AMD acquires Taalas to boost inference performance by etching models in silicon

AMD acquires AI chip startup Taalas, which bakes model weights directly into silicon using model-specific integrated circuits (MSICs) to boost inference performance by an order of magnitude, with early demos showing up to 17,000 tokens per second. The deal, framed similarly to Nvidia’s licensing deal with Groq, aims to make high-performance inference services faster and cheaper to run. Taalas’ first test chip was fabbed on TSMC’s 6nm process.