Week 19
AI Gigafactories: Can You Build Supply Before Demand?
Europe has just crossed a symbolic threshold: 20 billion euros committed to “AI Gigafactories,” a 180-million-euro sovereign cloud contract, and a Tech Sovereignty Package scheduled for May 27.
But a fundamental question persists — raised this week by both European parliamentarians and McKinsey: what good are massive infrastructures if the application ecosystem doesn’t exist to exploit them? According to McKinsey’s latest Global Survey on AI, 60% of organizations have still seen no enterprise-level EBIT impact despite massive adoption. The gap between AI activity and AI impact exists.
The parallel with European strategy is striking. Building large-scale data centers without an ecosystem of companies capable of exploiting them — only Mistral and a handful of players are mentioned by critics — is reproducing at the macro level what McKinsey describes at the micro level: the “pilot trap,” that infinite loop of investing without ever proving value.
Yet the counter-argument exists. The history of major infrastructure shows that supply can precede demand: the highway system, rural electrification, and fiber optics were all built before usage fully materialized.
Because demand, in reality, won’t come from governments: it will come from companies. It’s the business functions — finance, supply chain, customer service, R&D — that must seize this framework to demonstrate, use case by use case, that AI creates measurable value. If I chose to feature McKinsey's framework in this newsletter, it's because it isn't a policy tool — it's a steering instrument for European enterprises, and specifically for the business functions that are the true drivers of AI adoption
Until these business functions answer the framework’s five questions — technical performance, adoption, operational KPIs, strategic outcomes, financial impact — the 20 billion in infrastructure will remain a political bet. The proof of value isn’t decreed in Brussels: it’s built enterprise by enterprise, function by function.
Next week, we’ll offer an in-depth reading grid of the new Omnibus package and its concrete implications for European businesses.
AI & Sovereignty News — May 4 – May 11, 2026
SAP – Prior Labs: Europe’s First Frontier AI Acquisition by a Software Champion
SAP has executed what is, at this stage, one of the most structurally significant operations in the European AI ecosystem: the acquisition of Prior Labs, backed by an investment commitment exceeding 1 billion euros, followed by a $1.16 billion funding round.
The positioning is clear: build a world-class AI research lab specialized in structured data and high-compliance environments — a terrain where SAP holds a natural competitive advantage.
Prior Labs will maintain operational independence, rooted in Germany’s academic ecosystem. The strategic signal is twofold: first, it marks the first time a European software giant controls a frontier AI lab; second, it demonstrates SAP’s moves on sovereign AI — which we’ve been covering in recent weeks.
Source: SAP acquires Prior Labs to build a European frontier AI research lab
AI Omnibus: The EU Recalibrates Its Regulatory Timeline
This week confirmed what we described last week: the trilogue agreement between Parliament, Council, and Commission postpones high-risk AI system obligations to December 2027.
This realignment responds to a pragmatic observation: 30% of European CEOs identify regulatory fragmentation as a major operational constraint on their AI projects. The guiding principle adopted — companies should not be regulated twice — signals a significant pivot toward simplification.
For economic actors, this means two additional years to structure compliance, but also a signal of legal clarity the market had been waiting for.
The challenge will be maintaining regulatory ambition while avoiding falling behind the United States — and particularly the broader latitude US companies enjoy in using Europeans’ personal data. The Digital Omnibus introduces three major changes that de facto expand US platforms’ access to European citizens’ data: a new GDPR Article 88c creating a “legitimate interest” legal basis for AI training on personal data, the removal of pseudonymized data from GDPR scope, and relaxed rules on sensitive data processing during AI development.
Source: EU countries, lawmakers agree to dilute AI rules, delay implementation
Tech Sovereignty Package: The Commission Draws the Contours of a Protected Market
The legislative package expected on May 27 covers three strategic axes: cloud, semiconductors, and AI.
The headline measure — requiring public institutions to process their sensitive data on European sovereign infrastructure — creates de facto structural demand for local providers.
According to CNBC, the Commission also plans to restrict US cloud platforms’ access to government data. This move is consistent with the recent award of a 180-million-euro contract to four European cloud initiatives for EU institutions.
The implication for US hyperscalers is direct: the European public cloud market, until now accessible, could partially close.
Source: EU Moves to Curb Reliance on U.S. Cloud Giants
AI Gigafactories: 20 Billion Euros and the Supply-Demand Question
The Commission has formalized a 20-billion-euro-plus investment plan for large-scale AI data centers, mobilizing public and private capital. The ensuing debate is revealing: parliamentarians and experts point to a substantial overcapacity risk. The structural problem is identified — only a limited number of European players (Mistral foremost) would have the capacity to exploit these massive infrastructures. The criticism goes further: investing in replicating the hyperscaler model without a proprietary application ecosystem amounts to building highways without vehicles. This is the decisive test of Europe’s AI industrial strategy: the legitimacy of the investment rests entirely on the ability to stimulate supply and demand simultaneously.
Source: EU slammed over multi-billion AI infrastructure splurge plan
Germany: 1.7B Euro VC in Q1 2026 — The Financial Dependency Paradox
The German venture capital market hit a record in Q1 2026 with 1.7 billion euros raised, 58% directed toward AI projects. These figures confirm Germany’s centrality in the European AI ecosystem.
However, KfW Research analysis reveals a structural paradox: the bulk of this capital comes from American funds.
Europe is thus funding its innovation ecosystem with capital over which it exercises no strategic control.
Source: Germany attracts record venture capital: AI as a growth driver
Semidynamics – SiPearl: Toward a Fully European AI Compute Stack
The alliance between Spain’s Semidynamics and France’s SiPearl aims to develop a rack-scale AI compute platform entirely designed in Europe, with an open, multi-vendor architecture. The stated objective is explicit: reduce dependence on non-European ecosystems for cloud-scale inference.
This partnership takes on particular significance given that SiPearl already designs the processor for Europe’s EuroHPC supercomputer — the alliance thus fits within a coherent industrial continuity. The value proposition is clear: offer European public and private AI initiatives a credible alternative to NVIDIA-dependent architectures. It’s one of the rare projects addressing sovereignty at the most fundamental level — silicon.
Source: Semidynamics and SiPearl Announce Strategic Cooperation
Cooperative AI: The “Solidarity Stack” as a Systemic Alternative
Trebor Scholz and Mark Esposito, covered by Hubert Guillaud in “Dans les algorithmes,” offer a structuring diagnosis: today’s AI rests on an extractive stack where a few players control hardware, cloud, models, and data without democratic oversight. Their proposal: replace this architecture with a “solidarity stack” — hosting cooperatives, citizen-managed data trusts (MIDATA in Switzerland), and locally developed AI tools built with communities (AI4Coops in Argentina).
Open source models already reach 90% of proprietary system performance according to Raffi Krikorian (Mozilla), making this alternative technically credible. The challenge is political: without legislative and financial support, cooperatives cannot compete with hyperscaler network effects. Initiatives like the Public AI Network, Lestac AI (France), and Apertus (EPFL/EPFZ) point toward a distributed, empowering sovereignty — an essential counterpoint to top-down state strategies.
Source: Pour une IA coopérative — Dans les algorithmes
Study in Focus
McKinsey / QuantumBlack: “From Promise to Impact”
How companies can measure — and realize — the full value of AI
April 2026 | J.-T. Lorenz, J.C. Abraham, R. Levin, D. Ziman
The gap between AI activity and AI impact has reached a critical point. McKinsey’s Global Survey on AI lays out unambiguous figures: 78% of organizations use gen AI in at least one business function, 62% are experimenting with agentic AI — yet 60% have still seen no enterprise-level EBIT impact. The cause is structural: most deployments focus on horizontal tools (chatbots, copilots, summaries) that improve employee experience but don’t move the P&L. Only a minority of organizations automate end-to-end workflows in specific domains — and that’s where value is created.
The 5-layer framework. McKinsey proposes a measurement system linking technical performance to financial impact with a clear chain of accountability:
(1) Financial impact (owner: Finance) — revenue uplift, cost of service reduction, margin expansion, cloud + token TCO;
(2) Strategic outcomes (owner: BU/Strategy) — NPS, customer satisfaction, retention, compliance;
(3) Operational KPIs (owner: Process Owner) — cycle time, defect rate, first-contact resolution, cost per transaction;
(4) Adoption & engagement (owner: Product/Ops) — active users, workflow penetration, acceptance vs. override rate;
(5) Technical performance (owner: Data Science/Engineering) — hallucinations, latency, cost per interaction, drift.
The 3 differentiators. Organizations that escape the “pilot trap” share three practices: they define expected value upfront and connect metrics across layers in a complete causal chain; they embed attribution measurement directly in deployment (A/B testing, staged rollout) so results withstand scrutiny; and they manage AI as an investment with fixed review cadence, explicit stage gates, and a single evidence file combining benefits and total cost of ownership. Only use cases that demonstrate defensible impact progress to scaling.
Source: McKinsey — From promise to impact
Nicolas Bombourg
The C&M Intelligence Leader’s Wine — nicolasbombourg.substack.com