Sherpa.ai Raises $18 Million for AI That Keeps the Data in Place

Sherpa.ai raised $18 million on July 6, 2026 to keep building federated learning infrastructure for AI that does not move the data out of its owner’s domain. The Spanish deep-tech company frames the work as the practical core of data sovereignty, with the platform sold as a way for governments and enterprises to train models together without sharing sensitive information. The round landed alongside a customer roster that now spans US federal health research, one of Spain’s largest IT contractors, and a pair of Spanish retail banks.

The investor line-up mixes a Silicon Valley cybersecurity specialist with a state-backed Spanish digital transformation fund. Forgepoint Capital, described in the funding announcement as a firm focused on cybersecurity and artificial intelligence, joined the cap table as a new investor, while existing investors Mundi Ventures, Ekarpen, Allegra Holdings and SETT all participated. Sherpa.ai plans to spend the proceeds across five verticals (healthcare, finance, industry, cybersecurity and the public sector) and add new platform capabilities through the rest of 2026.

Sherpa.ai Closes $18 Million for an AI Model That Refuses to Move the Data

The $18 million round, announced on July 6, 2026, is the company’s largest public raise by some distance from its $8.5 million round in 2021. The technology at the centre is federated learning, a way of training models that keeps the underlying data inside its owner’s domain and ships only model updates between participants. Sherpa.ai sells the approach as a way to use AI that complies with Europe’s GDPR and the EU AI Act rather than fighting them. The framing puts the company squarely inside the sovereign AI conversation that European governments and procurement officers have been treating as strategic.

This round allows us to accelerate our vision: to develop and commercialise a secure and scalable artificial intelligence platform that enables companies and governments to harness the full potential of AI without giving up control, privacy and sovereignty over their data.

That is founder and CEO Xabi Uribe-Etxebarria, in a statement attached to the funding announcement. The five named investors co-signed the wager without disclosing valuation, lead-investor terms or post-money structure. The bet, in short, is that distributed AI, with the model moving to the data rather than the data moving to a central server, will outcompete the centralised approach in the regulated corners of the market. Forgepoint’s own website describes the firm as a multi-stage venture capital firm that invests in transformative companies protecting the digital future.

Where the Platform Is Already Running

What looks like a six-name customer roster reads, on closer reading, as a stress test of the privacy thesis. Sherpa.ai says the contracts span sectors where data rules are tightest. The mapping, in the company’s own framing, is six organisations across five target verticals.

Each contract lands in a place where data residency and audit obligations sit ahead of model accuracy on paper. Indra, one of Spain’s largest IT contractors, brings defence-adjacent public-sector deployments. Caja Laboral and Unicaja operate as retail banks under European banking supervision, with reporting obligations that complicate centralised training. Prosegur runs a global physical-security business whose telemetry is classified by default. Centogene Genomics handles rare-disease genetic data under both GDPR and US rules at once.

The most public of the recent wins is the US National Institutes of Health deal, signed in 2022. The agreement put Sherpa.ai’s federated learning system inside NIH’s rare-disease diagnostic work, a project now written up with University College London. NIH senior investigator Carsten G. Bönnemann said the work “could make it possible to explore new diagnostic and treatment possibilities for a group of diseases that currently have no specific treatment,” in remarks posted by Sherpa.ai. The technical setup is detailed in the NIH rare-disease collaboration on the company’s blog.

• Indra – defence-adjacent IT and public-sector systems
• NIH (US National Institutes of Health) – rare-disease diagnostic research
• Centogene Genomics – rare-disease genetic data
• Caja Laboral – Spanish retail bank
• Unicaja – Spanish retail bank
• Prosegur – global physical-security telemetry

A Cap Table That Pairs a Silicon Valley Specialist With a Spanish State Fund

The investor mix is the unusual part of the round. Forgepoint Capital joined as a new investor, the first disclosed US specialist backer on Sherpa.ai’s cap table. The other four participants are existing backers who have supported the company through earlier rounds. SETT, the Sociedad Española para la Transformación Tecnológica (also known as Sepi digital), is attached to the Spanish Ministry for Digital Transformation and Public Function. The presence of both a US sector specialist and a public-sector vehicle on the same cap table is the clearest sign yet that data sovereignty has stopped being a niche procurement concern in Europe. European AI procurement is now asking who owns the model, who audits it, and where the data sits when the model trains.

Mundi Ventures, Ekarpen and Allegra Holdings are the three Spanish investors who returned for the round. According to a database entry tracking the round on Dealroom, the raise ranks in the top 1% by size across a sample of 160 all-time early-stage VC rounds for AI companies in Spain. Per the $18M funding announcement, the deal closes a 2021 plan: Sherpa.ai had then raised $8.5 million framed as a bridge to a larger B-round of €16 million to €25 million.

The Three Papers Sherpa.ai Puts in Front of Regulators

Sherpa.ai publishes more peer-reviewed research than most Spanish startups, and the company is leaning on that record in the funding round. The most prominent recent paper is “Towards the Next Frontier of LLMs, Training on Private Data: A Cross-Domain Benchmark for Federated Fine-Tuning,” published on May 18, 2026 according to the company’s research page. The paper makes the case that the next gains in large language models will come from training on private, distributed datasets that cannot legally be centralised. The work is positioned as a road map for sectors where patient records, banking transactions and industrial telemetry carry the most weight, and where they cannot legally leave the institution that holds them.

The medical-AI line gets its own paper. “Training Together, Diagnosing Better: Federated Learning as a New Paradigm for Medical AI” was published on December 19, 2025, jointly between the National Institutes of Health, University College London and Sherpa.ai. The paper reports an F1-score of 0.82 on the federated model applied to collagen VI-related dystrophies, compared to isolated institutional models. That gap is the case Sherpa.ai is making to clinical buyers who have already said no to centralised training.

The third strand is engineering efficiency. Sherpa.ai’s published research on Blind Federated Learning describes a distributed-training method that reduces communication requirements by up to 99 per cent, the figure the company cites when discussing the work. Less data exchanged between parties means lower energy cost, lower bandwidth and a smaller attack surface. The three papers sit on Sherpa.ai’s published research page alongside a multi-party entity alignment paper from April 2026 and earlier work. The multi-party alignment paper addresses one of the hardest problems in federated AI, which is how to combine data across organisations without ever exposing the identifiers that link the records. Together, the three papers are the proof stack Sherpa.ai offers regulators and procurement officers, alongside the platform code itself.

• May 18, 2026 – “Towards the Next Frontier of LLMs, Training on Private Data,” positioning private distributed data as the next frontier for LLM gains.
• December 19, 2025 – “Training Together, Diagnosing Better,” a joint paper with NIH and University College London on collagen VI-related dystrophies.
• Blind Federated Learning, reporting communication reductions of up to 99 per cent during distributed training.

The Silicon Valley CVs on a Spanish Cap Table

Sherpa.ai has also been quietly assembling a senior team that helps explain why a Silicon Valley firm was willing to lead. Tom Gruber, the co-founder and former chief technology officer of Siri, joined Sherpa.ai as Chief AI Strategy Officer. Two other former Apple executives are also on the team: Doug Solomon, the company’s former head of strategy, and Joanna Hoffman, a former marketing executive.

The 2021 round had already pulled in marquee US names. Marcelo Gigliani, then managing partner at Apax Digital, and Alex Cruz, then president of British Airways, were among the named investors in the $8.5 million round. According to the round’s Spanish-language coverage on Cinco Días, Thomas Kalil, a former White House director of science and technology policy, is also on the team roster. The accumulation of US tech and policy veterans positions Sherpa.ai to sell into both European public-sector buyers and US enterprise customers at the same time.

• Tom Gruber, co-founder and former CTO of Siri, joined as Chief AI Strategy Officer
• Doug Solomon, former head of strategy at Apple
• Joanna Hoffman, former marketing executive at Apple
• Thomas Kalil, former White House director of science and technology policy
• Marcelo Gigliani (then Apax Digital) and Alex Cruz (then British Airways) on the 2021 investor list

What $18 Million Buys Through the Rest of 2026

Sherpa.ai says it will use the round to expand its platform capabilities through the rest of 2026. The work includes new features for enterprise and public-sector users, the explicit framing in the funding announcement. The verticals already named are healthcare, finance, industry, cybersecurity and the public sector. International expansion, beyond Spain, is the second use of proceeds, again across the same five verticals per the funding announcement.

The platform is sold as a way to train, deploy and operate AI models collaboratively without sharing sensitive information. So is the positioning that fits the SETT mandate, which is governments looking to reduce dependence on foreign AI infrastructure. The Sherpa.ai platform overview on the company’s website lays out the product as a SaaS for collaborative AI model trainings across enterprises, governments and regulated environments.

The Spanish public-sector mandate is also why the round is structured as it is. The funding is split between platform expansion (one workstream) and international expansion (a second), per the funding announcement. Sherpa.ai’s own framing puts the platform between the customer and a foreign hyperscaler, a positioning consistent with both SETT and Forgepoint. The customer list, cap table and research record now point in the same direction, and the data sovereignty wager is the bet visible across all three.