NVIDIA’s most advanced AI chips continue to draw intense demand in China, even as U.S. export rules and Chinese policy pushes aim to curb reliance on foreign hardware. A new report claims restricted NVIDIA-based AI servers ultimately reached Alibaba through an alleged smuggling route that ran through Thailand, drawing scrutiny to the role of server maker Supermicro and a network of intermediaries.
According to the report, U.S. investigators suspect a large-scale operation valued at roughly $2.5 billion, involving shipments of AI servers equipped with NVIDIA GPUs that were moved from Thailand into China. The allegations include involvement by multiple Supermicro employees, including senior staff, as well as coordination with third-party brokers and a Southeast Asian partner company.
At the center of the route is Bangkok-based OBON Corp, described as a major AI firm in Thailand. Investigators reportedly found that equipment sold to OBON did not remain there and instead ended up in China—specifically at Alibaba—highlighting how high-end AI hardware can be diverted through regional channels despite export controls designed to prevent exactly that.
NVIDIA, for its part, has stressed that partners across its ecosystem are expected to follow strict compliance requirements and that the company will continue working with government authorities to enforce the rules. The situation underscores the pressure placed on chipmakers and server suppliers to police end-users and identify diversion risks in complex international supply chains.
Alibaba has denied any involvement. In a statement, the company said it has no business relationship with Supermicro, OBON, or any brokers referenced in the case, and said it has not participated in alleged smuggling activity. Alibaba also stated it does not currently use, and has never used, banned NVIDIA chips in its data centers.
The report also claims that more than $500 million worth of AI equipment moved within a narrow window between April and May 2025. Much of the attention is on NVIDIA’s H200-class hardware, which was restricted for China during that period. The policy environment around these AI accelerators has been shifting, with later changes indicating the H200 could be sold into China again under specific conditions, including a financial carve-out tied to national revenue.
Thailand’s role is particularly notable as the country continues to attract significant investment from global tech firms building AI and cloud infrastructure across the region. That growth, combined with strong demand for high-end AI GPUs, creates opportunities for gray-market actors to exploit gaps in oversight—especially when shipments can be routed through multiple entities before reaching their final destination.
While similar cases have surfaced before, the broader problem remains difficult to measure. With China’s ongoing appetite for top-tier AI computing and an active black market for restricted hardware across the Asia-Pacific, it’s hard to determine how many banned or controlled chips still find their way into the country. The latest allegations add fuel to concerns that current enforcement approaches may not be enough, and that tighter, more robust controls may be required to deter large-scale diversion.
Adding to the tension, NVIDIA’s CEO Jensen Huang has said the company’s “official” share in China has effectively dropped to zero, emphasizing that NVIDIA is no longer selling chips there under the current climate—while also arguing that global competition should continue.
