Herman Gref arrives in Beijing in search of chips for GigaChat

Sberbank CEO German Gref announced plans for the Russian bank to use Chinese processors to develop GigaChat. This statement was made on federal television against the backdrop of a two-day visit by the Russian president to Beijing and chronic shortages of imported equipment caused by Western sanctions, which have made access to technologies from the US and Europe difficult.

Western Blockade Pushes Sberbank Eastward

The expansion of Western sanctions against Russia over the past two years has made it nearly impossible for Sberbank to have direct access to advanced NVIDIA H100 and A100 graphics processors, which for many years have been considered the gold standard for training large language models. These chips provide maximum computational speed and energy efficiency — qualities critical for scaling models like GigaChat. Without access to them, developing the model becomes extremely difficult and economically inefficient.

Under conditions of a de facto blockade of American and European technologies, German Gref articulated a pragmatic position for the bank: the company is willing to work with available equipment from China, primarily Huawei (Kunpeng) processors and other alternatives that can still be purchased on the open market. This is not an ideal option — Chinese chips significantly lag behind the best NVIDIA models in performance and have their own compatibility limitations with Western software. However, by Gref's logic, modest equipment is better than its complete absence. Under conditions of severe informational and technological pressure on Russian companies, to be without tools for AI development would mean capitulation.

A Long Queue for Scarce Resources

But there is a serious catch, which Gref is probably well aware of: the global queue for Chinese processors is already enormous, and Sberbank may find itself far from the front — more likely at the back. ByteDance (owner of TikTok) and Alibaba — two of the largest Chinese technology conglomerates, which annually consume hundreds of thousands of chips for their data centers, cloud platforms, and research divisions — have already reserved significant production volumes for the coming quarters. Dozens of other major companies are joining them. This means that Sberbank's actual access to the necessary volumes could be significantly limited and unpredictable.

• Huawei and other manufacturers operate at near maximum capacity, prioritizing domestic market needs
• The Chinese state and national champions receive priority in supplies — this is an instrument of Beijing's industrial policy
• Logistics create chaos — shipments are delayed for months, volumes are cut sharply and without warning
• Political risks are rising — any new sanctions against China could cut off the supply source entirely

For Sberbank, this transforms the problem from a commercial logistics issue into an existential strategic one. The competitiveness of GigaChat now depends not on the talent of engineers, but on geopolitical forces over which the bank has little influence.

What This Means

The situation illustrates the split in the global AI market. An archipelago of separate ecosystems is forming: the Western one (NVIDIA → US, Europe) remains the most advanced, but is closed to Russia and China. The Chinese ecosystem (Huawei, Alibaba, ByteDance) is developing rapidly, but resources are distributed within China. The Russian ecosystem is forced to compete for access to hardware on the periphery of these systems. For Russia, this means that the future of its AI is determined not by ambitions and engineering skills, but by equipment availability. And this access will become a chronic bottleneck.