01 / WHO WE SERVE Back to home

On-premise AI for hospitals and healthcare

Healthcare is the sector where AI is most needed and where data can leave the least. An on-site AI data center brings AI-assisted diagnostics, report dictation and record summarization without patient data ever leaving the hospital.

LOCAL DATA OWN ENERGY OPEN MODELS
Updated September 2026
THE PROBLEM

Why on-site, why off-grid.

  • Health data is special-category data (Art. 9 GDPR): entrusting it to a foreign cloud service requires legal bases, assessments and transfer safeguards.
  • A physician spends on average 13 hours a week on indirect activities such as documentation.
  • AI-assisted diagnostics is already used in 74% of EU states, but the data should not leave the hospital.
THE NUMBERS
996

SSN hospital facilities and 9,121 outpatient facilities

Annuario SSN 2023
€2.7 bn

digital health spending in Italy in 2025, +9%

Osservatori PoliMi
61%survey

of specialists and GPs already use generative AI

Osservatori PoliMi
€15.63 bn

NRRP Mission 6, of which €1.38 bn for the electronic health record

Osservatori PoliMi
74%

of EU states already use AI-assisted diagnostics

OMS Europa · apr 2026
THE RULES
  • health data (Art. 9 GDPR) that must not leave the hospital
  • the European Health Data Space (EHDS, Reg. 2025/327) applies to primary use from 25 March 2027
  • law 132/2025 leaves the final decision to the doctor and requires informing the patient
WHERE ON-SITE AI HELPS
  • AI-assisted reading of exams and screening — in the randomized MASAI trial (Sweden, 105,000 women) AI cut radiologists’ reading workload by 44% and increased detected cancers by 29%
  • dictation and drafting of reports and discharge letters — in a UCLA study of 238 physicians documentation time fell by 9.5%
  • patient record summarization
  • triage of patient messages
WHY BARREL

Sovereign by design.

  • Open medical models (MedGemma, MedASR) and general models with a commercial license, installed on site.
  • Health data that never leaves the hospital perimeter.
  • Own energy: the system keeps working during a blackout.
RECOMMENDED SIZE

D1 for a facility or clinic, D2–D3 for a hospital.

See the product range
NEXT STEP

Size it. Then verify it
with a feasibility study.

06 / FAQ

Questions people ask.

01Can ChatGPT be used with patient data?+

Health data is a special category (Art. 9 GDPR). With a foreign cloud service you need legal bases, assessments and transfer safeguards. An on-site model avoids the transfer, but a DPIA is still required.

02Which AI models can be installed in a hospital?+

MedGemma, MedASR and general open-source models with a commercial license. They are not certified medical devices: they support staff, they do not replace clinical decision-making.

03How much does an AI server for a hospital cost?+

It depends on users and processes. Use the configurator to get the order of magnitude for load, GPUs, models and cost.

04What changes with the EHDS in 2027?+

From 25 March 2027 the EHDS rules on the primary use of electronic health data apply. You need your data organized and under control.