AI resilience: managing geopolitical AI dependencies with confidence

AI resilience is gaining importance

In view of geopolitical tensions, particularly with the US and China, AI and cloud dependencies are a critical reality for banks. But how should they be dealt with? The best approach is to tackle the risks in a focused manner, as described below!

In view of international tensions, particularly in transatlantic relations, dependencies on cloud providers outside Europe are increasingly becoming the focus of credit institutions.

Some institutions are already responding: for example, Landesbank Hessen-Thüringen (Helaba) announced in a press release on 12 August 2025¹ that it is entering into a partnership with Schwarz Digits, the infrastructure service provider of the Schwarz Group.

It is therefore time for financial institutions to make their AI infrastructure crisis-proof against potential risks. But is the switch that easy – and is it also economical?

At first glance, cloud-based approaches offer clear advantages, especially when scaling GenAI solutions, for example through flexible infrastructure and high computing capacities, regardless of whether they are hosted in the US or the EU. Since specialised infrastructures such as GPUs are required for the productive use of such models, the economic operation of generative AI – for example, large language models – only makes sense for many banks under certain conditions.

A sample calculation illustrates the significant investments that would be associated with on-premises operation of a company GPT chatbot – independently of external cloud providers.

The sample calculation in Figure 1 shows a difference of over EUR 90,000 between in-house operation and cloud operation of AI models.

Hyperscalers offer very well-scaled prices for AI infrastructure, making independent operation unattractive.

This makes it clear that, from a cost perspective alone, sourcing GenAI capabilities from large hyperscalers is actually the only attractive option. Nevertheless, financial institutions should be aware of the associated risks and dependencies – and actively manage them. A multidimensional view of technological and operational aspects helps to ensure the confident operation of AI applications.