US financial services company Moody’s has warned that the race to adopt AI is placing banks around the world at the mercy of a select few American firms providing foundation models and cloud services. Although Moody acknowledges the upsides of AI in banking, particularly its potential to support revenue growth and reduce operational costs, it says those gains require substantial investment in operational resilience and governance.
While banks have shifted beyond simply evaluating what AI can do, today’s concern is what happens when so much of the industry builds on the same set of suppliers. As in other industries, many financial institutions depend on the same relatively small selection of cloud infrastructure and model providers. As such, if a critical platform or model goes down, disruption could spread across multiple institutions rather than remaining isolated. According to Moody, that “risks creating a systemic dependency.” Banks are especially exposed in this respect, because AI is rapidly moving into business-critical operational activities such as administration, claims processing, and credit assessment.
Pricing is another growing concern. Dependence on a mere handful of vendors often results in market concentration, where dominant providers have greater influence on AI pricing. In addition to substantial initial investments in AI technology and strategy, the risk of supplier price hikes could quickly eat into any early cost-savings brought about by productivity gains and revenue growth.
However, operational resilience goes beyond operational uptime and price to encompass regulatory concerns as well, especially in sectors like banking and finance. This is a major concern in countries and regions outside the US, such as the EU, which is increasingly worried about maintaining its digital sovereignty—and it is actively regulating for it. Moody expects regulators to focus more on operational resilience and third-party risk concentration across the AI tech stack, both in the US and beyond.
Currently, Anthropic, OpenAI, and Google hold almost all the market share in enterprise AI model development, while NVIDIA commands an overwhelming market share for the data center and training GPUs that power those model vendors. While that’s unlikely to change in the near term, Moody points to proprietary data and open-source models as possible ways to reduce dependency on market leaders. That means having flexible adoption strategies that allow institutions to choose from multiple models and vendors, incorporate fallback arrangements, and avoid architectures that make switching prohibitively complicated.
Ultimately, AI architecture is becoming a resilience and supplier-management decision as much as a model-selection decision. After all, what might be the best model for a given use case today might not be the same tomorrow. Also, not every use case requires a leading foundation model that only a major vendor might be able to provide. Many use cases, for example, are better suited to relatively simple automated workflows comprising small, tailor-made models that can either be developed in-house or assembled from open-source components. What increasingly matters to the financial services and fintech sectors is dependency mapping and resilience testing.