Cerebras Supernova 2026: Revealing CS-4, WSE-3 Turbo, & The Next Frontier of AI Speed
Watch the full Cerebras SUPERNOVA 2026 keynote, featuring the unveiling of CS-4, WSE-3 Turbo, and the Nexus rack-scale platform for ultrafast AI inference. Cerebras CEO and co-founder Andrew Feldman, CTO and co-founder Sean Lie, and leaders from OpenAI, Arista Networks, AMD, Figma, Cognition, and CrowdStrike demonstrate how inference speed is transforming AI agents, software development, design, cybersecurity, and data center infrastructure. CS-4 delivers up to 30x faster inference than GPU-based systems. Powered by three WSE-3 Turbo processors, each CS-4 provides 129.6 petabytes per second of aggregate memory bandwidth, including 43.2 PB/s per wafer. FEATURED SPEAKERS AND DEMOS • OpenAI: Thibault Sottiaux, Head of Core Products & Platforms, demonstrates GPT-5.6 Sol Ultrafast mode running live on Cerebras. • Arista Networks: Jayshree Ullal, CEO, discusses the high-bandwidth network fabrics required to build ultrafast AI clusters. • AMD: Mark Papermaster, EVP and CTO, explains prefill-decode disaggregation using GPU infrastructure for prefill and Cerebras Wafer-Scale Engines for decode. • Cerebras: Jessica Liu, SVP Product Management, explains how faster inference unlocks more responsive and capable AI agents. • Figma: Pavi Bhatter, AI Research Product Lead, demonstrates real-time agentic design with Figma Design Agent. • Cognition: Silas Alberti, SVP of Research and Founding Engineer, discusses Devin, SWE-grep, and the architecture behind ultrafast coding agents. • CrowdStrike and Cerebras: Keith Culley, VP of Engineering at CrowdStrike, and Angela Yeung, SVP Product Management at Cerebras, discuss real-time AI security, time-to-decision, and threat response. • Cerebras: Sean Lie, CTO and co-founder, introduces CS-4, WSE-3 Turbo, the Nexus Platform Architecture, and disaggregated inference. • Cerebras: Julie Shin Choi, SVP and Chief Marketing Officer, delivers closing remarks. LEARN MORE Explore Cerebras CS-4: https://www.cerebras.ai/cs4 Learn more about Cerebras ultrafast AI inference: https://www.cerebras.ai/inference CHAPTERS 0:00 Opening Remarks: The Cost of Going Public 1:33 Why Speed Is Essential for AI 4:06 The Wafer-Scale Engine Journey 7:13 OpenAI: GPT-5.6 Sol Ultrafast Mode 9:11 Speedrun: Humanity’s Last Exam 11:35 The Future of Ultrafast AI with OpenAI’s Thibault Sottiaux 24:33 Cerebras Global Data Center Expansion 26:58 AI Networking with Arista CEO Jayshree Ullal 34:36 GPU vs. Cerebras: Speed and Throughput Roadmaps 37:17 Disaggregated Inference with AMD EVP and CTO Mark Papermaster 47:29 Accelerating AI Agents with Jessica Liu 52:44 Figma Design Agent with Pavi Bhatter 1:01:19 Cognition: Devin, SWE-grep, and Ultrafast Coding Agents 1:08:20 Real-Time AI Security with Angela Yeung and Keith Culley 1:17:00 Introducing Cerebras CS-4 and the Nexus Platform Architecture 1:29:17 WSE-3 Turbo: 43.2 PB/s of Memory Bandwidth per Wafer 1:41:04 How Prefill-Decode Disaggregation Works 1:47:24 CS-4 Performance, Throughput Roadmap, and Availability with Sean Lie #Cerebras #CS4 #AIInference