Etched.ai, Inc.

Etched.ai, Inc., founded in 2022 by Harvard dropouts Gavin Uberti, Chris Zhu and Robert Wachen, designs transformer-specific ASIC chips: hardware built for one class of model rather than for general-purpose computing. The company's bet is that transformers will dominate artificial intelligence, and that a single-model chip can therefore outperform versatile but less efficient alternatives.

Its product, Sohu, is described as the world's first transformer-specific ASIC. The company reports that Sohu delivers over 500,000 tokens per second on Llama 70B, an order of magnitude more throughput and lower latency than leading GPUs such as NVIDIA's B200. According to Etched, that performance is sufficient to make real-time video generation models and extremely deep chain-of-thought reasoning agents practical.

Technical work spans ASIC chip design, transformer-specific inference acceleration, semiconductor design and memory and interface technologies. The company partners with industry leaders including Rambus on memory and interface components, and is positioned as a challenger to established players such as NVIDIA.

Etched raised a $120m round at a valuation of $5 billion. It operates in the artificial intelligence, AI inference and AI computing, and semiconductors sectors.