Taalas HC1 Hardwired AI Chip 17000 Tokens Per Second

Next Platform·high signal

Taalas HC1 etches Llama 3.1 8B weights into silicon at TSMC 6nm. 17K tok/s per user, 0.0075/M tokens, 250W per card. Tradeoff: one model permanently. LoRA adapters supported. Developer API available now.