AI chip smugglers are exploiting a trade-control loophole by attaching banned HBM from South Korea to low-end processors, because customs checks only processor performance specifications.
A memory smuggler has detailed how he gets export-controlled HBM chips from South Korea into China by exploiting a loophole. According to The Information, the smuggler owns a chip design firm that builds cheap, low-performance processors that fall under U.S. and South Korean export controls. These processors are then paired with expensive HBM from Samsung and SK hynix, which are banned for export to China on their own. Because export control rules only examine the performance of the server processor and not the other components attached to it, the HBM chips installed alongside it can slip through and make their way into China.
The operation is not risk-free. Once the mediocre processors with high-end memory are shipped to Hong Kong and smuggled across the border into Shenzhen in the luggage of commuters who cross between the two regions daily, they end up in workshops. There, technicians heat the boards to soften the solder, carefully lift out the crucial memory chips, and discard the rest of the components. Despite a high failure rate, the high price of HBM makes the procedure quite lucrative.
A Shenzhen-based chip broker has been making big money matching suppliers such as the smuggler with companies desperate for HBM. "A year ago, a single HBM3E unit changed hands in China for several hundred dollars," he tells the publication. "Today it's about $4,000." Since an AI chip requires six to eight high-end memory chips, the broker is raking in $24,000 to $32,000 for every AI GPU.
Chinese AI labs face this same memory bottleneck when acquiring AI GPUs. Chinese memory manufacturer CXMT has already begun risk production of HBM3E and is expected to mass produce them by the end of this year, but it is reportedly still suffering from low yields, which makes its chips expensive to produce and procure. This is evident in Huawei's Ascend 950DT, its current top-of-the-line processor, which sells for about $37,000. It offers pricing similar to Nvidia's B200 AI GPU but delivers only half the performance of the previous-generation H100.
Huawei has reportedly said its next-generation flagship AI processor, the 960DT, will deliver performance similar to the H100. However, because the company cannot purchase HBM from Samsung and SK hynix, and Chinese memory manufacturers lack the capacity to supply all of its chip demand, Huawei has told customers they must find the HBM these servers need themselves. This will further fuel demand, push up prices, and make similar smuggling operations more lucrative.
Given that this method of moving HBM into China is becoming more popular, the smuggler fears South Korean and U.S. authorities will soon clamp down on their operations. Washington has been cracking down hard on such activities, with high-profile arrests including several Super Micro executives and the CEO of a California-based tech company. Taiwan and Malaysia have also ramped up their anti-smuggling efforts, so it is probably only a matter of time before Seoul follows suit.