Cyprus approves its AI strategy. Now it must fund it

The final plan strengthens safeguards and clarifies responsibilities, but extends a key deadline and leaves major implementation costs to a future Action Plan. Cyprus has approved its National AI Strategy 2032, setting out how artificial intelligence should improve public services, strengthen businesses and support research. But the final document leaves a question that will determine whether those ambitions are delivered: how much money will be committed, and when? A comparison of the July consultation draft with the November approved version shows substantive changes. The document has grown from 101 to 120 pages and has been reorganised around clearer policy priorities. It strengthens safeguards, adjusts institutional responsibilities and adds support for locally developed technologies. It also moves the deadline for ministries’ AI plans, removes several specific launch deadlines and leaves significant financing requirements to a future Action Plan. Approval establishes national direction. It does not, by itself, provide the resources needed to implement it. The clearest timetable change concerns ministries and public bodies. The consultation draft required their institutional AI strategies to be published by March 2027. The final version moves publication to the end of 2027, approximately nine months later. Plans will be updated annually and fully refreshed every three years. Other delivery commitments have become less time-specific. The draft called for procurement of six transformational AI solutions within eight months and identification of five priority use cases per ministry within 12 months. These deadlines do not appear in the final implementation chapter. Instead, delivery proceeds through three phases: activating governance and setting priorities, delivering the first portfolios, and expanding initiatives that demonstrate value. Six national transformational projects will be selected from 12 proposed flagship programmes. Early deployments should deliver measurable benefits within 12 to 24 months, but the phases have no rigid start and end dates. This allows projects to progress according to readiness. It also makes it harder for the public to judge whether government is falling behind. The forthcoming Action Plan must supply the deadlines and accountable owners that this approach leaves open. The treatment of private businesses has also changed significantly. The draft proposed institutional AI strategies as part of supervisory reporting in regulated sectors. It also linked AI strategies and acceptable-use policies to access to grants, national computing resources and certain operating licences. The approved text explicitly states that the strategy creates no new legal obligations for private organisations beyond applicable European and national law. Private organisations other than SMEs are encouraged to formalise adoption plans. This narrows the proposed administrative requirements while preserving existing regulatory obligations. The final version also draws clearer boundaries around compliance. The National AI Compliance Framework will provide guidance and readiness support. It will not perform conformity assessments, confer certification or imply regulatory approval. Its registry will not itself establish that a system complies with the law. Technology testbeds and data sandboxes are likewise distinguished from the national regulatory sandbox required under the EU AI Act. A controlled environment for developing a product is not automatically a regulatory assessment of that product. For startups, the proposed AI Sovereign Investment Matching Fund remains. Its envisaged design includes matching private investment euro for euro, up to €1 million per round per company, with 25 per cent of capital allocated to pre-seed and seed rounds and 75 per cent to Series A and B. However, the final wording is more conditional. Investment parameters remain subject to coordination with the relevant authorities. Founders should therefore read these figures as proposed terms, rather than evidence that an operational fund is ready to accept applications. The expectation attached to startup support has also softened. The draft said any startup receiving government support should host a master’s or PhD student for at least 12 months. The final text allows proportionate knowledge-transfer requirements for substantial development support, potentially including a student placement for an appropriate period. There are useful additions for local companies. A dedicated Sovereign AI Technologies programme will support solutions developed within Cyprus’s research and innovation ecosystem, including evaluation for potential public-sector adoption under procurement and regulatory rules. Open-source AI models and platforms receive explicit support for public-sector and mission-critical systems. Enterprise adoption support is more practical, with explicit provision for micro-enterprises and instruments such as vouchers, implementation grants, cloud and computing credits, staff training and integration support. These reflect costs that smaller companies encounter when moving from experimenting with AI to using it in daily operations. Education receives a more detailed framework. The education minister replaces the digital-policy deputy minister as chair of the relevant steering board, with the latter becoming co-chair. Existing education structures will handle pedagogical oversight, avoiding the need for a new permanent committee. The final strategy adds school-readiness assessments, stronger pupil-data protections, supplier portability and preparation for teachers before classroom deployment. It specifies a supervised 12-month school pilot before deciding whether to expand the pedagogical validation system. These changes recognise that introducing AI into education requires teaching expertise, technical support and evidence of value. A regional AI Skilling Hub becomes a distinct initiative, extending Cyprus’s education and training capabilities towards international markets. Domestic skills development remains central, alongside the ambition to develop a workforce of at least 3,000 AI professionals by 2032. Infrastructure planning is more disciplined. Major computing investments will require a techno-economic assessment covering total ownership costs, energy requirements and the balance between national, European and commercial capacity. Upgrading existing infrastructure takes priority where it offers better value. Existing capabilities must also be mapped before new centres are established, with scope to build capacity within institutions already operating. The headline ambitions largely remain: a 15 per cent productivity improvement, 12 per cent GDP expansion from AI-enabled transformation, three AI unicorns and a place among Europe’s top 15 countries for AI research. The final document makes clearer that the productivity and GDP figures are conditional scenarios, not forecasts or guaranteed results. One deadline deserves attention. The approved National AI Strategy 2032 sets the enterprise AI adoption target at 75 per cent by 2030, from 9.3 per cent in 2025. The consultation draft used both 2030 and 2032 in different passages. The final executive summary and KPI table specify 2030. Yet the central weakness remains financial. On page 104, the strategy acknowledges that funding requirements for governance and delivery structures, data and AI infrastructure, Centres of excellence, testbeds and broader education programmes are not quantified there. These requirements will be developed through the Action Plan and institutional plans. That is a substantial gap in an approved national strategy. Naming funds and describing possible European financing does not establish how much money will be available, which institution will control it or whether support will continue beyond an initial pilot. The final document rightly recognises that funding must cover the full life of AI systems. Data preparation, integration, cybersecurity, staff training, operation and upgrades all cost money. But recognising those costs is not the same as committing resources to meet them. Government should now publish a costed Action Plan with annual allocations, confirmed funding sources, staffing requirements and delivery milestones. It should distinguish committed national money from European funding that still has to be secured. Where resources are limited, priorities must be explicit. Cyprus has approved a more developed AI strategy, with clearer safeguards and responsibilities. It has also postponed a key deadline and left essential financing decisions unresolved. Without sufficient, sustained funding, the strategy risks falling apart during implementation. A national AI ambition cannot be delivered through approval alone. It requires a budget that matches the work government is asking the country to do.