The AI craze started yesteryear has caused leading GPU giant NVIDIA to “suffer” a huge chunk of unobtainable profits due to US sanctions against Chinese firms from obtaining powerful GPU/accelerators to train AI models as a result of geopolitical warfare. But of course, with enough dedication and spirit, one can get their hands on contrabands as long as they are willing to.
A Chinese businessman shows off his newly acquired NVIDIA GPUs, even under the U.S. ban.
"It is indeed illegal, but violates the U.S. law. Trump will be furious about it, right?" pic.twitter.com/ADLfM8l7YI— The Great Translation Movement 大翻译运动 (@TGTM_Official) December 10, 2024
As reported by Tom’s Hardware covering X account The Great Translation Movement‘s post, the individual in question showcased and shared 25 units of Supermicro servers with each hosting 8x H200 GPUs from Team Green totaling up to 200 of them. Clearly violating the ongoing sanctions, this “big boss” feels “righteous” that he has “created chaos and furiousness towards US President-Elect Donald Trump”.
Although the amount of GPU might be way too underpowered to train the next state-of-the-art LLM or associated AI technologies, it is actually adequate and satiates the needs of researchers from, let’s say, startups, who need to familiarize themselves with Team Green suite of hardware and software ecosystem of NVIDIA AI Enterprise, Omniverse, you name it.
Despite Supermicro having its fair share of trouble recently when they were accused of fraud based on Hindenburg Research’s report as well as renowned auditor Ernst & Young parting ways with them while the latest situation being “no evidence of fraud found” and had their CFO replaced, it is still unclear in terms of when and how did the Chinese businessman managed to get himself not one, not two, but eight of these systems.
With them costing around US$300K per rack, partners like HPE, Dell, and yes, Supermicro who gets these GPUs from NVIDIA and assembles them into AI servers, are likely to order extra units to ensure consistent production but they may end up on the black market should a well-established smuggling network of dealers and stock movers find such opportunity which is clearly the case here.
We also don’t know if the full 8 systems were acquired from a single or multiple sources but the point is clear – when there are stacks to be made and someone is providing that stack, someone with “particular skills” will take up the job.
