The European Commission tabled its Technological Sovereignty Package on June 3: the Chips Act 2.0, the Cloud and AI Development Act, an Open Source Strategy, and an energy-AI roadmap. The headline ambition is to triple EU data centre capacity within five to seven years and end the bloc's reliance on suppliers who, in one commissioner's phrasing, could hold a "kill switch."
The timing is awkward. The week the Commission announced how much compute Europe should build, reporting confirmed its existing flagship build is stalling. Brussels can write the strategy; the open question this package does not answer is whether it can deliver the infrastructure.
Research & Labs
PhysicsX raises $300M to scale "Large Physics Models"
London-based PhysicsX closed an oversubscribed $300 million Series C on June 8 at a roughly $2.4 billion valuation, led by Temasek with Siemens, Nvidia, and Applied Materials returning. The company builds what it calls Large Physics Models: the LLM analogy applied to the equations governing how engines, turbines, and chips behave under stress.
Founded by two former Formula 1 engineers, PhysicsX has grown from 150 to 350 staff in a year and more than quadrupled revenue over two. Its platform is already deployed across aerospace, semiconductors, automotive, and energy.
The raise is a signal about where European AI has a real edge. Not consumer chatbots, where the frontier gap is wide, but physics-aware engineering tied to the industrial base Europe still owns. (The Next Web / PhysicsX)
Merantix closes €103M fund for industry-specific AI
Berlin's Merantix Capital closed a €103 million fund on June 4, more than triple its first vehicle. Half the capital goes to founders it incubates from the pre-idea stage at its Berlin AI Campus; the other half to direct pre-seed and seed deals across Europe.
The thesis is narrow on purpose. Merantix is backing teams applying AI to logistics, manufacturing, energy, healthcare, and robotics, the sectors where it argues Europe's strength lies, rather than horizontal model builders.
The limited partners tell the same story: forklift maker Jungheinrich and asset manager Union Investment sit alongside the usual institutions. The bet is that Europe's moat is the gap between its industrial base and the startups rebuilding those industries. (The Next Web)
Deployment & Applications
Agentic AI reaches regulated banking, governance attached
At Money20/20 Europe in Amsterdam on June 2, Experian launched an Agent Operating System for financial services: infrastructure for autonomous machine decisioning in lending and credit. ServiceNow is the first integration partner, connecting its AI agents to Experian's data and decisioning layer inside existing enterprise workflows.
The notable part is what ships alongside the autonomy. The pitch leads with governance and auditability, not just capability, a tacit acknowledgment that agentic systems making credit decisions land squarely inside the AI Act's high-risk category.
Banking is converting AI pilots to production faster than any other sector. Whether the governance layer holds up under a regulator's first real audit is the test that matters, and it has not happened yet. (FF News)
Governance & Safety
The AI Act's expert backbone is finally staffed
On June 2, the Commission named the members of the AI Act's Scientific Panel and Advisory Forum, recruiting around 60 independent experts to advise the AI Office and national authorities. They serve two-year terms and will work on systemic-risk classification, model evaluation, and cross-border surveillance of general-purpose AI.
This matters because of the calendar. GPAI enforcement powers and the Article 50 transparency rules both activate on August 2, and until now the AI Office lacked the independent technical bench to exercise them credibly.
The transparency consultation closed June 3, with the final Code of Practice on labelling AI-generated content due this month. The enforcement machinery is assembling roughly eight weeks before it has to run. (European Commission / AIwire)
Compute & Infrastructure
The EU's €20B gigafactory plan is falling apart before the first bid
Reporting on June 2 confirmed that the EU's plan for five one-gigawatt AI gigafactories is stumbling. Bidding has slipped from May to July, only two of the five sites can be funded before the 2028 budget cycle, and interest has narrowed from about 70 companies to roughly 10.
The scale mismatch is the story. The EU's €4.1 billion in direct subsidies, spread across five countries, sits against SoftBank's €75 billion data centre commitment in France alone.
Some bidders have stopped waiting. Germany's Schwarz Group is building its own data centre south of Berlin without subsidies; Deutsche Telekom will only commit if customers guarantee demand. The private appetite exists. The bureaucratic process has struggled to harness it, which is the exact gap the new sovereignty package now promises to close. (The Next Web)
AI & Society
Mathematicians draw a line: the Leiden Declaration
A coalition of mathematicians from Oxford, Cambridge, ETH Zurich, Columbia, and Northwestern issued the Leiden Declaration on June 2, endorsed by the International Mathematical Union. It is the most organised response yet from an academic discipline to how AI companies use published research.
The objections are specific: training on papers without consent, announcing results by press release instead of peer review, and eroding attribution. The recommendations are blunt too: develop licensing that blocks non-consensual training, regulate the industry, invest in public compute.
It is easy to read this as reflex resistance. It is more accurate to read it as a profession trying to set terms before the terms are set for it, which is a form of agency that scientific communities rarely organise quickly enough to use. (The Next Web / Scientific American)
One to Watch
July: the gigafactory bids, and whether CADA learns from them
The gigafactory call for proposals lands in July, the first concrete test of whether Europe can convert compute ambition into built capacity. Watch two numbers: how many of the roughly 10 remaining consortia actually bid, and whether the Commission funds more than the two sites currently within budget.
The deeper question is whether the Cloud and AI Development Act, tabled the same week the gigafactory plan faltered, has internalised that lesson or simply restated the ambition at larger scale. The same actor, the Commission, is running both. The July bids will show which version of Europe's compute strategy is real.