U.S. sanctions prompted Chinese tech companies to fast-track artificial intelligence advancement without relying on the latest American chips.
Chinese companies, including Alibaba Group Holding Limited BABA, Baidu, Inc BIDU, and Huawei Technologies Co, Ltd, are studying techniques to achieve state-of-the-art AI performance with fewer or less powerful semiconductors, the Wall Street Journal cites research papers and employee interviews.
They are also researching ways to combine different types of chips to avoid counting on any one kind of hardware.
The race to commercialize ChatGPT-like models fueled the need for more powerful chips and ways to squeeze more out of them to reduce the costs of AI development.
The U.S. sanctions cut China from the most advanced chips made by Nvidia Corp NVDA, forcing them to consume American chip stocks to create their own ChatGPT equivalents.
OpenAI released ChatGPT a month after the chip sanctions came into force, triggering a global rage to develop generative AI.
Alibaba and Baidu, which stashed A100s before the sanctions, have earmarked them for the most computationally intensive tasks.
Nvidia created downgraded versions of its A800 and H800 chips for the Chinese market, called the A800 and H800, to bypass the sanctions.
However, China’s move is a costlier one. If a U.S. firm needs 1,000 H100s to train a large language model, a Chinese firm could need 3,000 or more H800s to achieve the same results, Yang You, a National University of Singapore professor. Experts also flagged the technical challenges of getting them to work reliably.
Dylan Patel, the chief analyst at SemiAnalysis, told WSJ that Chinese researchers’ pain points would only aggravate without access to the new Nvidia H100, which includes an extra performance-boosting feature especially helpful for training ChatGPT-like models.
Price Action: BABA shares traded lower by 0.44% at $82.86 on the last check Monday.
Photo by Body Stock and Dragon Claws via Shutterstock