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

AMD has acquired Taalas, an AI chip startup that etches machine learning model weights directly into silicon to dramatically improve inference performance. Taalas's model-specific integrated circuits (MSICs) have demonstrated impressive benchmarks, with early test chips achieving 16,960 tokens per second—48x faster than Nvidia GPUs—and the company plans to scale to 20 billion parameters per chip, enabling efficient service of trillion-parameter models with significantly fewer accelerators than competitors like Nvidia and Groq.