Cerebras CEO Pushes Back on AI Doomer Narrative at Disrupt 2026
Cerebras Systems CEO and co-founder Andrew Feldman rejected AI doomsday warnings during an interview at TechCrunch Disrupt 2026. Speaking in a session titled 'Can AI Keep Scaling?', Feldman said some leaders in the AI world "maybe shouldn't be allowed to talk in public" with reference to the doomer narrative. TechCrunch
He offered a contrasting outcome. Feldman said AI "could lead to a world where your grandkids will never meet anybody who dies of cancer." The remark positions large-scale models as instruments for biological discovery and clinical progress. It was deliberate.
Feldman did not argue for an unconstrained buildout. He said "powerful technologies come with the challenge of what happens when they are in the hands of bad people." He added that AI technology "is powerful, must be approached thoughtfully and regulated reasonably." TechCrunch
The business context is inference. Cerebras lists customers including OpenAI, Block, Figma, GSK and Lovable. For practitioners, that mix spans frontier-model serving, fintech, design tooling, pharma and AI-native code generation. Latency matters in all five.
The company describes the Cerebras CS-4 as a rack-scale solution that delivers up to 30x faster inference compared to GPUs. OpenAI's GPT-5.6 Sol is available in Ultrafast mode running on Cerebras at up to 750 tokens per second. Cerebras has partnered with Lovable to power faster AI software creation and with AMD on disaggregated inference combining AMD Helios and the Cerebras Wafer-Scale Engine. Cerebras
Feldman also turned to siting and power. He said "the industry was not good at working with local communities in the U.S. on AI-driven data center buildouts." He named Brazil, Canada, Finland, Iceland, Norway, Mexico and Sweden as markets encouraging companies to build data centers.
TechCrunch Disrupt 2026 featured six editorially curated stages, with Feldman's scaling session among the enterprise and infrastructure programming announced on Sept. 30.
Looking at what this means for technologists, Feldman's comments collapse two separate debates into one conversation. The first is capability risk, who gets access to powerful models and under what guardrails. The second is deployment friction, where to place capacity, how to power it, and how to retain local consent. Both now constrain roadmaps as directly as silicon or memory bandwidth.
In this author's view, optimism about cures and caution about misuse can coexist without contradiction. The long arc of computing has repeatedly expanded beneficial use and abuse surface together, from the PC to the commercial internet to mobile and cloud. What changes with each wave is not whether tradeoffs exist, but how quickly engineering and policy teams learn to manage them. Worth flagging for inference architects in particular is the shift in where scaling pressure sits. Training scale still matters, but product viability increasingly turns on serving efficiency, token throughput, time to first token, and cost per million tokens at sustained concurrency. Rack-scale wafer engines and disaggregated designs that separate compute, memory and fabric are responses to that pressure. So is geographic diversification of data centers toward grids and jurisdictions that welcome large loads.
If Feldman is right that the industry mishandled community engagement in the U.S., the fix is operational, not rhetorical. Noise, water, interconnection queues and rate impacts are local issues. They require early disclosure, measurable commitments and long-term presence. Jurisdictions that provide clear permitting and power planning will attract builds. Those that do not will watch capacity go elsewhere.
