EU & U.K. AI Sovereignty Strengthens: How Should International AI Cloud Vendors Position?
By Paul Schell |
10 Jul 2026 |
IN-8205
Log In to unlock this content.
You have x unlocks remaining.
This content falls outside of your subscription, but you may view up to five pieces of premium content outside of your subscription each month
You have x unlocks remaining.
By Paul Schell |
10 Jul 2026 |
IN-8205
NEWSChips Act 2.0, Cloud Sovereignty Framework & U.K. AI Hardware Plan |
A series of recent developments affecting Artificial Intelligence (AI) cloud operators inside the European Union’s (EU) 27 member states and the United Kingdom (outside the EU’s jurisdiction) point to a changing industrial policy landscape that will shape the future of semiconductor and infrastructure procurement. Although some rules are yet to be ratified in law, the direction of travel has been set, and previous frameworks, such as the EU’s Cloud Sovereignty Framework (see ABI Insight “Lenovo’s ODM+ in the Context of Increasing EU Supply Chain Sovereignty”), are being disseminated to industry. This suggests that vendors should consider their positioning within a changing policy landscape inside the largest trading bloc and the United Kingdom, a significant economy in its own right (and one likely to align more closely with its geopolitical neighbor over the coming years). Nonetheless, there are key differences between each jurisdiction’s approach:
- EU Chips Act 2.0: The proposal for the second Act (June 3, 2026) builds on the original 2023 Act, which the European Commission (EC) credits with mobilizing over €52 billion in public and private investment. The framing is explicitly AI-driven: it cites AI-related components that drive roughly 70% of global semiconductor industry growth and laments the EU’s extensive reliance on third countries for advanced manufacturing and design. It aims to reduce reliance on external vendors (read: U.S. and Chinese) with the explicit goal of cutting overreliance on external suppliers for critical semiconductors and infrastructure.
- Cloud and AI Development Act (CADA): The CADA proposal introduces a single EU-wide sovereignty framework with four assurance levels for public sector bodies and critical industries to apply based on risk levels, factoring in how dependent operators are on foreign vendors. This pairs with a common EU-level procurement framework for public administrations to pool purchasing power, directly affecting buying criteria. Importantly, it also keeps the majority of the market open to partners, confirming this is conditional access rather than exclusion. Some of the operational mechanisms behind this, as set out in the separate but related Cloud Sovereignty Framework of October 2025, already exist.
- U.K. AI Hardware Plan: U.K. policy takes a different approach—rather than regulation combined with a sovereignty-grading, the AI Hardware Plan commits over £1.1 billion to build an end-to-end pipeline (Develop → Demonstrate → Deploy → Scale) across four pillars. There is an awareness of the need to partner with foreign technology vendors, and the framing is deliberately "open and globally integrated.” It seeks to leverage the United Kingdom’s strengths: a deep design-Internet Protocol (IP) base, photonics and compound semiconductors, a number of world-class High-Performance Computing (HPC) facilities, and homegrown cybersecurity Intellectual Property (IP). Key funding beneficiaries include Fractile (inference compute), Lumai (optical compute), Oriole Networks (photonic networking, validating with AMD), and Callosum (heterogeneous compute systems software).
On a fundamental level, the EU’s approach differs from the United Kingdom’s: the former looks to regulate and guide procurement decisions toward EU vendors. At the same time, the United Kingdom focuses on nurturing domestic talent and key parts of the supply chain through its venture capital arm (although the EU also provides significant funding alongside rules). But both point to a strengthening of sovereignty frameworks centered around AI and its diffusion.
IMPACTHow Will This Affect Incumbents? |
Despite their differences, several major trends and themes are present in the U.K. and EU approaches to sovereign AI.
- Public Funding is Significant and Conditional: Tripling EU data center capacity, AI Gigafactories, and Demand Accelerators; governments are now anchor customers for AI compute, and they are attaching conditions (EU added value, sovereignty level, and U.K. IP) to their procurement.
- Sovereignty is Being Operationalized Through the Bill of Materials (BOM): The EU framework's supply-chain and technology goals (SOV-5, SOV-6) assess where chips are made, the origin of software, and European freedom from foreign vendor lock-in. Sovereignty is now a feature of the hardware, audited down the supply chain.
- The Compute Stack is Diversifying, Supported by Policy: Both jurisdictions explicitly back inference, edge, embodied/physical, photonic networking, and heterogeneous compute, which creates space for new entrants and erodes single architectures/designs and vendor lock-in.
- Partnerships with Key Foreign Vendors Are Protected: The EU keeps most of the market open and lets the EC recognize third-country providers; the United Kingdom frames everything around trusted partnerships and global integration. Incumbents are not shut out; instead, they are being asked to localize, partner, show, and add demonstrable European/British value to keep their access to the protected segments.
EU- and U.K.-based cloud service providers rely heavily on U.S. vendors such as NVIDIA, Intel, AMD, Cisco, HPE, Dell, and a host of other silicon and server integrators’ expertise to bring cutting-edge AI and HPC data centers to market. It is impossible, under the prevailing conditions, to deploy leading-edge AI systems without the vendors and their ecosystems. Sovereignty Effective Assurance Level 4 (SEAL-4) (Full Digital Sovereignty) is a future aim and not something attainable by buyers today. How does this impact AI semiconductors and server vendors?
- Incumbent Semiconductor Vendors: The calculus shifts from dominant by default to dominant where your sovereignty scoring is above the minimum requirements. The EU’s SEAL guidance acknowledges that today’s hardware supply chain renders SEAL-4 unreachable. Foreign-controlled silicon is placed directly in the regulator's crosshairs as the dependency is reduced. Public-sector and sovereignty-sensitive demand in Europe will increasingly hinge on supply-chain transparency, audit rights, firmware and software provenance, and localization. The "open to partners" stance and third-country recognition mean these vendors can still compete—but only by meeting criteria they have historically not had to consider. U.K. and EU funding will also be funneled toward local champions, including startups.
- Incumbent Server Vendors: The EU’s sovereignty assessment impacts components beyond the compute and factors in software, networking, and the location of manufacturing or assembly. This will favor those with localized supply chains, especially those able to assemble in Europe with deep partnerships with domestic integrators and solution providers. The United Kingdom explicitly calls out heterogeneous compute, openness to multiple compute types from diverse vendors, and the ability to integrate quantum in its flagship supercomputers. Vendors that adopt an open approach with deep-rooted operations across the continent will be favored over those that assemble and source solely in Asia-Pacific or North America.
RECOMMENDATIONSHow Should AI Semiconductor & Server Vendors Position? |
The U.K.’s approach is distinct from the EU’s in that it uses public procurement as a demand lever to build momentum for the deployment of U.K.-developed technology into systems. On the other hand, the EU, while also providing significant funding for home-grown technology, is crystallizing frameworks and rules that will shape future investment decisions. Both approaches have their own merits and are shaped by each political institution’s legacies and traditions, as well as by each jurisdiction’s relative size and supply chains. Regardless, vendors should treat sovereignty as a product and differentiator. Useful exercises include:
- Portfolio Mapping to Sovereignty Tiers: Build an internal matrix that scores each product and configuration against CADA's four levels and the eight SEAL objectives. For each Stock Keeping Unit (SKU), identify an achievable ceiling and the single objective that limits it (the overall SEAL is the lowest achieved in any Objective). For most non-European silicon and systems today, that cap is SOV-5 (supply chain) or SOV-6 (technology). An awareness of the ceiling means vendors know which public tenders can be won as is, which require a partner, and which require structural change. This map should drive product roadmaps, not just bid/no-bid decisions.
- Build for the Diversified, Heterogeneous, Inference-led Stack: Both jurisdictions are funding the move beyond monolithic and more general-purpose compute (like Graphics Processing Units (GPUs) geared toward
- training). Silicon vendors should align roadmaps to the funded frontiers of inference and edge accelerators, photonics, and novel architectures. Server vendors should make multi-accelerator orchestration, chiplet/heterogeneous integration support, and photonic-interconnect readiness first-class capabilities, and should be able to integrate third-party and novel architectures, not just their preferred stack (NVIDIA’s NVLink Fusion is a good example). The Oriole-AMD validation partnership is a template: incumbents co-engineering with diversified startups to capture new demand instead of being displaced by it.
- Partner Strategically: Assessments run end-to-end, and sovereignty levels depend on vendors’ partners as much as their own. Form relationships with EU/U.K. sovereign cloud providers, system integrators with EU-based operations, and trusted foundry and packaging partners (see U.K. Semiconductor Center). For non-European vendors, a credibly EU-controlled partner can feasibly lift a joint offer to a higher level that neither could reach alone.
AMD’s partnership with Oriole Networks and Imperial College London is a blueprint for a successful partnership in the United Kingdom, leveraging local IP and explicitly aligning with the AI Hardware Plan and similar initiatives. It is worth noting that CADA and Chips Act 2.0 have yet to be ratified, a process that requires approval by all EU member states. Nonetheless, the trajectory is a course toward more, rather than less, sovereignty for AI clouds in Europe, with the October 2025 Framework setting an example that has already influenced procurement decisions.
Written by Paul Schell
- Competitive & Market Intelligence
- Executive & C-Suite
- Marketing
- Product Strategy
- Startup Leader & Founder
- Users & Implementers
Job Role
- Telco & Communications
- Hyperscalers
- Industrial & Manufacturing
- Semiconductor
- Supply Chain
- Industry & Trade Organizations
Industry
Services
Spotlights
5G, Cloud & Networks
- 5G Devices, Smartphones & Wearables
- 5G, 6G & Open RAN
- Cloud
- Enterprise Connectivity
- Space Technologies & Innovation
- Telco AI
AI & Robotics
Automotive
Bluetooth, Wi-Fi & Short Range Wireless
Cyber & Digital Security
- Citizen Digital Identity
- Digital Payment Technologies
- eSIM & SIM Solutions
- Quantum Safe Technologies
- Trusted Device Solutions