Lagarde Backs Public First Losses for €1.3 Trillion AI Buildout
Christine Lagarde proposes €100 billion in public first-loss capital for a €1.3 trillion European AI buildout, targeting 45 GW of compute by 2030 amid grid delays.
By Olivia Hart
5 min read

What's News
- Lagarde’s Nov. 20, 2026 proposal would use public first-loss capital for a €1.3 trillion European AI buildout.
- The strategy targets 45 GW of EU AI compute by 2030, up from less than 2 GW in 2026.
- It recommends €100 billion in public support: €25 billion at EU level and €75 billion across member states.
- Europe aims to raise its share of global AI computing capacity from around 5% to 15% by 2030.
- The strategy assumes $34.4 billion of capital expenditure per gigawatt and global compute supply of 300 GW.
By Olivia Hart
Christine Lagarde proposed on November 20, 2026, that European public money absorb first losses on chip plants and data centers in an AI buildout estimated at €1.3 trillion. She made the pitch in Frankfurt at the 36th Frankfurt European Banking Congress.
Lagarde said public funds, including money from the EU budget, should take the initial risk. That would de-risk strategically important projects and attract private capital, rather than leave governments responsible for financing the entire buildout.
The proposal remains a policy idea, not an adopted EU funding program. It was Lagarde’s eighth and final address to the congress as European Central Bank president and illustrated her case for stronger European institutions when a challenge requires a European solution.
Who pays for Europe’s AI buildout?
The €1.3 trillion estimate comes from A Transformative AI Strategy for Europe, an independent project whose experts wrote it in their personal capacities. It calls for Europe to raise its share of global AI computing capacity from around 5% to 15% by 2030, a threefold increase.
The strategy translates that goal into 45 gigawatts of AI compute on EU soil, measured as total facility power. It starts from less than 2 GW at its 2026 baseline and assumes global compute supply reaches 300 GW.
The project defines compute as the specialized chips and data-center infrastructure used to train and deploy models. Its investment assumption is $34.4 billion of capital expenditure for every gigawatt of facility power.
The strategy recommends a €100 billion public contribution, less than one-tenth of the buildout estimate. The money would de-risk strategic investment and expand European-only AI compute, defined as capacity owned and operated by European organizations.
Of the public package, €25 billion, or one quarter, would come at EU level through loan guarantees and advance purchase commitments. The remaining €75 billion, or three quarters, would come across member states through those instruments and potentially equity investment.
Takeaway: Public capital would absorb the early losses, while private investors would still need to provide the bulk of the funding.
Why does foreign compute threaten European sovereignty?
Lagarde argued that continued access to foreign computing providers and frontier AI models makes near-term European investment harder to justify politically. Yet, she said, “lost industries are difficult to rebuild, and by the time a dependency is turned against Europe, its sovereignty has already been compromised.”
Ready access to foreign capacity lowers the pressure to build immediately, Lagarde said. It does not, however, remove the vulnerability she described or make European production any less necessary.
She framed AI infrastructure as a slow-moving risk: the investment comes due now, while the benefits remain distant. Costs accumulate without a single crisis forcing a decision, unlike the pandemic and Ukraine, where the price of delay became plainly visible.
Lagarde attributed the ECB’s ability to commit early to its euro-area mandate, long horizon and Governing Council majority voting. National governments must instead weigh concentrated domestic opposition against benefits spread across Europe while negotiating with partners facing similar constraints.
Her examples for stronger European action were an effective European capital-markets supervisor and European-level public-private partnerships for large technology projects. Both reflect her stated preference for European institutions when a challenge requires a European solution.
Takeaway: Europe faces a present cost for computing sovereignty, even when foreign providers remain available and the pressure to act appears limited.
Can permits and grids deliver 45 GW?
The strategy identifies lengthy data-center approval and grid-connection processes as the principal obstacle to European expansion. It does not identify a shortage of potential sites or investor interest as the main constraint.
Instead, the project recommends reducing “time to power,” which it defines as the time a data center takes to connect to the grid. It calls for accelerated permitting and prioritized grid connections.
The strategy also seeks a larger local benefit from hosting infrastructure. It recommends guaranteeing host municipalities a share of local business-tax revenue based on installed facility power and using that arrangement to facilitate developer investment in local infrastructure.
Large operators would have to compensate for the incremental electricity-system costs caused by their projects. The strategy says that requirement would keep those costs from spreading across households and local businesses.
First-loss public capital could make large projects more attractive to private investors. The strategy nevertheless treats grid access, permitting and local community benefits alongside financing, rather than as separate obstacles.
Takeaway: Capital may open the door, but faster approvals and grid connections determine whether new capacity can come online.
What remains undecided?
Lagarde’s proposal leaves governments weighing explicit fiscal risk now against a technological dependency that becomes harder to reverse over time. First-loss capital would put public money behind projects whose benefits, by Lagarde’s account, remain distant.
Neither the speech nor the strategy sets an adoption timetable for the proposal. They also do not specify which projects would receive support, how claims against first-loss capital would be managed or when the 45 GW target could become operational.
The financing proposal must now meet the strategy’s nonfinancial tests: shorter grid waits, faster permits and credible benefits for host communities. The next measure of success will not be the public pledge alone, but whether Europe can connect capacity quickly enough to make the sovereign alternative viable.
Takeaway: Europe’s sovereign AI goal depends on permitting, grid connections and local support, not the size of the public pledge alone.
Source: unite.ai
More from Olivia Hart
Show full bio
Staff writer covering people, launches and day-to-day practice in entrepreneurship, startups and venture funding at Business Bearings.
2,535 articles