Cerebras CEO Andrew Feldman Tells AI Doomsayers to Stop Talking at TechCrunch Disrupt

Cerebras CEO Andrew Feldman used a TechCrunch Disrupt appearance to push back on fellow AI leaders who warn that their own technology could end badly, suggesting they should talk less about catastrophe. His argument, delivered onstage, was that the steady drumbeat of risk messaging has taken up most of the public conversation while the practical upside of the work gets squeezed out.

Feldman runs a company whose business depends on that upside. Cerebras builds wafer-scale processors and sells AI compute, competing for training and inference workloads against Nvidia's GPUs and the cloud providers that resell them. Its pitch rests on the idea that faster, cheaper compute translates into useful applications rather than existential hazards.

The doomsday framing he objects to has been a fixture of the industry for years. Several prominent researchers and executives have signed statements warning that advanced systems could pose severe risks, and some have called for slowdowns or moratoria on the largest training runs. Those positions have shaped how lawmakers, regulators and reporters approach AI policy debates.

Feldman's counterargument lands in a crowded field of executives making similar claims. Venture investors and application-layer startups have argued that safety rhetoric favors incumbents with the resources to comply with regulation, while smaller builders get frozen out. Whether that critique changes any minds is another matter, since the researchers issuing the warnings are often the same people building the frontier systems.

What Feldman did not offer, at least in the summary of his remarks, is a concrete alternative agenda. Saying the benefits deserve more airtime is not the same as specifying which benefits, on what timeline, or how the industry should handle the legitimate concerns that researchers raise. That gap is where the debate usually stalls.

Watch whether Cerebras ties its public positioning to specific customer results, since compute vendors ultimately get judged on benchmarks and cost per token rather than on stage rhetoric. The broader question is whether the risk-versus-benefit split hardens into two camps with separate conferences, separate funders and separate policy asks, or whether the loudest voices on each side are simply talking past each other.