Pulse News

**Europe’s Banks, Unlike U.S. Rivals, Play With DeepSeek**

By Michael Roddan

Mar 11, 2025, 6:00am PDT

The sudden rise of DeepSeek represents an opportunity for banks and financial firms to access a low-cost but still sophisticated artificial intelligence model—if only it wasn't from China. Geopolitical tensions between Washington and Beijing, along with perceived security risks posed by the Chinese-made AI tool, have kept U.S. banks from even experimenting with DeepSeek.

**The Takeaway**

- HSBC, NatWest are among European banks testing DeepSeek

- U.S. banks are unwilling to work with DeepSeek

- DeepSeek's low cost could offset the expense of overhauling older systems

Banks on the other side of the Atlantic appear to be less worried about such matters. Since the release of DeepSeek's groundbreaking R1 model in January, major European institutions including HSBC, NatWest (one of the U.K.'s big four banks), and Spain's second-largest bank, BBVA, have been experimenting with DeepSeek's technology along with other AI models, according to people familiar with the banks' efforts.

The release of DeepSeek triggered widespread reassessment of the apparent technological lead U.S. firms such as OpenAI, Google, and Meta Platforms are maintaining in the AI arms race. European banks' willingness to experiment with DeepSeek adds to pressure on AI leaders to compete with the Chinese upstart.

U.S. banks' unwillingness to work with DeepSeek, amid concerns about potential political biases and security flaws, could put them at a global disadvantage. But the current political climate makes partnering with a Chinese AI firm difficult. In February, several states, including New York, banned DeepSeek applications from government devices. The White House is considering a similar move, according to The Wall Street Journal.

"We have an incredibly volatile federal government, and the Chinese-U.S. relationship is very much on tenterhooks," said Alenka Grealish, principal analyst at banking technology firm Celent. "What U.S. bank is going to risk the ire of bad press from the current administration or actual regulatory scrutiny by using DeepSeek? European banks don't have that concern."

Generally, banks put AI tools through an arduous security and governance review process before making the technology available to staff or before customers interact with any systems that leverage it. During the experimentation phase, banks test the AI tools in a secure environment where the institutions control the data on which an AI model is trained. Companies using DeepSeek's open-source models can store data on whatever servers they choose (in contrast, the personal data of people using DeepSeek's app are stored on the company's servers in China).

Banks are hoping AI tools can address some of the issues their older computer systems create. Overhauling those systems can be expensive, and DeepSeek's low cost can offset those expenses. To be sure, big U.S. firms such as Google and OpenAI have responded to DeepSeek by releasing lower-cost new AI models.

"If I were a generative AI leader at a U.S. bank, I certainly would be examining DeepSeek," Grealish said. "The cost of models is a huge variable on what a bank can do with generative AI, and the current political environment will pass."

DeepSeek, which offers open-source models backed by Chinese hedge fund High-Flyer, purports to match the sophistication of U.S. models at a lower cost. According to S&P Global Ratings, at least 19 Chinese lenders are already seeking to use DeepSeek to drive efficiencies.

So too, are some of Europe's leading banks. London-based NatWest, which has experimented with building its own proprietary AI, has been toying with DeepSeek's technology through cloud-based machine-learning platforms. It has researched the way DeepSeek has managed to create smaller, more efficient versions of its flagship model, R1—a process known as distilling—people familiar with the team's work said. NatWest declined to comment.

HSBC, which is headquartered in London but has extensive operations in China, has also experimented with DeepSeek's generative AI models, according to a person familiar with the matter. HSBC's testing of AI tools and products at the early proof-of-concept stage occurs in a tightly controlled environment within its AI center, before the bank gives a small number of staff access in the event of a subsequent pilot test. HSBC declined to comment.

While BBVA focuses mainly on OpenAI's Chat GPT and Google's Gemini, its team in recent weeks has been experimenting with DeepSeek, according to people familiar with the bank's work. "We are currently analyzing DeepSeek-R1, along with many other AI tools, to assess its features and potential benefits," a BBVA spokesperson said. "However, as of today, there are no plans to deploy or adopt it within the organization." The BBVA spokesperson added that any AI models it includes in its suite need to comply with European privacy laws and regulations.

Many large U.S. banks have key partnerships with prominent U.S. AI firms. OpenAI powers Morgan Stanley's tools for helping financial advisers take notes and provide feedback to clients. BNY last month announced a new partnership with OpenAI that will boost the bank's generative AI platform, Eliza. Goldman Sachs' generative AI platform leverages models such as ChatGPT, Gemini, and Meta's Llama. And while Goldman Sachs Chief Information Officer Marco Argenti told The Information earlier this year that his team was interested in using DeepSeek, the Chinese model is not currently on the list of prospective models the bank is examining for inclusion on its platform, according to a person with knowledge of its AI platform. A person briefed on JPMorgan Chase's suite of LLM tools said the bank is not using DeepSeek.

Michael Roddan is a reporter at The Information based in New York covering banking and financial services. He can be reached at michael@theinformation.com or on Twitter at @michaelroddan. You can also message him on any encrypted app at +1 347 864 0601.