The €20bn Question: Can Irish VCs Match Moonshot's AI Model Ambitions?
The €20bn question is not whether Ireland can build world-class AI companies. It already has the ingredients. The question is whether Irish venture capital can write cheques big enough to keep them Irish long enough to matter.
The Numbers That Should Be Keeping VCs Awake
DeepSeek raised the equivalent of roughly €1.4bn before its R1 model made headlines in January 2025. Moonshot AI, the Beijing-based foundation model company behind the Kimi assistant, has pulled in over $3bn in total funding. That is more than the entire Irish venture capital market deployed across all sectors in the last two years combined. Put it another way: one Chinese AI startup raised more in a single round than every Irish VC firm has under management.
Ireland's VC ecosystem, measured by Enterprise Ireland's own figures, hovers around €1.5bn in total annual deployment across early and growth stages. That is a solid number for a country of five million people. It is a rounding error compared to what foundation model companies now demand at Series B and beyond. A competitive Series B for an AI infrastructure play in 2025 starts at $150m. Most Irish funds top out well below that as a single-ticket investment.
The brutal binary here is this: Irish VCs are early-stage investors operating in a market that has moved to late-stage capital requirements almost overnight.
What Foundation Models Actually Cost
Foundation models are not software startups with a SaaS margin story. They are industrial infrastructure projects. Training a competitive large language model from scratch requires compute costs that run into the hundreds of millions before a single customer pays a cent. Mistral in France raised €600m at a €6bn valuation in 2024. That round was led by General Catalyst and DST Global, not any European-first fund. The pattern is consistent: European talent, American and sovereign capital.
Irish-based AI research is genuinely world-class. Ireland's semiconductor and deep-tech infrastructure gives homegrown AI companies a real foundation to build on, with access to chip design and photonics capability that most European countries cannot match. The researchers coming out of Trinity, UCD, and the ADAPT Centre are competing with Stanford graduates on paper quality. The problem is not the people. The problem is the capital stack above Series A.
An Irish AI startup that builds something compelling at seed and Series A faces a stark choice at Series B: take American or Gulf sovereign money and move the centre of gravity to San Francisco, or stay Irish and accept a slower trajectory. Most choose to move. That is not failure. That is rational behaviour given where the capital lives.
The Three-Stage Problem
The gap between Irish VC capacity and foundation model funding needs breaks down into three distinct stages, and the order matters because each one makes the next harder to solve.
Stage 1: Pre-seed and seed. Ireland is actually well served here. Enterprise Ireland's HPSU programme, Atlantic Bridge's early funds, and a cluster of angel networks mean a compelling AI team can raise €500k to €3m without leaving Dublin. This stage works.
Stage 2: Series A. Workable but stretched. A €5m to €15m Series A is achievable with Irish-led syndicates, often with a London or European co-investor. Frontline Ventures, Elkstone, and Act Venture Capital can participate meaningfully here. This stage is competitive but not broken.
Stage 3: Series B and beyond. This is where the Irish market falls off a cliff. A foundation model company that needs €80m to €200m to compete on compute, talent, and distribution has no credible Irish-led option. The capital simply does not exist at that scale in domestic funds. This forces the founder toward a binary: international capital on international terms, or stay subscale and become an acquisition target before the model matures.
Irish tech founders are already being acquired before they scale in adjacent sectors, and the same pressure applies to AI. The acquirer gets the team and the model weights. Ireland gets the tax receipts for a few years and a LinkedIn post about a successful exit.
What Can Actually Be Done
The honest answer is that Ireland will not build a sovereign foundation model company the way France is attempting with Mistral. The fund sizes are not there and building them takes a decade. What Ireland can do is three things, and doing all three simultaneously is what separates a strategy from a wish list.
First, anchor the research layer. Keep the universities and ADAPT funded at a level that makes Ireland a destination for AI researchers rather than a stepping stone. Talent retention is cheaper than talent replacement.
Second, build specialist application-layer companies on top of existing foundation models. The real money in the next five years will not be in building GPT-4 competitors. It will be in companies that take foundation models and apply them to specific industries where Irish firms already have domain knowledge: financial services, agri-tech, medtech, and legal. These companies have defensible moats that do not require €500m in compute.
Third, push the National Treasury Management Agency and Ireland Strategic Investment Fund to take a more aggressive position as anchor LPs in larger Irish and European AI-focused funds. The ISIF has a mandate to invest on a commercial basis in ways that support the Irish economy. A €200m commitment to a dedicated European AI growth fund fits that mandate exactly. It has not happened at the scale required.
The Turn
The fear is that Ireland becomes a talent farm for American AI giants. The opportunity is that a small, well-networked country with genuine deep-tech infrastructure and a pro-business regulatory posture can punch above its weight in applied AI, where domain expertise beats raw compute. Those are not the same outcome and the next 24 months will determine which one we get.
Ireland has done this before. It built a pharmaceutical manufacturing base that looked too ambitious in 1990 and feeds the state's balance sheet today. The question is whether the same long-term thinking gets applied to AI before the window closes.
Back yourself. Build the thing. Just do not pretend a €5m seed cheque is a foundation model strategy.
