The US government has been doing everything that is in their hand to prevent Chinese companies and research organizations for more than three years artificial intelligence (AI) more advanced that produce nvidia, AMD or brainsamong other companies. This is the purpose of Export restrictions Of these GPUs approved by the Department of Commerce, which in this area is the executing arm of the administration.
Despite the efforts of the US government, the avant -garde chips for AI They have continued arriving in Chinaalthough it is reasonable to accept that in amounts lower than those needed by companies that are dedicated to the development of large AI models. And they have done it mainly through secondary markets and parallel import roads that run in India, Malaysia or Singapore, in which The US action It is very limited.
Nvidia has achieved the export license you need to sell in China Your GPU for IA H20but the Chinese government has vetoed this chip. And he has done so that the administration of the cyberspace of China, which is the main Internet regulatory body in this country, This GPU is thoroughly investigating Because he suspects that he could incorporate a rear door of difficult location by Chinese experts. If so, the possibility of China to use this chip.
China is getting in the second -hand market the best Nvidia GPU
China already has Three very clear alternatives to Nvidia. Cambricon Technologies, Huawei and Moore Threads are the three Chinese chip designers for ia who are better positioned to get the market that NVIDIA maintains in China. Despite all the developers of great AI models that have projects with projects with CUDA They need chips from this last company. And, according to Digitimes Asiathey have found the appropriate place to get these GPU: the international second -hand market.
However, the most interesting thing is that they are not being formed with the H20 chips that, as we have seen, have been prohibited by the Chinese government; They are getting the GPUs H100 and A100 of Nvidia, which They are more powerful than H20. For the US it is essentially impossible launch an unpublished strategy. This measure will consist, if it finally comes to fruition, to introduce the technology necessary to allow them to be tracked at all times.
Chinese companies are buying huge games of NVIDIA A100 and H100 in the second -hand market
A priori would be a form of geolocation that would identify where each integrated circuit resides during its journey once it leaves the semiconductor factory. Michael Kratsios, the director of the Office of Scientific and Technological Policy of the White House and one of those responsible for the US government’s action plan for the development of AI, has confirmed that are considering the possibility of introducing tracking solutions both implemented in the hardware and software itself: “We are discussing what software or physical changes could be introduced into the chips themselves to improve the tracking of their location.”
The US administration is determined to protect its current leadership in the development of hardware for AI as the adoption of this technology accelerates throughout the planet. However, implement the solution proposed by Kratsios and other White House advisors It implies important technical challenges. Constant tracking by GPS is not practical in this scenario of use, According to Kratsios himselfso a possible solution is to introduce in each chip the logic necessary to confirm its location periodically. In this way, external networks can be reduced.
The problem is that this design increases the complexity of the logic of each chip and can presumably impact its performance, which would penalize its productivity in the calculation centers. In addition, the cost of each GPU will increase and hackers They could find new ways to violate the safety of these chips by resorting to possible weaknesses of tracking logic.
Image | Nvidia
More information | Digitimes Asia
GIPHY App Key not set. Please check settings