficiency, and “all of this has to come together to make meaningful strides forward.” Put simply: cheaper compute leads to more use, not less.
What the Next 1-3 Years Could Look Like
Expect the AI messaging war to get sharper. Here’s what to watch:
- Two camps, louder voices. Safety-focused labs will keep raising alarms, while infrastructure and application companies push the benefits story harder. Policymakers will hear both.
- Community deals become standard. Data center developers will likely borrow from energy companies, with local benefit agreements, grid investments, and early outreach becoming part of every build.
- Capacity moves abroad. If U.S. permitting fights drag on, more compute could land in the Nordics, Canada, and Latin America, where power is cleaner and governments are friendlier.
- “Reasonable regulation” gets defined. Feldman’s phrase will mean very different things to different players. The fight over that definition is coming.
What Businesses Should Do
- Lead with outcomes. If you’re deploying AI, talk about specific problems solved, not vague promises or vague fears.
- Plan for location risk. If you depend on new compute capacity, local opposition and power limits can delay timelines. Build that into your plans.
- Take risk seriously without dramatizing it. Clear governance and honest communication build more trust than either hype or doom.
Feldman’s core argument is that the AI industry has a trust problem it partly created itself. Whether you agree with his take on doomers or not, the data center point is hard to dispute. You can find the full interview details at TechCrunch AI.