Omnibus result this morning, London's $1.6B superintelligence cluster, Europe's chip gap

The second political trilogue on the AI Omnibus took place in Strasbourg on Tuesday April 28 — the target date set at the opening of negotiations in late March. The Centre for Democracy and Technology’s April AI Bulletin, published on the day of the trilogue, confirmed the session was proceeding but noted that the key unresolved question entering the room was sectoral integration: whether high-risk AI systems embedded in products regulated under EU product safety legislation (machinery, medical devices, toys, radio equipment) should be exempted from the AI Act’s scope. Parliament had proposed moving these systems into Annex I-B and writing AI requirements horizontally into existing sectoral law. The Council had not accepted this. Civil society broadly opposed the exemption; industry strongly backed it.

If a deal was reached yesterday: The confirmed timeline is IMCO/LIBE committee vote in May 2026, full Parliament plenary vote in June 2026, Official Journal publication in July 2026 — just days before the August 2, 2026 original deadline for Annex III high-risk AI obligations. What would be agreed: Annex III deadline moves to December 2, 2027. Annex I embedded AI moves to August 2, 2028. Targeted ban on AI nudifiers. Synthetic content watermarking deadline to be confirmed.

If no deal yesterday: A second trilogue becomes necessary, likely in May 2026. Every week of additional negotiation compresses the window between formal adoption and the August 2 deadline. At a certain point, even an eventual deal would arrive too late for the Official Journal to publish before August 2 — triggering the original deadline by default. The CDT noted that remaining open questions also concerned exact wording on the prohibition of AI systems generating non-consensual intimate images and child sexual abuse material, where both institutions have agreed on the principle but not the precise legal text.

The European Innovation Council announced on Monday that eight companies have been selected under the EIC STEP Scale Up scheme to receive equity investments of between €10 million and €30 million each. The scheme is designed to close Europe’s persistent gap in late-stage deep tech funding — the zone between early-stage venture and large-scale commercial deployment where European companies most often lose capital competition to US acquirers.

The eight selected companies span Germany, France, Bulgaria, Denmark, the Netherlands, and Spain. The AI entry is Aignostics (Germany), which is applying AI to transform drug development and improve patient outcomes — operating at the intersection of computational pathology and pharmaceutical R&D. Also selected: Alice & Bob (France) building the first universal quantum computer; Endurosat (Bulgaria) on space intelligence; Reverion (Germany) on next-generation biogas power; and Quantware (Netherlands) on 3D quantum computing architecture.

The EIC STEP Scale Up call has a total 2026 budget of €300 million, with evaluation sessions running quarterly. Of the 44 companies that applied, 28 were invited to interview; 8 were selected and will receive investments subject to due diligence. A further 18 companies not selected receive the STEP Seal to facilitate access to alternative or complementary funding. The next batch deadline is May 6, 2026.

The full significance of Monday’s Ineffable Intelligence announcement becomes clearer when placed alongside the two other labs forming in the same geography. What is emerging in London is not a single bet on one researcher — it is a gravitational cluster of reinforcement learning talent, frontier capital, and superintelligence ambition, anchored by the proximity of Google DeepMind’s headquarters.

Ineffable Intelligence (David Silver, ex-DeepMind RL team lead): $1.1B seed, $5.1B valuation, Sequoia + Lightspeed, Nvidia, Google, UK Sovereign AI Fund. Mission: build a “superlearner” that discovers knowledge autonomously through reinforcement learning without human-generated data. Founded November 2025. No product yet.

Recursive Superintelligence (Richard Socher, former Salesforce chief scientist, and Tim Rocktäschel, former DeepMind principal scientist and UCL professor): raised $500 million at a $4 billion pre-money valuation, led by Google’s venture arm GV with Nvidia participating. The round was oversubscribed with demand potentially reaching $1 billion. Also incorporated in the UK. Twenty employees. Public launch planned mid-May 2026.

Project Prometheus (Jeff Bezos): TechCrunch reported that Bezos’s AI lab is in talks to secure office space close to Google’s AI hub in London — directly adjacent to the DeepMind alumni network. Several former DeepMind staffers are reportedly set to join Ineffable’s executive team. The talent pipeline is the same pool. The capital is American. The address is British.

A cluster of European AI chip startups is growing — but its funding is structurally mismatched against US competition. According to Dealroom data reported by CNBC, European AI chip startups have raised a combined $800 million in 2026 to date. Their US counterparts raised $4.7 billion in the same period — a gap of nearly six to one. In the US alone, Cerebras Systems picked up $1 billion in February. MatX, Ayar Labs, and Etched each raised $500 million rounds.

The European companies are real and technically credible. Axelera AI (Netherlands) raised $200M+ in 2026 and is already deployed inside EuroHPC’s Mimer AI Factory in Sweden — the first EuroHPC system to include Axelera accelerators. Olix (UK) also raised $200M+. Optalysys (UK) is planning a $100M+ raise later this year. Fractile (UK) and France’s Arago are both reportedly in the market for nine-figure rounds. The thesis across all of them is the same: AI inference workloads are now dominant, and GPUs were not designed for inference at scale.

Axelera CEO Fabrizio Del Maffeo told CNBC that European governments remain “conservative” in investing in products from new companies and have no equivalent to DARPA — the US defense research agency that has historically funded high-risk hardware bets. Chip development timescales are long, the tape-out-to-deployment path is hard, and Europe’s foundry ecosystem is still maturing. The structural disadvantage is not just capital — it is the absence of a risk-tolerant public buyer.