To be able to avoid the US artificial intelligence (AI) chip export control, Chinese AI corporations are reported to coach the AI ​​model in Malaysia by smuggling hard disks with large data. One other shortcut was revealed following the GPU smuggling.
The Wall Street Journal (WSJ) reported on Tuesday that 4 Chinese technicians from Beijing arrived in Kuala Lumpur with 15 hard disks of 80 terabytes (TB). It is a total of 4.8 petabytes (PB) data, which is sufficient to learn a big language model (LLM).
The operation is thought to have been planned for several months. Chinese engineers selected to physically transport data since it took a protracted time and risk of being found online.
That they had 4 hard disks to avoid suspicion of Malaysian customs, after which leased 300 NVIDIA AI servers to local data centers to conduct model training.
It didn’t reveal where this Chinese company is. Nonetheless, prior to now, the Malaysian data center was utilized in an identical way, and at the moment, the corporate signed a server lease contract through a subsidiary registered in Singapore.
Nonetheless, when Singapore recently strengthened AI technology regulations, Malaysia also demanded that China’s headquarters directly registered locals.
The US government’s pressure is expanded to Asian countries, and hardware bypass income is becoming an increasing number of difficult. Consequently, some Chinese corporations are selecting a method of exporting data as a substitute of importing chips.
This was identified for instance of China’s semiconductor export control that China might be a technical bypass.
Jensen Hwang Envidia CEO and the Senate of the US have repeatedly questioned the effectiveness of semiconductor export control. In fact, CEOs are intended to resolve export regulations, and the Senate should strengthen the crackdown.
By Park Chan, reporter cpark@aitimes.com
