Why Ireland's AI Ambitions Are Stuck in Neutral (And What Founders Are Doing About It)
The oldest mistake in business is trying to be everything to everyone. In AI, it is being repackaged as ambition.
Ireland's official AI posture is bullish. The National AI Strategy, backed by hundreds of millions in public and private capital, talks about building capability "across every layer of the stack." That phrase is doing a lot of heavy lifting. It describes, in four words, a plan that would take a country with 5.1 million people and ask it to compete simultaneously with the United States, China, and the combined engineering budget of a company like Google, which spent €28 billion on research and development in 2023 alone. That is roughly fourteen times Ireland's entire annual capital budget.
The fear underneath all this ambition is real and worth naming: if Ireland does not build its own AI infrastructure, it becomes a tenant in someone else's technology. The opportunity is equally real: Ireland has world-class researchers, a proven track record of attracting engineering talent, and proximity to the EU regulatory environment that is fast becoming the global standard. Both things are true. The question is whether the strategy matches the reality.
The 'Every Layer' Trap
Trying to own every layer of the AI stack is not a strategy. It is a procurement wish list.
The stack, for anyone who has not spent time in a pitch deck recently, runs from raw compute and silicon at the bottom, through foundation models and training infrastructure in the middle, to applications and interfaces at the top. Owning all of it requires the kind of capital that makes Ireland's entire annual exchequer receipts look like a rounding error. OpenAI is burning through approximately €5 billion a year just on compute. That figure, stretched across Ireland's population, works out at roughly €1,000 for every man, woman, and child in the country, and it still would not be enough to train a competitive foundation model.
The government's instinct to build broad capability is not wrong. The mistake is assuming that breadth and depth can be achieved simultaneously at Irish scale. They cannot. Every euro spent trying to build a sovereign Irish large language model is a euro not spent making Irish healthcare AI, agricultural AI, or legal AI the best in the world.
Irish companies are already burning through €720 million on AI projects that never ship. The pattern inside that number is consistent: broad mandates, unclear ownership, no defined customer, no defined problem.
What the Winning Founders Are Actually Doing
Pick a lane. Own it completely. That is the entire playbook.
Amethyst Care, based in Dublin, did not try to build a general-purpose AI. They built a system specifically for monitoring older adults in residential care, trained on the data and workflows that exist inside Irish care homes. The result was a product that a care home manager could understand, trust, and defend to a regulator. The founders understood that their customer was not a tech enthusiast. Their customer was a director of nursing with a CQI inspection coming in six weeks.
That specificity is not a limitation. It is the moat.
The framework for Irish founders who want to compete in AI is not complicated, but the order matters.
Step 1: Name the exact customer, not the sector. Not "healthcare" but "GP practices with three to seven partners in secondary towns." The more precisely you can describe the person who will sign the cheque, the better your product will be.
Step 2: Find the regulatory or compliance asymmetry. Irish and EU regulation is a burden for big American AI companies and an advantage for Irish ones. A founder who understands GDPR, the EU AI Act, and Irish data protection practice at the same level as the customer's legal team has something that no San Francisco startup can replicate quickly. Irish compliance startups are already demonstrating exactly this pattern.
Step 3: Build for the workflow, not the demo. The graveyard of Irish AI pilots is full of products that looked extraordinary in a conference room and collapsed the moment they touched a real operational environment. The founders winning right now are the ones who spent the first six months inside the customer's process before writing a line of production code.
Step 4: Price for value, not for adoption. Cheap is not a strategy in a market where the buyer is choosing between your product and doing nothing. If your AI genuinely saves a mid-sized Irish manufacturer four hours of production planning per day, that is worth pricing at the value of those four hours. Discounting to drive adoption signals that you do not believe in your own product. The buyer notices.
The Data on Who Is Succeeding
Enterprise Ireland's own portfolio tells the story clearly enough. The AI-related companies that have reached Series A in the past two years are not the ones building broad platforms. They are verticalists. Legal AI for Irish conveyancing. Predictive maintenance for food processing lines. Triage tools for community mental health services.
The pattern repeats because the logic is sound. A founder competing globally in foundation models is a producer in a market where the inputs (compute, data at scale, engineering talent) are controlled by a handful of American and Chinese firms. A founder building AI for Irish food safety compliance is a producer in a market where the regulatory knowledge, the customer relationships, and the domain expertise are right here, and where the American giants have neither the appetite nor the local context to compete.
This is producer versus consumer thinking applied to competitive positioning. Build where you have genuine advantage. Buy where you do not.
The Turn
There is a version of Ireland's AI story that is genuinely exciting. It is not the version where we announce a sovereign AI model and spend the next four years explaining why it is behind schedule. It is the version where twenty Irish companies each become the undisputed world leader in AI for one specific, valuable, defensible problem, and then sell that to the world.
Tines did not try to build a general automation platform for every industry. They built workflow automation for security operations teams and became so embedded in that world that their valuation crossed €1 billion. That is the template. Not breadth. Depth.
Ireland's AI ambitions are not too large. They are aimed at the wrong target. The founders quietly getting on with it already know this. The strategy documents are catching up slowly.
