In the escalating tech conflict between the U.S. and China, Chinese companies have found a surprisingly low-tech yet effective workaround to American restrictions on artificial intelligence development. While Washington tightens controls on chip exports, Chinese engineers are turning to physical data transfers using hard drives—a strategy as old-fashioned as it is ingenious.
Smuggling data, not hardware
Instead of smuggling hardware or rerouting supply chains, Chinese firms are physically transporting massive datasets abroad to train AI models on U.S.-made chips. In one case, four engineers flew from Beijing to Malaysia carrying suitcases packed with hard drives—each containing up to 80 terabytes of training data. Once there, they used rented servers powered by Nvidia chips to train their AI model and then returned with the output.
Why not just use the cloud?
The answer lies in speed and security. Transferring that volume of data online would have taken months, and introducing those same chips into China is increasingly risky due to customs scrutiny. This method, while manual, allowed for faster, more controlled deployment of training data, with weeks of preparation spent fine-tuning the dataset.
A growing regional workaround
Countries like Malaysia, Singapore and Saudi Arabia are becoming key data centers for Chinese firms seeking AI power offshore. Some entities even create local shell companies to lease server capacity discreetly. Meanwhile, the U.S. continues issuing warnings but has stopped short of setting firm limits on chip usage abroad—leaving space for these strategies to flourish.
The new normal in global tech warfare
As pressure mounts, China’s use of physical data transfers reflects a broader shift toward unconventional tactics in the AI arms race. It’s not just a workaround—it’s a calculated, global logistics operation.