- Deployment
- on-prem
- Hq Country
- France
- Hq City
- Paris
- Ownership
- public
- Founded
- 2016
- Training Default
- never
- Residency Options
- customer's own infrastructure (air-gap capable), customer infrastructure with sovereign European cloud GPUs, sovereign European cloud (managed SaaS)
- Gov Access Exposure
- none-eu
- Eu Ai Act Role
- deployer
- DPA available
- yes
- Zero Retention Available
- Yes
- Self Host Option
- Yes
- Retention Default
- In the on-premise deployment both compute and data stay on the customer's infrastructure, so LightOn retains nothing. The corporate privacy policy's retention schedule (6 months for application and connection data) covers LightOn's own website and customer-relationship data, not Paradigm workloads.
- Retention Exceptions
- "Most of the processing directly under the control of LightOn is carried out in France or in the UE." Sub-processor processing "can take place in various locations" under SCCs, Data Privacy Framework or BCRs — but that is the corporate estate, not the on-premise product.
- Subprocessor Count
- 0
- Transfer Mechanism
- intra-eu
- Eu Procurement Ready
- pass
- Eu Procurement Reason
- Europe's only listed generative-AI pure play — Euronext Growth Paris since November 2024 — selling a GenAI platform that deploys on the customer's own servers, air-gap capable, with a hybrid mode that keeps data on-premise while borrowing sovereign European cloud GPUs for inference. Jurisdiction and data control are as good as the rubric allows. Two things hold it below the self-hosted open-weight rows: Paradigm is proprietary, so exit means rebuilding rather than changing a base URL, and there is no certification this pass could verify — for a vendor selling to defence, critical infrastructure and the French public sector, the absence of a published ISO 27001 or SecNumCloud position is conspicuous. The privacy policy is also a corporate one; it says nothing about Paradigm workloads, which in an on-premise product is defensible but leaves the buyer inferring the most important commitment.
- Training Claim Basis
- inferred
- Training Confidence
- medium
- Residency Claim Basis
- stated
- Residency Confidence
- high
- Retention Claim Basis
- inferred
- Retention Confidence
- medium
- Subprocessors Claim Basis
- inferred
- Subprocessors Confidence
- medium
- Residency Class
- customer-controlled
- Portability / Exit Path
- proprietary-only
- Subprocessor Jurisdiction
- structurally-none
- Underlying model providers
- LightOn (proprietary Paradigm models), open-weight models deployed on customer infrastructure
