Russia wants to strengthen its push for artificial intelligence, but it faces an obstacle that can’t be solved with talent, data, or political ambition alone: it needs advanced chips. Sberbank, Russia’s largest bank and one of the country’s leading tech drivers, expects to use microchips manufactured in China to power GigaChat, its generative AI model and main national alternative to ChatGPT.
The idea was expressed by German Gref, Sberbank’s CEO, during Vladimir Putin’s visit to China. The message reflects the current geopolitical context of AI: hardware access has become a strategic issue. Western sanctions complicate or block Russian access to advanced accelerators from NVIDIA, AMD, and other providers, while China emerges as the most likely tech partner to fulfill part of that need.
GigaChat, a Russian AI with reliance on foreign hardware
GigaChat isn’t just a simple experimental chatbot. Sberbank presents it as a family of models designed for the Russian language and focused on text tasks, analysis, automation, programming, reasoning, and image recognition. Its API allows integration of these capabilities into products and services, aligning with Russia’s strategy to develop a local AI infrastructure for businesses, government, and regulated sectors.
The GigaChat family has been described in technical reports as a line of foundational and post-trained models for Russian, based on the Mixture of Experts architecture. This approach enables activating only a part of the model for each task, which can improve training and inference efficiency at scale. Sberbank has also released open models for research and industrial use, aiming to expand the local ecosystem and attract developers.
The challenge is that a model of this nature does not exist in software alone. Training, fine-tuning, and serving modern language models require massive computing resources. Russia may have research teams, linguistic data, and political will, but without sufficient GPUs or specialized accelerators, it remains at a disadvantage compared to the US and China.
That’s where Chinese dependence comes into play. Russia currently lacks a domestic semiconductor industry capable of competing with the advanced nodes needed for AI. Its electronics ecosystem remains heavily dependent on imports, and Western sanctions have further tightened the margins. Practically speaking, China becomes the most logical route to access modern hardware, though it’s not necessarily a quick or guaranteed solution.
Huawei Ascend, Russia’s potential pathway to AI
Although Sberbank has not publicly specified which Chinese chips it expects to use, the market naturally looks toward Huawei and its Ascend family. In particular, the Ascend 950PR has become one of China’s major bets to reduce dependence on NVIDIA domestically.
Reuters reported in March that Chinese tech giants like ByteDance and Alibaba were preparing orders for Huawei 950PR chips after favorable testing. The chip’s appeal lies not only in its performance but also in its improved compatibility with software used by developers accustomed to the CUDA ecosystem, which is a key barrier for migrating workloads from NVIDIA hardware.
| Key Element | Situation |
|---|---|
| Russian model | GigaChat, developed by Sberbank |
| Primary need | AI accelerators for training and inference |
| Russian restriction | Western sanctions and limited access to advanced hardware |
| Likely supplier | Chinese manufacturers, with Huawei as a natural candidate |
| Chip highlighted by the market | Huawei Ascend 950PR |
| Sitricting competitors | ByteDance, Alibaba, Tencent, and other Chinese firms |
| Main risk | Limited supply and prioritization of domestic Chinese demand |
The Ascend 950PR, based on available information, is more oriented toward inference than massive training. This may fit part of GigaChat’s needs, especially if Sberbank aims to deploy large-scale services for users and businesses. For training larger models or developing new generations, Russia would also need chips with higher capacity and infrastructure in networking, memory, and storage.
The problem is that China does not have infinite capacity either. Huawei is trying to meet the enormous internal demand driven by models like DeepSeek, the plans of ByteDance, Alibaba, Tencent, and the political pressure from Beijing to replace US hardware. Reuters has reported that Huawei plans to produce around 750,000 units of the 950PR in 2026 — a significant figure but limited by China’s manufacturing constraints in advanced chip fabrication.
This puts Russia in a tricky position. It can see China as a way to bypass Western restrictions, but it will have to compete with China’s own industrial priorities. For Beijing, supplying its tech giants and consolidating AI independence may be more urgent than meeting Russian demand.
Strengthening the Moscow-Beijing tech alliance
Seeking Chinese chips for GigaChat fits into a broader Russia-China approach to AI cooperation. Over the past months, both countries have pushed for increased collaboration in AI, open-source software, tech governance, and strategic applications. Putin had previously ordered the Russian government and Sberbank to bolster cooperation with China in the field, as part of a strategy to reduce reliance on Western technologies.
AI has become a domain where sanctions, export controls, and industrial policies carry as much weight as scientific advances. The US limits sales of advanced chips to certain countries and companies. China invests in domestic alternatives. Russia, more isolated and with less industrial capacity, seeks support from Beijing to stay competitive in a technology that will influence economic, military, and administrative domains.
This also highlights a key difference between China and Russia. China has built a broader industrial foundation over years: chip makers, telecom providers, local hyperscalers, data centers, foundational models, and a public policy aimed at replacing foreign technology. Russia, on the other hand, starts from a weaker semiconductor base with less access to global supply chains.
GigaChat may be competitive in Russian and useful for domestic services, but its future depends heavily on hardware availability. If Sberbank cannot secure sufficient capacity, the model can still advance in specific applications but will remain behind US and Chinese frontier systems. In AI, language, data, and software matter, but computational scale remains a decisive factor.
Hardware, sovereignty, and real limits
The most important takeaway from this news is that AI sovereignty is not declared; it’s built. A country can develop its own models, publish APIs, and build national platforms, but if it depends on another vendor for chips, its autonomy is inherently limited.
Russia seeks to safeguard its technological space from Western sanctions, but in doing so, it increases dependence on China. This dependence may be functional in the short term but reduces its leverage if China’s supply tightens, Beijing prioritizes its own companies, or if available chips do not meet the performance of Western accelerators.
For China, selling AI chips to Russia can have geopolitical and commercial value but also opportunity costs. Each accelerator sent abroad is one less for Chinese companies competing in models, cloud services, or AI applications. In a constrained supply chain, industrial policy may outweigh diplomatic affinity.
The AI race teaches a clear lesson: models are visible, but infrastructure determines who can scale. GigaChat might be Russia’s showpiece for generative AI, but its future depends on something less glamorous than a demo: chips, memory, data centers, energy, networks, and supply agreements.
Russia has found in China the only feasible door to keep its AI ambitions alive. The question is whether that door will be wide enough to power a truly national-scale AI.
Frequently Asked Questions
What is GigaChat?
GigaChat is a family of generative AI models developed by Sberbank, focused on the Russian language and for enterprise, analytical, and automation applications.
Why is Russia seeking AI chips from China?
Because Western sanctions limit Russia’s access to advanced hardware from providers like NVIDIA or AMD, essential for training and running modern AI models.
Which chips might Sberbank use for GigaChat?
Sberbank has not confirmed a specific model, but the market points toward Chinese AI chips like Huawei Ascend, especially given the growing interest around the Ascend 950PR.
Can China meet Russia’s demand for AI chips?
Not guaranteed. Huawei and other Chinese manufacturers are already facing strong internal demand from companies like ByteDance, Alibaba, and Tencent, along with limitations in advanced manufacturing capacity.