From GPU to Construction Site: Sovereign AI in the French Building Industry

Recap of the "Sovereignty & AI in Secure Cloud" roundtable at BIM World 2025

At a trade show dominated by data and digital models, this roundtable opened the hood on sovereign AI for construction. As Annalisa De Maestri from Bouygues Immobilier emphasized, "Sovereignty and AI aren't just trends. They're commitments. Values that we choose to embody—or not—in our strategies."

LinkedIn posts following the event confirm its impact: the debate is no longer technical but strategic, reaching all the way to executive committees. This meeting, orchestrated by Cyriaque Rios, CEO of Resolving, brought together key industry players around "Octave's plan for French Sovereignty & AI."

OVHcloud's Take on Sovereignty – The Cloud Must Go Where Data Lives

Octave Klaba from OVHcloud clarified this often misunderstood concept: "Sovereignty is really about freedom of choice." Far from nationalistic withdrawal, it's about guaranteeing companies control over their technologies, data, and infrastructure.

OVHcloud develops solutions for sensitive data that cannot leave data centers. Facing the resource inequality between large banks and SMEs, they offer fully automated cloud solutions with all necessary security features.

Their major innovation: a cloud-in-a-box (¼ rack, 7 servers) that condenses 25 years of OVH automation. It has the same API as a hyperscaler but sits directly in the factory or client headquarters. This enables 4K video ingestion for industrial vision without transatlantic round trips.

Resolving's Use Cases – From POC to Production

Cyriaque Rios from Resolving shared several practical AI applications in construction:

• Meeting transcription and summarization

• Construction site vision for quantifying progress

• Image generation to create architectural perspective variants for Bouygues Immobilier

• Resolving Process AI, which enables on-the-fly creation of multiple use cases

• Resolving Context, their new global corporate approach

The major challenge: deploying these building blocks into 24/7 production without sacrificing sovereignty or exploding GPU costs. His vision boils down to: "We used to be rivals with Finalcad; today we combine our components so real estate and construction professionals gain productivity, not complexity. That's what operational efficiency means."

These applications raise confidentiality concerns, addressed through deployment on private servers. Resolving maintains a strong commitment to "federating a hybrid secure cloud + on-premise ecosystem" because "strategic autonomy begins with concrete choices."

Bouygues Immobilier's Strategic Approach – Industrialize or Die

Annalisa De Maestri stressed the importance of a thoughtful strategy to avoid "the POC graveyard." She asserts that "data and AI aren't trends. They're values" that should structure company trajectories.

Her method is clear: apply an ROI × security × scalability grid. If a POC scores less than 8/10, it heads to the graveyard. This approach has validated high-impact use cases:

• Automatic extraction of supplier quotes

• Salesforce copilot that summarizes customer history in two scrolls

For Annalisa, the digital ecosystem is reshaping around several axes: cloud frugality, open source, credibility, economic and environmental performance, and democratization of technologies. All of this is anchored in European sovereignty "without which there can be no trust."

Bureau Veritas – Reliability Above All

Céline Labrune from Bureau Veritas Construction highlighted two major challenges in a context where technical control tolerates neither hallucination nor regulatory ambiguity:

1. Information reliability: Facing the hallucination phenomenon in generative AI, Bureau Veritas set up an internal LLM testing bench on fire/structure corpus, a closed sandbox, and prompt training for their 3,000 engineers.

2. Data security: Bureau Veritas handles sensitive data and implements rigorous policies to ensure security.

At the show, Bureau Veritas also presented concrete applications like inspecting a "door of the future" at the Resolving booth. This proved that AI can validate atypical structures if rigorously audited. Result: document analysis time reduced by two-thirds, with demonstrable compliance to external auditors.

Equans Digital's Industrial AI Projects – The Architect Between IT and OT

Raphaël Contamin from Equans Digital presented three projects illustrating AI integration in industrial processes. His compass: "Putting digital technologies at the service of infrastructure performance."

1. The "Pallet Tetris" - A patented AI that optimizes furniture arrangement on pallets, reducing required trucks by 15%

2. Watermark defect detection for the National Printing Office, replacing manual inspection

3. Network architecture for battery gigafactories, with ML workloads at the edge and ERP in private cloud

These projects reflect current challenges where sovereignty and cybersecurity rise to the top of client priorities. Raphaël's mantra: define the architecture (edge/private/public) before the first line of code. As he elegantly summarizes: "Feet in the cables, head in the cloud!"

Our Vision at There – Frugality, Open Source, Ultra-Fast

After 10 years in construction digitalization with FinalCAD, I observed professionals spending much of their time on Office 365. They swing between rigid specialized tools and the freedom of the blank page.

Our approach at There starts with an observation: "Digitalizing business processes one by one takes too long." We've imagined a solution unifying three simple editors: Notes, Word processing, and Email.

AI agents navigate between these interfaces and automate complex tasks. This frees users from ERP and CRM complexity. The user stays in their natural flow while AI handles background systems.

Our vision revolves around three pillars that resonated with the audience. Arthur Montessuit even wondered if they "opened the way to a serene and sustainable use of AI in our professions":

1. Pay-per-result: We adapt AI resources to specific needs and charge per successful AI task, not per user. In a context where license prices double, this approach helps break free from software dependency and pay for real impact.

2. Orchestrated open source LLMs: My experience convinced me of transparency's importance. We use Mistral for heavy tasks and compact local models - DeepSeek - for simple corrections. This reduces GPU costs while ensuring transparency.

3. Agents closest to the field: Our AI agents unify notes, docs, and emails in a single editor, then push data to ERP/BIM/CRM systems. This eliminates user friction and accelerates the observation → decision loop.

The missing link with Resolving becomes apparent: There structures informal knowledge captured in writing, while Resolving sends it back to 3D and planning to multiply field value.

Conclusion

The roundtable concluded with a call for collective action. As Cyriaque Rios emphasized, "sovereignty will be teamwork" among all actors.

Our vision at There can be summarized as: "Speed! Frugal, affordable, pay-per-result. Full open source. Local and global in permanent mix."

As Arthur Montessuit noted: "AI will only be sustainable if it relies on responsible technological choices. And that starts now."

This conference demonstrated that the future of AI in French construction rests on a delicate balance between freedom and structure. Our ambition at There is modest but clear: allow professionals to focus on creation rather than bureaucracy.

This article was written following the "Sovereignty & AI in Secure Cloud" roundtable at BIM World 2025, with Octave Klaba (OVHcloud), Cyriaque Rios (Resolving), Annalisa De Maestri (Bouygues Immobilier), Céline Labrune (Bureau Veritas), Raphaël Contamin (Equans Digital) and yours truly for There.