Rusia adquirirá GPUs NVIDIA H100 de segunda mano, desechadas por China, para desarrollar Supercomputadoras en 2030 e impulsar la IA

Russia needs hardware, but faces import bans from all possible channels, leaving them behind in artificial intelligence (AI). To address this issue, Russia plans to purchase second-hand NVIDIA GPUs from China to build supercomputers, aiming for ten of the world’s top 500 supercomputers by 2030.

China and Russia’s partnership comes into play once again, potentially causing tensions between themselves, the United States, and the Biden administration. Russia will be forced to buy second-hand NVIDIA hardware from China for their supercomputers due to existing sanctions.

This hypothesis emerges as Russia’s intentions become clear, seemingly exploiting a loophole in U.S. restrictions on China. The resale of owned hardware doesn’t appear to be covered by American measures, paving the way for cooperation between the two countries.

For Putin’s 2030 goal, these ten supercomputers will require between 10,000 to 15,000 NVIDIA H100 GPUs. These GPUs cannot be H800, A800, or the banned A100 models, as the maximum computing power is necessary to be on the list within six years.

In a little over those six years, hardware technology is expected to advance further, posing multiple challenges. However, from a power perspective, Russia would be able to develop a ChatGPT-like AI system with their hardware and software engineers.

Initially, it was suggested that Russia and Trusted Infrastructure would acquire the H100 GPUs through smuggling. However, the high volume of GPUs needed and low availability make this an unlikely option. The solution, as rumors suggest, lies with their ally, China.

Chinese companies that upgrade their hardware will remove the H100 GPUs they currently own and replace them with newer models, allowing them to recoup some of their initial investment by selling them to Russia. Current NVIDIA GPU prices predict that the Russian project could reportedly cost around $6 billion, but with Chinese support, the cost could be reduced by $500 million to $700 million.

With these second-hand GPUs, added to Russia’s existing 1,592 NVIDIA A100 units, Putin aims to avoid falling behind in AI advancements.