It’s become increasingly evident that two parallel AI universes are forming between the U.S. and China. While the U.S. has spawned notable players like OpenAI and Anthropic, China has its own emerging candidates. One of these foundation model developers, Zhipu AI, announced today that it has raised 2.5 billion yuan ($340 million) in total financing to date this year.
Founded in 2019, Zhipu was spun out of China’s prestigious Tsinghua University and is led by Tang Jie, a professor in the university’s Department of Computer Science and Technology.
The announcement has come at a delicate time. This week, the Biden administration imposed additional restrictions on the export of Nvidia AI chips to China, further impeding its rival’s ability to train large language models. In anticipation of semiconductor bans from Washington, China’s deep-pocketed AI companies have been stockpiling semiconductors, spending hundreds of millions of dollars on these much-coveted chips.
To sustain its participation in such a costly AI race, Zhipu is keeping its coffer full by raising from local investors. The $340 million investment came from yuan-denominated funds, marking a shift from a two-decade trend where USD funds were the preferred source of financing until geopolitical tensions created a tech divide.
In August, President Joe Biden signed an executive order barring U.S. investments in critical tech sectors of China, including AI, semiconductors and quantum computing. The goal is to curb China’s military build-up, but the order also sent a chill through China-focused American VCs, which are now avoiding investments in sensitive areas. Some have sought a solution to continue operating in the market by separating their China units, such as Sequoia Capital China, which was renamed to HongShan, and GGV Capital.
HongShan, alongside other prominent VCs like Shunwei Capital and Hillhouse Capital, as well as a state fund managed by Legend Capital, invested in Zhipu.
The AI startup also raised capital from an impressive roster of Chinese internet behemoths, uniting even archrivals like Alibaba and Tencent that rarely co-invest together. The lineup includes Ant Group, Alibaba, Tencent, Xiaomi, Meituan, Kingsoft, TAL Education Group and Boss Zhipin.
Zhipu recently open sourced its bilingual (Chinese and English) conversational AI model ChatGLM-6B, which is trained on six billion parameters and claims to be able to carry out inferences on a single consumer-grade graphics card. It also has an open sourced foundation model trained on 130 billion parameters, the GLM-130B.
Comment