The Sohu chip, built using Taiwan Semiconductor Manufacturing Corp.’s advanced four-nanometer process, is claimed to deliver superior inferencing performance compared to Nvidia’s GPUs while using less energy. The company has already received orders worth tens of millions of dollars for its chips and plans to launch the Sohu Developer Cloud platform soon. Despite the challenges of entering the semiconductor market, investors are hopeful that startups like Etched.ai can challenge Nvidia's market dominance.
Key takeaways:
- AI chip startup Etched.ai has raised $120 million in a Series A funding round to challenge Nvidia's dominance in the AI chip market. The round was led by Primary Venture Partners and included participation from high-profile investors such as Peter Thiel and Amjad Masad.
- Etched.ai's first product, Sohu, is a dedicated AI chip designed to train, deploy, and optimize transformer models. The chip is built using Taiwan Semiconductor Manufacturing Corp.'s advanced four-nanometer process and is claimed to deliver enhanced inferencing performance while drawing less energy.
- Etched.ai CEO Gavin Uberti claims that Sohu is significantly faster and cheaper for transformer-based text, image, and video generators than Nvidia's upcoming Blackwell GB200 GPUs. The company has partnered with TSMC to manufacture its chips.
- Despite the dominance of Nvidia in the AI chip market, several startups are emerging in the sector, including Cerebras Systems, Tenstorrent, Taalas, SiMA.ai, and DEEPX. Investors believe there's room for more than one major AI chip supplier, leading to increased attention and funding for these startups.