Projects launched in 2026 no longer resemble those of 2024. The gradual implementation of the European AI Act, the emergence of shared AI Gigafactories, and the refocusing of funding on specific industrial problems are changing how a company structures its investments, teams, and roadmap.
AI Act Compliance: Integrating Regulation from Project Design
The transparency obligations of the European AI Act came into effect on August 2, 2026. AI-generated content must become detectable or be appropriately marked, with a grace period until December 2, 2026, for certain systems already on the market. Any company deploying an automation tool, a customer-oriented chatbot, or a recommendation engine must now integrate compliance from the design phase, not at the end of the process.
In practical terms, this means that an automated marketing project or predictive inventory management can no longer be prototyped without a prior regulatory audit. The cost of compliance after delivery far exceeds that of an architecture designed from the outset for the marking and traceability of AI outputs.
We recommend appointing an AI compliance officer in each project team, including in SMEs. This role does not require a full-time lawyer: a team member trained in the requirements of the AI Act, capable of documenting technical choices and verifying transparency obligations, is sufficient in most cases. Several resources compiled on the Flexmind news blog on Daily Breizh detail these obligations by industry sector.
Shared Computing and AI Gigafactories: Concrete Access for SMEs
The European Union launched a call in July 2026 for up to seven AI Gigafactories. These infrastructures are intended to provide startups, SMEs, manufacturers, and laboratories with computing capabilities to train, adjust, and deploy advanced models. This represents a structural change for entrepreneurs who, until now, had to either invest heavily in their own GPU servers or rely entirely on American hyperscalers.
Shared access to computing reduces the entry barrier for AI projects in Europe. An SME in the industrial sector can now consider training a computer vision model for quality control without mobilizing a six-figure budget in cloud infrastructure.
What This Changes for Project Management
The availability of these resources alters the development timeline. Where a machine learning project previously required several months of negotiation with a cloud provider, Gigafactories allow for training to begin on reserved slots. The project schedule must incorporate this slot allocation logic, comparable to shared time on heavy equipment.
We observe that the best-prepared teams are those that have already structured their datasets and defined their performance metrics before requesting computing time. Data quality remains the limiting factor, not raw power.
Phased Funding: The Model of the European Innovation Council
The European Innovation Council (EIC) launched a pilot in September 2026 selecting 20 projects in robotics, health, and industry. The principle: teams must validate their solution, measure its performance, and demonstrate its potential before accessing the next stages of funding.
This milestone-based model is gradually replacing flat grants. For a project leader, this means rethinking the very structure of the business plan:
- Each phase must produce measurable deliverables (functional prototype, test results, field feedback) and not just progress reports
- The budget forecast must be divided into tranches aligned with the milestones, with predefined performance indicators to trigger the next tranche
- The project team must include from the outset a competence in impact measurement, capable of quantifying gains compared to the existing situation
A project without a clear validation metric no longer accesses European funding. This criterion also applies to national schemes that are gradually aligning with the EIC methodology.
AI Training and Skills Development: A Standalone Budget Item
Europe has placed digital skills and AI at the center of its priorities for 2026. Employee training is no longer an optional annual plan: it is a structural investment that conditions an organization’s ability to absorb new tools and regulations.
Target Operational Skills Rather Than General Knowledge
General training on AI (history, concepts, promises) saturates the market. What teams need in 2026 are operational skills: knowing how to audit a model, document a data pipeline, interpret the marking obligations of the AI Act. Useful training is that which produces an actionable deliverable by the following week.
For the construction and industrial sectors, skills development also involves integrating AI into existing processes: assisted planning, anomaly detection on site, logistical optimization. These use cases do not require data scientists but rather team members capable of configuring and supervising pre-trained tools.
Organize Training as a Project
We recommend treating AI training with the same rigor as a client project:
- Define a measurable objective (number of certified employees, number of use cases deployed)
- Allocate a specific budget, distinct from the traditional training budget
- Designate a program manager who tracks results and adjusts content based on field feedback
- Plan short and recurring sessions rather than an annual marathon training
Applying project management to training accelerates the adoption of AI tools. Organizations that treat skills development as a side task accumulate a backlog that is difficult to catch up on, especially when regulatory deadlines impose fixed timelines.
Projects that will succeed in 2026 share a common trait: they treat compliance, access to computing, funding, and training not as side issues, but as structural components of their roadmap. Ignoring any of these four pillars is akin to building on ground whose foundations are not understood.
