Medical Data in France: Why It's the Oil of the Next Decade

Summary

Text Link

Health and AI

Medical Data in France: Why It's the Oil of the Next Decade

France sits on a medical data goldmine. Without proper structure, it's fueling someone else's AI revolution.

Guillaume Gabriel Community Manager

Updated on

2/4/2026

The Essentials in 30 Seconds

Question	Short Answer	Key Takeaway
Are medical data really comparable to oil?	Only partially	Unlike oil, medical data doesn't run out — but it loses all value if it remains unstructured.
Does France have an advantage in health data?	Yes, but theoretical	The SNDS is one of the world's largest data reserves, but most of its value remains unexploitable as-is.
What is the main obstacle to medical AI in France?	Structure, not volume	Legacy EHRs generate unstructured data — free text, PDFs — incompatible with AI training.
What is the FHIR standard and why does it matter?	Healthcare's common language	HL7 FHIR (Fast Healthcare Interoperability Resources) is the interoperability standard mandated by the ANS for new EHRs in France.
Who risks capturing the value created by this data?	Foreign tech giants	Without proper infrastructure, French hospitals become raw material suppliers for players who will resell the intelligence produced.
What is Galeon's approach on this subject?	Structuring data at the source	Galeon structures data directly at the point of generation, across 19 hospitals, so that value stays within the French hospital ecosystem.
What does France risk if it fails to act this decade?	Digital dependency in healthcare	Losing control of the data means losing control of national care strategy — and funding someone else's AI.
Are there limits to exploiting medical data?	Yes, several	GDPR, patient consent, re-identification risks and institutional resistance still hinder the valorisation of this data.

Introduction

France is sitting on a treasure. The SNDS — Système National des Données de Santé (National Health Data System) — consolidates decades of care pathways for more than 70 million people. It is one of the richest medical data reserves in the world. And yet French research remains dependent on foreign players for its most ambitious AI models.

The paradox is stark: we have the data. We don't have the infrastructure to extract its value. Volume is an illusion — structure is the reality. And this difference, technical in appearance, is in reality a national sovereignty issue.

Galeon understood this before the official reports did. Deployed across 19 hospitals, covering more than 3 million patient records and supporting over 10,000 healthcare professionals daily, Galeon is building what France has been waiting for: an infrastructure for structured, interoperable and sovereign medical data.

This article explores why health data is the oil of the next decade — and above all, why France is still letting it burn in place.

Why Do We Compare Medical Data to Oil?

The comparison is seductive. Like oil in the 1920s, health data is abundant for some, scarce for others, and nearly everyone agrees it's going to "change everything." Big tech companies — Google Health, Microsoft, Amazon Web Services — are investing billions to take control of it. States are enacting regulations to protect access to it.

But the metaphor has its limits, and they matter.

Oil runs out. Medical data, on the other hand, renews itself constantly — with every consultation, every hospitalisation, every lab test. The issue isn't scarcity. The issue is siloing: thousands of compartmentalised databases, in incompatible formats, produced by systems that don't communicate.

The truly scarce resource is not raw data. It is structured, interoperable data that is ready for artificial intelligence.

What Is the Real State of Health Data in France in 2026?

The SNDS: A Giant with Feet of Clay

The National Health Data System is, on paper, an exceptional infrastructure. It consolidates reimbursement data from the French national health insurance, hospital data from the PMSI (Programme de Médicalisation des Systèmes d'Information — the hospital activity database) and medical causes of death. For epidemiological research or pharmaco-economic studies, it is an unrivalled resource in Europe.

But the SNDS was designed for reimbursement, not for AI. Its data is structured according to an administrative logic, not a clinical one. Biological data, imaging reports, clinicians' observations — the essence of what makes a care pathway — are largely absent.

Hospital EHRs: Most Value Lost Before AI Touches It

In the vast majority of French hospitals, the EHR (Electronic Health Record, or DPI — Dossier Patient Informatisé) is still a transcription tool, not a structuring one. Physicians enter free-text reports. Prescriptions are scanned as PDFs. Lab results are imported from third-party systems without semantic harmonisation.

The result: data that is theoretically available, but practically unusable by an AI algorithm without costly, time-consuming cleaning work.

According to the Court of Auditors report published in January 2025, French hospitals devote on average only 1.7% of their operating budget to digital — compared to 9% in the banking sector. This chronic underinvestment explains both the fragility of hospital information systems and their inability to produce AI-ready data.

Most of the value of medical data is lost before any algorithm ever touches it. This is not a quantity problem. It is an architecture problem.

Why Does France Risk Becoming a Digital Colony in Healthcare?

The Dependency Pattern

Something silent is happening in French hospitals. Companies — often American, sometimes Chinese — offer "turnkey" AI solutions for radiology, biology, readmission prediction. Institutions sign. Data flows. Models train.

And the value created — patents, algorithms, epidemiological insights — stays with the foreign vendor.

French hospitals, lacking their own infrastructure, become raw material suppliers. They produce the resource. Others extract the intelligence. And they will buy that intelligence back in the form of licences, forever.

The CLOUD Act: A Concrete Threat to Sovereignty

The American CLOUD Act (Clarifying Lawful Overseas Use of Data Act) compels US companies to provide data to American justice, regardless of the country where that data is physically stored. Hosting French medical data with an American provider — even one that is HDS-certified (Hébergeur de Données de Santé — the French health data hosting certification) — structurally exposes it to this legal risk.

This is not theoretical. In February 2026, the French government announced that the Health Data Hub — the national platform for health data — would leave Microsoft and migrate to a European cloud provider qualified under the SecNumCloud standard defined by ANSSI (the French national cybersecurity agency). A historic decision that validates the core argument of this article.

Sovereignty is not decreed in reports. It is coded into interoperability and hosting choices.

How Do You Transform Raw Data into Medical Intelligence?

From Big Data to Smart Data: The Structuring Challenge

The idea that "more data = more AI" is false. An AI model trained on heterogeneous, poorly labelled and semantically incoherent data produces unreliable — even dangerous — results in clinical settings.

The real transformation requires three levels of structuring:

Syntactic interoperability: systems use a common exchange format (HL7 FHIR — Fast Healthcare Interoperability Resources).
Semantic interoperability: medical concepts are codified according to shared nomenclatures (SNOMED CT, LOINC for biology, DICOM for imaging).
Consent traceability: every use of patient data is timestamped, audited and reversible.

Without these three levels, "Big Data" in healthcare remains an empty promise. With them, it becomes "Smart Data" — exploitable by algorithms from the moment of generation.

The Native-AI Approach: Structuring at the Source

The paradigm shift Galeon proposes is simple to state, difficult to implement: do not structure data after its creation, but at the moment of its creation.

In Galeon's partner hospitals, clinicians document within an EHR designed to directly produce structured, codified and interoperable data. No post-hoc cleaning. No costly transformation. Value is captured at the source — and it stays in the hospital.

This is what Galeon calls the Native-AI principle: every clinician-system interaction produces data that is immediately exploitable by an artificial intelligence algorithm.

Galeon vs. Traditional EHR: A Comparison

Galeon vs. Traditional EHR

Criterion	Traditional EHR (legacy approach)	Galeon Approach (Native-AI + BSL®)
Data format	Free text, PDFs, proprietary formats — barely usable by algorithms.	Natively structured according to HL7 FHIR and SNOMED CT / LOINC nomenclatures.
Interoperability between hospitals	Non-existent or manual — each institution remains a data island.	Ensured by the inter-hospital blockchain (BSL®) without transferring raw data.
AI exploitation	Requires lengthy, costly cleaning before any model training.	Immediate — data ready for training from the moment of clinical input (Native-AI).
Data ownership	Unclear — often contractually shared with the solution vendor.	Data stays on hospital servers at all times. Never centralised.
Digital sovereignty	Exposed to CLOUD Act risk if hosted by a US provider.	Decentralised HDS architecture, no central third party, no extraterritorial exposure.
Value sharing	Benefit captured by the AI solution vendor — hospitals bear the cost, not the gain.	Redistribution to hospitals via the $GALEON token (40% to contributing institutions).
GDPR / HDS compliance	Variable by vendor — patient consent is rarely tracked at a granular level.	Patient consent tracked, timestamped and reversible by design.
Research potential	Limited — heterogeneous data, difficult to aggregate across institutions.	High — access to 3M+ structured records, multi-site, trainable without exposing raw data.
Impact on clinical workload	Often time-consuming — re-entry, duplicates, poor UX.	Reduction of administrative tasks through native automation and coding assistance.
Economic model for hospitals	Pure cost — vendor subscription with no return on the value of data produced.	Revenue generator — hospitals monetise their data via the BSL® and the $GALEON token.

What Are the Real Limits and Challenges of This Transition?

1. The GDPR: A Legitimate Constraint, Not an Excuse

The General Data Protection Regulation strictly governs the use of medical data — and that is a good thing. Any use for research purposes or AI training requires either explicit patient consent or a specific legal framework (CNIL authorisation, reference methodology MR-004, etc.). This regulatory complexity slows projects — and it won't disappear.

2. Institutional Resistance: An Underestimated Factor

Hospital CIOs face real budget constraints and information systems sometimes dating back 20 years. Migrating to a structuring EHR is a multi-year project, not a matter of months. Clinician buy-in — who must change their documentation habits — is a human challenge as much as a technical one.

3. The Re-identification Risk

Even anonymised, medical data carries re-identification risks when cross-referenced with other sources. Advanced pseudonymisation techniques and architectures such as Galeon's Blockchain Swarm Learning® (BSL®) reduce this risk — but do not eliminate it entirely. Vigilance remains essential.

4. Territorial Inequality in Data Access

Private institutions, mid-sized clinics and rural hospitals produce equally valuable data — but often remain outside national structuring initiatives. An effective health data policy must address this divide, at the risk of creating systemic biases in the AI models trained on this data.

5. Algorithm Governance: Who Controls What AI Decides?

Structuring data and training models is only the first step. The question of algorithm governance — who validates, audits and challenges them — remains largely open in France, despite the work of the HAS (Haute Autorité de Santé) on AI-based medical devices (DM-IA).

FAQ: Medical Data and AI in France

Can French medical data be used to train AI?Yes, under strict conditions. French law allows health data to be used for research, study or evaluation purposes — provided the relevant CNIL procedures are followed and appropriate authorisations obtained. Hospitals that structure their data upstream have a significant advantage in accessing these uses legally.

What is the SNDS and why is it insufficient for medical AI?The SNDS consolidates health insurance reimbursement data for more than 70 million people. It is valuable for epidemiology and health policy evaluation. However, it does not contain detailed biological data, clinical observations, or medical images — making it insufficient to train precision clinical AI models.

What is HL7 FHIR and why is it the reference standard?HL7 FHIR (Fast Healthcare Interoperability Resources) is the international standard for health data exchange. It defines a common format — structured "resources" — that allows different information systems to communicate without manual transformation. It is now mandated by the ANS as a prerequisite for new EHRs in France.

What is Galeon's Blockchain Swarm Learning® (BSL®)?The BSL® is Galeon's proprietary architecture for training medical AI in a decentralised manner. Data stays on each hospital's servers — it does not circulate. AI algorithms "travel" from node to node via the blockchain to train locally, without ever exposing raw data. The blockchain tracks each use and enables value redistribution to each contributing institution.

How can hospitals monetise their health data without violating patient privacy?Galeon's model is based on explicit, traceable patient consent. Each patient can choose whether their anonymised data participates in research, can withdraw consent at any time, and benefits — via the DAO governance system — from an indirect redistribution of the value created. Monetisation does not involve selling raw data, but making models trained on that data available.

Can France catch up with the United States or China in medical AI?Yes — provided it acts now on structuring, not volume. The US has players like Epic or Google Health that have been structuring data for years. China has opted for massive centralisation. The French path — sovereign, ethical, decentralised — is technically viable. But every year lost widens a gap that becomes harder to close.

Conclusion: The Battle of the Decade Has Already Begun

The next decade in healthcare will not be won by those with the best algorithms. It will be won by those who master the entire data lifecycle — from generation to exploitation, through governance.

France holds an exceptional reserve and a tradition of data protection that can become a global competitive advantage — provided it stops treating structuring as a technical detail. It is a national strategic choice. Hospitals that structure their data today — with standards like HL7 FHIR, decentralised architectures like Galeon's BSL®, and robust patient consent frameworks — will be the hospitals that attract research investment tomorrow.

Galeon, deployed across 19 hospitals and anchored in more than 3 million patient records, demonstrates that the equation is solvable: data sovereignty, value redistributed to clinicians and institutions, AI trained without ever exposing a single raw data point. This is not a prototype. It is an operational infrastructure.

In healthcare, data is a treasure. Let's not let it become an archive.

Would you like to know more about our Smart EHR ?

Book a demo

Sources

‍

Medical Data in France: Why It's the Oil of the Next Decade

The Essentials in 30 Seconds

Introduction

Why Do We Compare Medical Data to Oil?