AMD announced on August 6, 2026 that it will acquire Taalas, a startup developing model-specific AI inference chips that bake neural network weights directly into silicon rather than loading them from memory. The Taalas HC1 demonstrator, built on TSMC 6nm technology, reportedly achieves up to 17,000 tokens per second on Llama 3.1 8B, trading hardware flexibility for significant performance gains in stable, high-volume inference workloads. AMD plans to integrate Taalas technology into its accelerator roadmap alongside its existing Instinct accelerators, EPYC CPUs, and ROCm software.
