Launching an AI startup in Barcelona has never been more accessible: computing infrastructure available through the Barcelona Supercomputing Center, a dense research ecosystem, an international community and a cost of living compatible with distributed teams. On the finance side, however, feedback from French Tech Barcelona founders highlights five recurring challenges. They are not insurmountable; they are simply rarely anticipated in initial plans, and expensive when discovered too late. This article reviews each one, alongside the practical approaches teams actually use to manage it.

Challenge 1: Unpredictable GPU Burn

Unlike a conventional SaaS startup, whose infrastructure costs rise linearly with user numbers, an AI startup sees its GPU burn change in steps, often sharply. A fine-tuning cycle on a proprietary model can consume in one week what infrastructure uses in a quarter at steady state. A traffic spike on production inference can multiply the monthly cloud bill by five.

Founders who have encountered this wall describe two useful habits. First, establish a separate GPU budget within the cash flow plan, with an automated alert threshold above which any new experiment requires explicit approval. Second, activate BSC credits (Barcelona Supercomputing Center), which several local AI startups access through research programmes. This little-known route can absorb a substantial share of a project’s GPU burn during R&D. Hyperscalers such as AWS Activate, Google for Startups and Azure for Startups also offer significant first-year credits, which should be negotiated when the company is established.

Challenge 2: Recruiting Scarce Talent

Barcelona salary ranges for an applied research specialist or staff ML engineer sit between €75,000 and €120,000 gross per year, lower than Paris by an average of 25–35%, but with fierce competition between the R&D hubs of HP AI Innovation Hub, Amazon R&D, Microsoft and local startups. The difficulty is not only salary: time to hire weighs on cash flow. Filling a senior ML engineer vacancy in Barcelona takes an average of three to five months, and the lost progress on the roadmap translates directly into runway consumed.

Approaches that work include sourcing through alumni of local laboratories, such as CVC, IIIA-CSIC, ELLIS and IRI, drawing on technical meetups in the local ecosystem and budgeting for a relocation package from pre-seed to attract candidates from Paris, London or Berlin. The Beckham Law, combined with the cost of living, produces highly competitive take-home packages for a French or British candidate.

Challenge 3: Misaligned French-Spanish R&D Tax Credits

France’s Crédit d’Impôt Recherche can support eligible R&D activities, at a rate of 30% on expenditure up to €100 million and 5% above that amount, according to the tax administration’s official guidance. Retaining a French entity is not enough: the eligibility of the company, activities and expenses must be checked, alongside their interaction with CDTI, ACCIÓ grants and Spanish deductions. Public subsidies relating to activities eligible for CIR must be deducted from its calculation base. Risks include ineligible expenses, incorrect allocation between entities and combinations that conflict with the rules of a particular programme.

The right habit is to check this interaction BEFORE the first funding round with a CFO and tax advisers who understand both countries. Generalist firms do not always cover both sides of the framework. Firms operating across France and Spain, such as Iter Advisors in Barcelona, can coordinate a structured analysis. Eligibility and amounts must then be confirmed against the specific application and official rules.

Challenge 4: Coordinating Bilateral Funding

Depending on its structure, an AI startup founded by French entrepreneurs in Barcelona may qualify for an impressive range of public programmes: Bpifrance, including French Tech Emergence and seed loans, ENISA in Spain with participative loans of up to €1.5 million, CDTI R&D project funding, ACCIÓ grants from the Generalitat, Horizon Europe and the EIC Accelerator at European level, alongside the CIR already mentioned.

The trap is to assume either automatic compatibility or a general prohibition on combining programmes. Compatibility depends on the costs, type of aid and programme limits. ENISA confirms that its loans can be compatible with other public or private funding, while CDTI explains the rules for its R&D projects. The method is to map programmes by expense, including R&D personnel, infrastructure, licences and subcontracting, declare concurrent aid and check the limits before applying. There is no automatic euro-for-euro reduction across all programmes.

Challenge 5: Preparing a French-Spanish Exit

This challenge arises later in the cycle but needs to be anticipated in the initial structure. An AI startup that has raised several rounds, operated from Barcelona and is preparing an acquisition or IPO must assess exit taxation according to the seller’s residence, the structure and the applicable regime. Under the general regime, Spain’s Personal Income Tax Law, articles 66 and 76, provides a combined marginal rate of 30% above €300,000, rather than applying it to the entire capital gain. The effects of the Beckham Law, a French holding company and bilateral agreements require individual analysis.

An unsuitable initial choice of holding company, BSPCE or stock option regime, or tax calendar can significantly affect net sale proceeds. The financial effect must be calculated for the specific case, without promising a standard saving per founder. In this type of scenario, an outsourced CFO who understands both countries can work with tax advisers on structural decisions taken several years before exit, alongside monthly financial management.

Summary

The five challenges, GPU burn, recruitment, cross-border R&D tax credits, bilateral funding and exit, are not unique to AI startups in Barcelona, but their combination in a French-Spanish context requires particular preparation. Good planning can reduce risks without eliminating them. Senior financial governance, through a fractional or dedicated CFO, should be considered according to complexity, burn, cash position and fundraising schedule, without a universal threshold or guaranteed improvement in runway or valuation.

For more detail on the fractional CFO model, our dedicated 2026 guide covers its five core responsibilities and cost ranges. For the AI toolkit applied to finance, our article on AI and startup finance in 2026 reviews the use cases that actually generate ROI.