Sberbank seeks Chinese chips for GigaChat as sanctions bite

Reuters reported on May 20 that Sberbank hopes to use Chinese microchips to power Russia's flagship GigaChat AI model. The comment came from chief executive German Gref during Vladimir Putin's visit to China.

This is not just a Russian technology story. It is a clear signal of how AI capacity is being split by sanctions, export controls and access to advanced chips. When hardware cannot be bought freely, AI strategy changes too. Models, suppliers and data centers become part of geopolitics.

Reuters writes that Western sanctions continue to block Russia's access to advanced hardware abroad. Sberbank, Russia's largest lender, has been a central force behind GigaChat and the country's AI push. Gref told state broadcaster Channel One that the bank hopes to use Chinese microchips for the model. He did not say which chips Sberbank was trying to buy.

The queue in China is already long. Reuters notes that ByteDance, Tencent and Alibaba are also rushing to order Huawei's Ascend 950 AI chips. Ascend 950 is described as China's most advanced AI chip, but Reuters says it still trails Nvidia's H200. Sberbank is therefore not just changing supplier. It is trying to secure scarce capacity in a market where China's own internet giants have the same need.

For Norwegian and European companies, the lesson is practical, not theoretical. AI projects should not be treated as pure software projects. They depend on a physical and legal supply chain: chips, cloud, data centers, model access, export rules, support and contracts. When one link is blocked, the whole business case can change.

That matters even for companies with no exposure to Russia. The supplier map is moving. U.S. chips, Chinese alternatives, European data centers and model providers with different legal frameworks are now part of the same decision. If a company puts critical workflows on one AI stack without an exit plan, that is not only a technical dependency. It is a governance risk.

Boards should ask three concrete questions. Which AI services depend on specific hardware or a specific cloud? Which suppliers could be affected by export control, sanctions or geopolitical pressure? And how quickly can the company move workloads, models or data if a provider loses capacity or becomes legally difficult to use?

CIOs and CISOs should also separate pilots from production. A chatbot may tolerate a manual fallback. An agent system that handles code, customer data, credit, logistics or security incidents needs a different level of control. Chip and cloud dependency should be part of the risk assessment, alongside logging, access control, data residency and incident readiness.

The story also shows why European AI sovereignty cannot be reduced to slogans. It is not enough to say that Europe wants its own AI. The questions are who controls the capacity, who gets priority when demand rises, what rules come with the stack, and whether alternatives are strong enough when they are actually needed.

Reuters frames Sberbank's plan as a hope, not as a confirmed delivery. That caveat matters. But the signal still matters: this is what major actors try to do when the best hardware is no longer available. The AI market is not only growing. It is fragmenting.

The executive takeaway is blunt: AI capacity belongs in supplier governance, sanctions review and continuity planning. Not after agents are in production. Before.

Sources and media

• Primary source: Reuters, "Sberbank seeks Chinese chips to power Russia's GigaChat AI model", published May 20, 2026 at 07:37 UTC: https://www.reuters.com/business/finance/sberbank-seeks-chinese-chips-power-russias-gigachat-ai-model-2026-05-20/
• Reuters photo in the original article: Ramil Sitdikov / Reuters. The image is not rehosted by hogby.ai.
• Thumbnail: GPT/OpenAI Image 2 / hogby.ai.