Spain's Sherpa.ai secures $18 million for data-sovereign artificial intelligence.

      Everyone is seeking the power of AI, yet few are willing to share their most sensitive data with a foreign cloud to obtain it. A Basque startup believes it has a solution.

      Sherpa.ai has secured $18 million to develop AI that never accesses your raw data, according to tech.eu. This Spanish company caters to banks, hospitals, and governments, which are particularly cautious about data security. Their approach aligns with the rapidly growing concept of sovereign AI.

      Training without data sharing

      At the core of their method is federated learning. Rather than centralizing data for model training, the model moves to where the data resides. Each hospital or bank trains the model locally and then shares only the insights gained, not the actual data. Sherpa.ai claims this approach can reduce data transfers between locations by up to 99 percent.

      This is particularly important in regulated sectors where privacy regulations hinder conventional AI initiatives. Founder and CEO Xabi Uribe-Etxebarria stated that the aim is to enable companies to "harness the full potential of AI without relinquishing control, privacy, and sovereignty over their data."

      A European proposal with American support

      Forgepoint Capital, a Silicon Valley investor focused on cyber and AI, participated in the funding round. Existing investors Mundi Ventures, Ekarpen, Allegra Holdings, and SETT were also involved. This represents a distinct American endorsement for a company rooted in the very concerns that are prompting Europe to retain AI capabilities domestically.

      The client roster reflects this trend. Sherpa.ai recently signed agreements with Spain’s Indra, the banks Caja Laboral and Unicaja, the security firm Prosegur, the genomics company Centogene, and the US National Institutes of Health. A privacy-centric European firm engaging with a US federal agency serves as a strong validation of its approach.

      Credibility behind the term

      The term “sovereign AI” is widely used, and federated learning is not a novel concept. What elevates Sherpa.ai’s version is the research backing it. The company has published peer-reviewed studies on training large language models using private datasets and collaborated with the NIH and University College London to apply this method for rare disease diagnoses.

      The funding amount is modest, and the field is competitive. Numerous companies now offer AI solutions that respect data boundaries, from national model initiatives to privacy-conscious security startups. However, the demand is genuine and on the rise. As governments implement increasingly stringent regulations regarding data location, the company capable of training models without direct access to the data has a compelling narrative.

      Sherpa.ai is wagering $18 million that this story resonates.