China’s Initiative to Support AI Startups with ‘Computing Vouchers’

China’s Initiative to Support AI Startups with ‘Computing Vouchers’

In response to U.S. export regulations hindering access to high-performance Nvidia GPUs for training large language models (LLMs), China is taking action to assist its AI startups. City governments, such as Shanghai, are issuing ‘computing vouchers’ to these startups to subsidize the costs associated with LLM training. These vouchers, valued between $140,000 and $280,000, aim to alleviate the financial burden stemming from escalating data center expenses and the scarcity of essential Nvidia processors within China. The shortage of these processors is a direct consequence of recent U.S. export restrictions, which prohibit companies like Nvidia from selling advanced AI chips to customers in China and other regions.

Further complicating matters, major Chinese internet firms like Alibaba, Tencent, and ByteDance have restricted access to Nvidia GPUs, reserving them for internal use and select clients. This strategic move was prompted by tightened U.S. export rules, prompting these companies to stockpile GPUs and seek alternative chip sourcing methods.

However, analysts caution that while the vouchers provide financial relief, they only partially address the underlying issue of resource scarcity. According to Charlie Chai, an analyst at research group 86Research, while the vouchers alleviate cost barriers, they do little to mitigate the scarcity of essential resources.

China’s strategy for fostering self-reliance in the AI sector involves subsidizing companies utilizing domestic chips and establishing state-operated data centers and cloud services as viable alternatives to offerings from major tech corporations. Initiatives like the ‘East Data West Computing’ project aim to enhance resource allocation efficiency and reduce energy consumption for AI workloads by creating data center clusters.

About Author


Leave a Reply

Your email address will not be published. Required fields are marked *