OpenAI announced that its AI models helped achieve a 14 percent power, performance, and chip area (PPA) improvement on AMD’s flagship GPU designs in under five months. According to OpenAI, the optimization delivered the equivalent of a full year of engineering progress, leading AMD to incorporate the AI-generated changes directly into manufacturing. OpenAI also utilized the technique on its internal custom inference processor, Jalapeño, and is collaborating with Synopsys on a dedicated chip-design model called GPT-Synopsys.
Separately, Anthropic is leveraging AI models to streamline its own hardware integration processes. The company reported using its Mythos model to process prompts on AMD MI355X GPUs in three days, down from an estimated six to twelve months required under manual workflows. Anthropic currently runs its Claude models across four separate semiconductor families to continuously balance operational costs and performance per dollar.
Why it matters
Accelerates semiconductor design cycles by replacing months of manual optimization with AI model automated workflows.
Reduces compute costs and improves inference performance through direct AI-driven hardware-software co-design.
Diversifies model deployment across multiple chip architectures to mitigate supply chain bottlenecks and vendor lock-in.
Source: the-decoder.com



