The Plan
Sovereign AI Investment Fund (SAIF)
Motivation
AI will change everything: the economy, jobs, society, geopolitics.
AI is already performing tasks that take humans decades of training to master. It is the advisor, psychologist, and main source of information for many people. Its capabilities are increasing at a record pace.
Look at how the world was 30 years ago before the internet and mobile phones, and look at how it is now. Think about the world 30 years from now and how AI and other technologies could have changed it.
We are not leading this change ā we are barely part of it. The US and China are investing hundreds of billions of dollars into AI companies, infrastructure, and research, while Europe's largest AI company, Mistral, has raised just $3B, compared to OpenAI's $168B.
As with energy and digital infrastructure, we are dependent on others at a moment when we should be leading the change.
Our Proposal ā Phase 1: Eliminate the Dependency
Building competitive AI requires enormous capital ā for compute, talent, and research.
We propose the Sovereign AI Investment Fund (SAIF) ā massive-scale investment across four pillars:
-
Datacenter Infrastructure: Direct construction (e.g., research institute datacenters) and indirect investment (funding companies and startups that build them).
-
A CERN for AI: A research institution modelled on CERN, dedicated to foundational AI research and creating hubs of excellence across Europe.
-
AI Startups and Companies: Funding promising startups, prioritising domestic companies while also investing in leading firms like OpenAI and Anthropic to reduce risk and facilitate cooperation.
-
Adjacent Technologies and Strategic Resources: Robotics, quantum computing, brain-computer interfaces, and securing strategic resources like rare earth elements and semiconductor manufacturing capacity.
The Funding
SAIF pools $100ā200B in public capital to mobilise $300ā600B in total, similar in scale to major EU initiatives.
SAIF Architecture ā Public Anchor, Private Scale
The core design principle is leverage: public capital de-risks and anchors private investment, rather than replacing it. The structure operates across two layers.
Layer 1 ā Sovereign AI Investment Platform ($100ā200B public capital).
The core public vehicle. Capitalised by member states, national development banks, and EU instruments. Functions as a strategic investor and cornerstone anchor for Layer 2 funds.
Layer 2 ā AI Growth Funds ($200ā400B private and institutional capital).
SAIF anchors multiple specialised funds, each with a strategic mandate, run by professional fund managers (e.g., EQT, Northzone). The sovereign platform provides ~one-third of each fund's capital:
| Fund Type | Sovereign Anchor | Private & Institutional | Total per Fund | Funds | Category Total |
|---|---|---|---|---|---|
| AI Infrastructure | $20ā80B | $40ā160B | $60ā240B | 1ā2 | $120ā240B |
| Companies | $20ā60B | $40ā120B | $60ā180B | 2ā3 | $180ā360B |
| Total | $100ā200B | $200ā400B | ā | ā | $300ā600B |
This is the model Bpifrance and KfW already use domestically ā scaled to the European level. SAIF acts as a cornerstone investor, not as a day-to-day operator, preserving commercial discipline while guaranteeing strategic direction.
Member states' first-year contribution is just 0.1ā0.3% of GDP, for a combined total of $20ā40B.
How to Assemble the Public Capital?
For EU member states, public capital is pooled from multiple existing channels. Non-EU partners (UK, CH, NO, IS, CAN, AUS, and others) participate through direct state contributions and their own national development banks. More details will follow soon.
| Source | Estimated Contribution | Mechanism |
|---|---|---|
| EIB Group | $20ā40B | First-loss guarantees by member states and InvestEU |
| National Development Banks | $25ā55B | KfW, Bpifrance, CDP, and others |
| RRF / NGEU Reallocation | $10ā15B | Unspent RRF digital allocations |
| Defence Budgets | $5ā10B | Dual-use AI infrastructure |
| Direct State Contributions | $40ā80B | Phased capital calls, 50% in Year 1 |
| Total | $100ā200B | ā |
- EIB Group: The EIB already deploys ~$80B/year across EU projects. Under InvestEU, member-state first-loss guarantees allow $1 of guarantee to mobilise $5ā10 of lending. $20ā40B is well within its historical capacity for a flagship strategic mandate.
- National Development Banks: KfW (Germany) alone has a balance sheet of ~$550B and routinely co-invests in exactly this type of strategic industrial initiative. Bpifrance manages ~$80B in AUM; CDP (Italy) ~$500B. A combined $25ā55B commitment across all NDBs is conservative.
- RRF / NGEU Reallocation: The EU's own data shows several countries have consistently underspent their digital transformation envelopes (originally ~$150B across all RRF plans). Redirecting unspent or upcoming tranches is legally straightforward under existing RRF rules.
- Defence Budgets: NATO countries are already committing to 2%+ GDP on defence. Dual-use AI infrastructure (logistics AI, sovereign cloud, edge computing for military) qualifies under existing defence budget frameworks. $5ā10B is a small fraction of the ~$300B+ European defence spend now being planned.
- Direct State Contributions: France, Germany, Italy and Poland alone have a combined GDP of ~$8T. A phased commitment of 0.1ā0.3% GDP over 3 years from major member states easily reaches this range ā comparable in structure to national contributions to NextGenerationEU.
Participation
SAIF is built on voluntary participation, open to EU and non-EU partners (UK, CH, NO, IS, CAN, AUS, NZ, JAP, KOR), and expandable without renegotiating the founding treaty.
-
A coalition of the willing moves faster than unanimity.
-
EU countries benefit from cheaper energy prices, chip manufacturing capabilities, geographic diversification, and increased funding scale.
-
Non-EU countries benefit from a funding scale impossible to achieve independently, a large talent pool, and access to the EU market.
Governance
SAIF is built for profit. Governance is delegated to independent investment professionals, under a strategic public mandate.
Why built for profit?
Public investment in AI must deliver returns ā for two reasons:
Political sustainability: a fund that generates visible dividends builds lasting public support, making it impossible for future governments to dismantle. A fund perceived as a subsidy machine is vulnerable the moment the political wind shifts.
Capital efficiency: profit discipline forces SAIF to back winners, not distribute favours. Every major sovereign wealth fund that has endured ā Norway's GPF, Singapore's Temasek, Abu Dhabi's Mubadala ā operates on commercial return expectations. The ones that didn't (many development banks of the 1970sā80s) became fiscal black holes.
This does not mean maximising returns at any cost. It means SAIF must be able to justify every investment on financial grounds in addition to strategic ones.
Why delegate governance to independent professionals?
Political bodies make poor fund managers. Governments have:
- Short election cycles
- Diverging national interests
- Incentives to protect incumbents rather than back challengers
Delegating investment decisions to independent professionals removes these distortions.
The state sets the strategic direction, professionals execute it.
What does the strategic public mandate look like?
The mandate is defined at the founding treaty level. It specifies, among others:
- Eligible geographies
- Minimum domestic exposure
- Excluded activities
- Return floor
- Transparency
Within these guardrails, investment professionals have full discretion. Governments cannot instruct them to back a specific company, a specific national champion, or a specific technology.
Comparative Models
SAIF has clear, successful predecessors ā including Bpifrance & KfW, the Norway SWF, Temasek, and Mubadala.
How SAIF Compares to Existing Models
| Comparative Model | Similarities to SAIF | Key Differences |
|---|---|---|
| Bpifrance & KfW | Public anchor investors; back startups and infrastructure with commercial discipline under a strategic mandate | Operate only nationally; no cross-border pooling of capital or mandate |
| Norway SWF (NBIM) | Massive, long-term state capital pool with a strict firewall between political mandate and professional execution | Entirely passive and globally diversified; no strategic sector focus, no active ownership |
| Temasek (Singapore) | Takes active, concentrated equity stakes in deep-tech and VC; long investment horizon; commercially driven | Operates globally without a coalition governance structure; financed by a single state |
| Mubadala (UAE) | Proactively deploys capital into strategic sectors ā AI, chips, biotech ā with explicit geopolitical intent | Financed by petrodollars from a single country; no democratic accountability structure |
| EIB / InvestEU | EU-scale public investment vehicle with member-state backing; funds infrastructure and VC through guarantees | Does not take equity stakes or run active investment mandates; focused on de-risking, not strategic ownership |
What SAIF does differently: it is the first vehicle designed to combine:
- Multi-country democratic governance
- Active strategic equity ownership
- A commercial return mandate
- A defined AI-specific industrial mission
ā at European scale.
Relationship with EU Frameworks
SAIF does not compete with existing EU programmes ā it transforms them into a more coherent ecosystem.
SAIF's Role Alongside EU Programmes
| EU Programme | SAIF Role |
|---|---|
| Chips Act | SAIF's $120ā240B compute buildout guarantees domestic demand for European fabs ā closing the demand-supply loop |
| InvestEU | First-loss guarantees (~$3ā5B) on Layer 2 funds, backing $20ā40B in fund capital ā exactly how InvestEU was designed to work |
| NGEU / RRF | SAIF as NGEU's strategic successor ā concentrated equity replaces dispersed grants; unspent digital funds redirected |
| ReArm Europe | Defence AI procurement sources from SAIF-backed companies; dual-use compute qualifies for defence budgets |
| InvestAI | $20B AI Gigafactories budget could be co-deployed alongside SAIF |
| Horizon Europe | SAIF funds the scale-up of Horizon-funded deep-tech startups that commercialise successfully |
A CERN for AI
Europe needs a unified institutional home for foundational AI research ā a CERN for AI. SAIF provides the financial engine and compute infrastructure; the CERN for AI provides the scientific independence. Together they close the loop between research and industrial scale-up.
-
Mission: Attract and retain top-tier global AI talent; produce foundational breakthroughs that enter the European ecosystem first.
-
Structure: Treaty-based intergovernmental organisation ā scientific directions set by researchers, not governments or investors.
-
Financing: Direct member state contributions (as with CERN today) + SAIF-funded compute infrastructure.
-
Momentum: Builds on existing proposals including the Manifesto for the Creation of a CERN for AI, already backed by significant scientific and political support.
Phase 2: AI to the Benefit of All
Phase 1 builds the infrastructure and the investment base. Phase 2 is about distributing the returns to citizens.
The specifics will depend on how AI reshapes our society, but several benefits are already foreseeable:
-
Jobs and tax revenue: Having AI companies headquartered in our countries means employment and taxes paid here.
-
Citizen dividends: Via SAIF, countries will own equity in top AI companies. Returns can be distributed among citizens, invested in welfare, or both.
-
A seat at the table: Governing AI may require international coordination on the scale of nuclear arms control. Being competitive in AI gives us a voice in that process.
This section will be expanded as the landscape evolves.
For any of this to happen, Phase 1 is necessary.
The Risks of Inaction
If Europe does not act, the consequences are economic, industrial, and strategic:
-
Poorer. AI is transforming industries, shifting job markets, and accelerating scientific research at a pace we cannot fully predict. We will miss out on the value created.
-
Dependent. European industries will rely on US and Chinese platforms for their core operations. This is the energy dependency problem repeated in digital form.
-
Voiceless on AI safety. AI safety governance may become as consequential as nuclear arms control. Without domestic capability, Europe will have no seat at that table.