Franco-German AI Report, April 17: Three Action Items DACH Mid-Market Must Complete Before Summer

On 17 April 2026, the working groups of the Franco-German AI Executives’ Dialogue submitted their final report to the French and German governments. The report, backed by Inria, IMT and Fraunhofer, identifies three industrial levers for a sovereign AI roadmap in Europe: shared compute, data sovereignty via GAIA-X nodes, and use-case prioritisation along industrial value chains. For DACH mid-market companies, the real question is not what Brussels does with it — but how to translate those recommendations into their own numbers by summer.

What the 17 April report actually demands

What is the Franco-German AI Executives’ Dialogue? The Dialogue is an industry-led initiative to align German-French AI industrial policy, launched in January 2025 in Berlin. It is backed by Inria (France), IMT (France), Fraunhofer (Germany), and executive delegates from the largest industrial companies in both countries. The result is a consensus report, updated three to five times a year, that serves as input for national funding policy and the EU IPCEI AI strategy. The version submitted on 17 April 2026 is the first to include concrete operational recommendations rather than statements of intent.

The report outlines three work packages with direct implications for mid-market companies. First, compute sharing through a joint funding framework for SMEs, granting access to sovereign high-performance infrastructure (AI Gigafactories, Jupiter in Germany, Jules Verne in France). Second, data space connectivity via GAIA-X nodes as the mandatory interface for industrial data ecosystems. Third, use-case prioritisation along defined Industrial AI domains (manufacturing, mobility, energy, health, environment). The details read as technical, but they translate into three concrete questions for mid-market businesses.

Question one: who within your organisation takes on the data owner role for the next GAIA-X integration? Question two: which Industrial AI domain accounts for your core revenue — and which use cases within it qualify for funding? Question three: how do you operationally resolve compute sharing when your own workloads need to move between on-premises, hyperscaler and sovereign data centres? Answering these three questions in writing gives you an operational anchor for the upcoming funding rounds.

Why this document carries more weight than expected

Most federal policy documents rarely reach mid-market companies directly. This report is different because it feeds straight into the IPCEI AI programme. IPCEI AI is the vehicle through which major funding programmes are processed in Germany and France; the state aid rules allow significantly more generous support rates on an IPCEI basis than standard individual grants. In practical terms: any company looking to access eligible projects over the next twelve months should be familiar with the report’s recommendations — and reference them cleanly in its own application structure.

Three Numbers from the Report That Will Stick with the Board

Every solid web architecture can be sketched on a Post-It note. The same holds for an SME AI roadmap: if it doesn’t fit on half a page, the prioritisation is missing.

The Three Operational Action Items in Detail

Action Item 1: Establish a Data Owner Role for GAIA-X Integration

GAIA-X nodes require a clear responsibility structure for each data category. Mid-market companies need to designate a role that governs data ownership and technical connectivity: a Data Owner with a mandate covering production, supply chain, and engineering data — not just IT-managed datasets. Ideally, this role sits at the executive level or directly beneath it as a dedicated mandate. From the first integration projects we’ve observed, companies that formally assign this role are GAIA-X-ready within three months. Companies that “tack it onto IT” are still stuck in data catalogue discussions nine months later.

The operational connection to a GAIA-X node can now be handled through commercial service providers (Teralab, International Data Spaces, Catena-X for mobility). The technical integration is a solved problem — the organisational side is not. A quick reality check: if you can’t name three data categories within your own organisation that are audit-ready and shareable externally, your GAIA-X journey hasn’t started yet.

Action Item 2: Prioritise Your Use-Case Portfolio Across the Five Industrial AI Domains

The report prioritises five domains: manufacturing, mobility, energy, health, and environment. Mid-market companies should map their existing AI use cases into these domains and prioritise the five that carry the greatest leverage within their own business model. That sounds straightforward — but it represents a clean break from the typical use-case discussion, which tends to rank AI projects primarily by technical feasibility. The industrial policy logic is different: value contribution plus strategic alignment with the funding landscape. A use case that fits cleanly into one of the five domains and delivers a genuine efficiency or quality advantage stands a significantly better chance of co-financing in 2026 than a technically elegant but generic optimisation scenario.

Action Item 3: Define a Compute-Sharing Strategy for the Next 24 Months

Compute sharing is the most operational of the three levers — and the most widely misunderstood. The point is not that mid-market companies will gain direct access to a national supercomputing centre, but that subsidised AI workloads are expected to run through defined hubs: Jupiter, Jules Verne, and the announced AI Gigafactories. For your own organisation, this means auditing which workloads are latency-critical (and therefore stay on-premises), which are suited to hyperscalers (and can remain there), and which could realistically migrate to sovereign data centres over time. Companies that make these decisions before summer will be positioned for the first wave of funded pilot programmes.

The compute-sharing strategy is the one that engages leadership most directly. It touches investment decisions, existing hyperscaler contracts, and potential exit strategies. A pragmatic interim step: over the next three months, build a compute inventory that classifies each workload by category — latency-critical, sensitive, or standardisable. That list will prove more valuable in subsidy applications over the next 18 months than any executive summary.

Reality Check: What a Mid-Market Project Actually Looks Like

From our advisory mandates over recent months, three project archetypes emerge that map cleanly onto the report’s framework. Archetype one, “Manufacturing Quality”: a mid-sized mechanical engineering firm sharing predictive quality models with supply chain partners via a GAIA-X connection. Archetype two, “Energy Optimisation”: an energy supplier or industrial operator running load forecasting and storage management models on sovereign infrastructure. Archetype three, “Health Compliance”: a medical technology company sharing training data with research partners in a data-sovereign manner and deploying models in regulated environments with full audit capability. All three archetypes are explicitly prioritised under the current IPCEI-AI framework.

Realistic project timelines from the first workshop to production rollout run between nine and 18 months, depending on the organisation’s data governance maturity. Total project costs typically range from 800,000 to 4 million Euro, with IPCEI funding ratios generally covering 40 to 60 percent. For your own budget planning, that translates to a net investment of 300,000 to 2 million Euro for a pilot project with a clear reference to the report.

One practical stumbling block appears in nearly every project: aligning budget co-financing between a corporate parent (where one exists) and the mid-market subsidiary. Funding eligibility typically covers only the mid-market entity. Setting up clear governance from the outset — a project steering committee with defined decision-making bodies and signing authorities — saves two to three months during the application phase. Ignoring this tends to trigger escalation cycles that the IPCEI timeline penalises heavily.

Interfaces with the EU AI Act and NIS2

The report of April 17, 2026 connects to two regulatory frameworks. First, the EU AI Act: high-risk systems in industrial contexts — safety-critical manufacturing robotics and energy infrastructure — are addressed by the report’s recommendations, which reduces the coordination burden between industrial policy and the competent supervisory authorities. Second, NIS2: data spaces and compute-sharing scenarios are classified as critical infrastructure wherever they become relevant to energy supply or healthcare. The operational consequence: organizations that complete the three homework assignments properly will simultaneously document NIS2 readiness and EU AI Act mappings — without duplicating effort.

A third perspective: what works councils and trade unions want to see

One dimension that has received little attention in public debate around the report: the role of works councils and trade unions in industrial AI implementation. On the German side, involving the works council in AI rollouts is mandatory under the Works Constitution Act (Betriebsverfassungsgesetz); on the French side, equivalent processes run through the CSE (Comité Social et Économique). Mid-sized companies that only bring employee representatives in at the rollout stage regularly find themselves in escalation, with project delays of two to six months. The operational consequence: the data-owner role and the use-case portfolio should be discussed with the works council or CSE in the first project quarter — not the last.

The report addresses this dimension at a meta level (“industrial acceptance”) but provides no model agreements or governance templates. Closing that gap is the task of employer associations and trade unions in the months ahead. For mid-sized companies, this means establishing their own agreements along the lines of DGB/IG Metall models — or their French equivalents from FO/CFDT — before the first funding application is submitted. Sound project management budgets at least four weeks for this and involves HR early. Those who underestimate this dimension may produce an excellent application on paper, only to fail on operational buy-in the moment the approval arrives and implementation begins.

What the report does not deliver

Three criticisms that came up in our conversations with mid-market clients about working with the report. First, concrete funding volumes per mid-sized project are still missing; they will be added in the forthcoming IPCEI calls. Second, the role of SME associations — BDI, BVMW, and MEDEF in France — in implementation is vaguely defined. Third, the IT security dimension is mentioned but not developed. Organizations that keep these three gaps on their radar can integrate the report’s recommendations cleanly into their own roadmap, without relying on promises the report never actually makes.

Frequently Asked Questions

Further Reading on MyBusinessFuture

• Our analysis of AI data maturity in the mid-market and the five prerequisites before deploying production agents
• Analysis of the 80 percent AI failure rate and the underlying structural question
• Report review of the Deloitte State of AI 2026 Enterprise and the execution gap

Quelle Titelbild: Pexels / Dom J (px:355948)