Workflow
High Bandwidth Memory (HBM)
icon
Search documents
'Inference Speed Makes Markets Bigger,' says Cerebras CEO
Bloomberg Technology· 2026-07-24 18:48
Market Trends and Partnerships - Cerebras Systems shares declined by approximately 10% following initial market movements [1] - Cerebras Systems partnered with AMD to develop a new server combining Helios and Cerebras servers to accelerate response times and compete with NVIDIA [1][3] - AMD has been an equity investor in Cerebras Systems for an extended period, participating in mid and later-stage funding rounds [14] - Cerebras Systems integrated open standards-based technology for its input/output (IO), facilitating rapid collaborations with industry leaders like AMD and AWS [8][9] Technology and Infrastructure - AI inference is divided into two distinct computational phases: prompt processing and answer generation [2] - Graphics Processing Units (GPUs) are utilized for processing prompts, while Cerebras Systems technology generates answers at high speeds and throughput [3] - Cerebras Systems avoids the use of High Bandwidth Memory (HBM), mitigating supply chain constraints and reducing token generation costs [16][17] - Cerebras Systems emphasizes open standards rather than proprietary IO technology to enable disaggregated integration with chip makers including AMD, AWS, and Google [9][14][15]