France Government AI
What changes when the state owns the shared AI layer?
Watch the France vs Germany Government AI case study
Watch on YouTube ↗Shared infrastructure moves coordination into the Owner layer.
France's architecture puts common inference, tooling, data capabilities and horizontal applications into a shared state layer. In the PBMC, model supply and secure hosting feed into DINUM, while administrations consume a reusable AI service through SIIAG and Albert API. This can reduce duplicated technical backends and make common security and access rules reusable across administrations, while also concentrating more operational responsibility in the Owner layer.
Structured data behind the canvas
One fixed table for every PBMC. Transaction remains one field; all transaction arrows and their explanations stay inside that field, so cases remain directly comparable.
| Perspective | Field | Value | Explanation |
|---|---|---|---|
| Core Value Unit | Core Value Unit | AI Service | At the center is an AI service: secure access to inference, shared tools, data capabilities or horizontal applications that an administration can actually use. |
| Owner | Actor | DINUM | The Owner is DINUM, short for Direction interministérielle du numérique, France's interministerial digital directorate. It coordinates the state AI strategy, operates the shared layer and animates the government AI community. |
| Owner | Job | Run Layer | Dinum's job is to run a common AI layer that many administrations can reuse. |
| Owner | Gain | Scale / Control | The upside is reuse at scale while keeping control over the gateway, data path and common rules. |
| Owner | Pain | Risk / Load | The trade-off is concentration. A common layer can reduce duplicated attack surfaces, but it also carries more critical load and has to remain available and secure under attack. |
| Owner | Transaction | Route / Serve ← Consumer · DemandMinistries send demand and use cases into Dinum. The shared owner learns what government teams need. ← Provider · ModelsModel vendors contribute model capability into the shared service catalog. ← Partner · ComputeSecure hosting partners contribute certified compute and hosting into the shared AI layer. → Consumer · ServiceDinum returns a usable AI service to administrations through the common layer. | Dinum routes demand to shared model and infrastructure capability and serves AI access back to administrations. |
| Owner | Governance | Security / Rules | Governance is concentrated in common security, interoperability, data protection and service rules. |
| Owner | Promotion Channel | State Network | Dinum grows adoption through the interministerial AI community and shared state programs. |
| Owner | Activities | Operate / Govern | It operates common services, curates access and coordinates adoption across administrations. |
| Owner | Resources | Albert / SIIAG | Its strategic resources are the shared layer itself, especially Albert API, common tooling, data services and horizontal applications. |
| Consumer | Actor | Ministries | The Consumer side is ministries and other state administrations that need AI capability without building the full technical stack themselves. |
| Consumer | Job | Deploy AI | Their job is to put useful AI into administrative work and digital services. |
| Consumer | Gain | Fast / Secure | The gain is faster deployment with a security level that individual teams do not have to recreate from scratch. |
| Consumer | Pain | Dupes / Shadow | The pain is duplicated development and shadow AI: teams solving the same infrastructure problem repeatedly or using tools outside state control. |
| Consumer | Transaction | Demand / Use → Owner · DemandMinistries send demand and use cases into Dinum. The shared owner learns what government teams need. ← Owner · ServiceDinum returns a usable AI service to administrations through the common layer. | Administrations bring use cases and demand into the system, then consume shared AI services. |
| Consumer | Filter | Public | Access is bounded to the public sector rather than an open consumer market. |
| Consumer | Access Channel | SIIAG | Administrations access the value through SIIAG, short for Socle interministériel d'IA générative, the interministerial generative AI foundation. Its products include Albert API and horizontal applications. |
| Consumer | Activities | Use / Deploy | They test, integrate and use AI in their own processes and products. |
| Consumer | Resources | Data / Cases | They contribute the use cases, domain knowledge and data context needed to make AI useful in administration. |
| Provider | Actor | Vendors | The Provider side is the organizations whose foundation and specialist models can be made available through the common layer. Albert currently exposes models from several model families and third parties. |
| Provider | Job | Supply AI | Their ecosystem job is to supply competitive model capability that the shared layer can expose to government users. |
| Provider | Gain | Reach / Usage | The gain is usage through a common government gateway without every application integrating each model separately. |
| Provider | Pain | Fit / Rules | The friction is fitting technical, licensing, security and performance requirements of the shared environment. |
| Provider | Transaction | Models / Update → Owner · ModelsModel vendors contribute model capability into the shared service catalog. | Providers contribute models and model updates into the service catalog. |
| Provider | Filter | Approved | Not every model automatically becomes part of the state catalog. Models are selected and hosted within the controlled service. |
| Provider | Access Channel | Albert API | The common access point is Albert API, which centralizes model access behind one OpenAI-compatible interface. |
| Provider | Activities | Train / Update | Providers develop and update models while Dinum can change what is offered behind the common interface. |
| Provider | Resources | Models / Weights | Their core resource is the model itself, including open weights or proprietary model access where applicable. |
| Partner | Actor | Hosts | The Partner side is secure cloud hosting. The shared layer depends on certified infrastructure that can run inference under state security requirements. |
| Partner | Job | Run Secure | The partner job is to provide secure, resilient compute on which the state AI layer can operate. |
| Partner | Gain | Volume / Trust | The gain is sustained public-sector workload and a trusted infrastructure role. |
| Partner | Pain | Audit / Load | The burden is meeting demanding assurance requirements while keeping enough capacity for shared government workloads. |
| Partner | Transaction | Compute / Host → Owner · ComputeSecure hosting partners contribute certified compute and hosting into the shared AI layer. | The partner contributes sovereign compute and hosting into the common AI layer. |
| Partner | Filter | SecNum | Hosting is filtered by the state security boundary, with SecNumCloud certification central to the current architecture. |
| Partner | Access Channel | Contracts | Hosting partners enter through public-sector infrastructure and service arrangements rather than through an open marketplace. |
| Partner | Activities | Host / Run | They host, operate and maintain the compute environment that supports inference. |
| Partner | Resources | GPUs / Cloud | They bring cloud infrastructure and accelerator capacity. |
Sources
Sources document the platform mechanics and factual case context. The PBMC mapping and Platform Lesson are Platform Generation's analysis.
- 01Socle interministériel d'IA ↗
Primary source for SIIAG as a shared interministerial AI infrastructure, its service layers, public-sector scope and SecNumCloud hosting.
- 02Inférence — SIIAG ↗
Primary source for Albert API as the shared inference layer, its OpenAI-compatible interface, model catalog and SecNumCloud infrastructure.
- 03Albert API ↗
Primary source for the shared AI gateway, centralized model access, advanced services and current public-project usage figures.
- 04Modèles — Albert API ↗
Source for the model-provider side and the range of model families exposed through the common gateway.
- 05Infrastructure sécurisée — Albert API ↗
Source for secure hosting, sovereign infrastructure and the SecNumCloud boundary used to derive the Partner role.
Use it. Cite it. Build on it.
The case-specific analysis and structured data are available under CC BY 4.0. You may cite, copy, analyze and adapt the case materials with attribution.
The Platform Business Model Canvas (PBMC) canvas/template is available under CC BY-SA 4.0. You may reproduce, teach with, use commercially and adapt it, provided clear attribution to the original PBMC remains and adapted versions are shared under the same license.
An unmodified Official PBMC should retain its displayed provenance rather than having Platform Generation attribution removed or replaced. An adapted version may carry your own branding, but the PBMC attribution must remain clearly visible and the adaptation must not be presented as an Official PBMC.
Platform Generation logos and the Official PBMC designation are not included in the Creative Commons licenses and may not be used to imply endorsement by Platform Generation.
BibTeX
@misc{eisape2026francegovernmentaipbmc,
author = {Eisape, Davis Adedayo},
title = {France Government AI --- Platform Business Model Canvas},
howpublished = {Platform Generation PBMC Library},
year = {2026},
month = sep,
url = {https://platformgeneration.com/france-government-ai/},
note = {PBMC snapshot, September 2026. Case-specific analysis/data: CC BY 4.0. PBMC canvas/template: CC BY-SA 4.0 with attribution. Platform Generation logos and Official PBMC designation are excluded from the Creative Commons licenses. Available at: https://platformgeneration.com/france-government-ai/}
}