MatX, a chip-design startup, has raised $80 million in funding to develop AI-specific chips optimized for large language models. The company aims to outperform NVIDIA GPUs, targeting a tenfold efficiency increase in training LLMs.