Clive Chan, who was OpenAI's second hardware employee in its custom chip program, has joined Anthropic. The move comes as both companies prepare for initial public offerings, with Anthropic reportedly considering building its own AI chips. Chan shared on LinkedIn that he is proud of his work at OpenAI, calling the team's 'density of hardware talent' extraordinary. He also expressed confidence in the potential of the chips developed there to become 'one of the most important engines of AGI.'
Chan's departure follows reports that Anthropic is evaluating the development of custom AI chips, following the strategies of OpenAI and Meta. His LinkedIn role description, 'perplexity per picojoule,' suggests a focus on optimizing model performance while minimizing energy use. This could involve either improving software for existing GPUs and TPUs or designing custom silicon tailored to Anthropic's models. Custom chips could potentially boost margins, especially for inference tasks, where purpose-built silicon might offer better long-term financial returns.
Chan previously worked at Tesla's Autopilot division, where he contributed to a custom chip for ML training, including software framework bring-up, datacenter co-design, and energy-efficient number formats. His experience in hardware development is now being leveraged by Anthropic as it seeks to strengthen its position in the AI infrastructure space. Source: thedecoder
