The Chinese tech giants had already been stockpiling Nvidia’s high-performance visual processing unit exports even before they were banned by Washington, anticipating a looming tech conflict between the two nations.
Baidu, a key player in China’s AI landscape, has secured a substantial number of AI chips to continue training its ChatGPT rival, Ernie Bot, for the foreseeable future, as confirmed by CEO Robin Li during a recent earnings call.
Furthermore, the demand for less powerful chips for inference purposes suggests that the existing chip reserves and alternative options will be sufficient to support a wide array of AI applications for end users. However, the quest for the most cutting-edge chips may potentially hinder AI advancements in China, prompting the company to actively seek solutions.
In light of the U.S. trade restrictions, other well-funded Chinese tech companies, including Baidu, ByteDance, Tencent, and Alibaba, have taken strategic measures. Reports indicate that they collectively placed orders worth up to \(4 billion for A800 chips from Nvidia, along with an additional \)1 billion for GPUs scheduled for delivery in 2024.
While significant upfront investments may deter some businesses from entering the Large Language Model (LLM) race, exceptions exist for younger companies able to secure profitable investments promptly. For instance, prominent investor Kai-Fu Lee established an AI company in March, which leveraged loans to acquire high-performance inference cards and swiftly repaid its debts after raising $1 billion in capital.
Baidu recently unveiled the Ernie Bot 4, touted by Li as surpassing GPT-4 in all aspects due to its ample supply of GPUs.
Given the complexity of AI designs, achieving competitive LLMs can be arduous. Some Chinese AI firms have resorted to position boosting by meticulously meeting LLM criteria, although the practical effectiveness of these models remains a topic of debate.
Smaller AI entities may have to make do with less potent processors unaffected by U.S. export restrictions due to financial constraints. Alternatively, they could explore potential acquisition opportunities. Li foresees the industry reaching a “consolidation phase” soon, driven by factors such as chip scarcity, soaring demand for data and AI expertise, and substantial initial investments.
