Etched.ai, Inc. is a semiconductor company founded in 2022 that designs application-specific integrated circuits (ASICs) optimised exclusively for transformer-based AI models. The company's central thesis is that transformer architectures will dominate the AI landscape, warranting dedicated hardware rather than general-purpose GPUs. It was founded by Harvard dropouts.
The company's flagship product, Sohu, is described as the world's first transformer-specific ASIC. Etched claims it delivers an order of magnitude greater throughput and lower latency than leading GPUs such as NVIDIA's B200. On the Llama 70B model, the chip reportedly achieves over 500,000 tokens per second.
Etched operates in the AI inference hardware and AI computing sectors, with technical work spanning ASIC design and transformer model optimisation. The company has raised $120 million in funding at a reported valuation of $5 billion.
