The EU will fund your AI. Most companies never ask.

There is a strange asymmetry in European business. Ask a room of operators about AI and everyone has an opinion, a pilot, a fear. Ask the same room who has applied for public co-funding to build the thing they keep talking about, and you get silence — maybe one hand from the company that hired a consultant back in 2019 and still tells the story.

Meanwhile, the money is not hypothetical and it is not hiding. The EU spent the last two years explicitly re-pointing its funding machinery at strategic technology — AI included — and countries like Portugal translated that agenda into open calls for companies. The position of this essay is blunt: if you're building serious AI or automation capability in Europe and you haven't structured any of it for public co-funding, you are voluntarily paying a price your competitors don't pay. Not because funding is magic. Because of what it does to project economics — and, less obviously, to project discipline.

The asking gap

Why do so few companies ask? Three myths do most of the damage.

"That's for research labs and startups with lobbyists." No. The instruments that matter for operating companies are incentive systems for companies — designed for firms investing in R&D or modernizing operations. A 30-year-old family manufacturer is squarely inside the target group, arguably more than the startup.

"The bureaucracy eats the benefit." Real applications take real work — workplans, budgets, indicators, reporting. But the work scales with the project, most of it is work you should be doing anyway (more on that below), and it is a known, bounded cost. Weigh actual paperwork hours against co-financing that commonly covers 45–75% of eligible costs, and the trade is routinely lopsided — in favour of asking.

"We'll move faster with our own money." Sometimes true, and worth respecting: funding calendars are not startup calendars. But "faster" often means "smaller" — the self-funded version of your AI roadmap is usually the timid version. The funded version is the one where you build the platform instead of the patch.

The result of these myths is a quiet market failure: the companies most capable of absorbing funding — real operations, real data, real deployment surface — are the least likely to request it, while a rotating cast of grant-chasers optimizes for the paperwork instead of the product. The gap between those groups is where the opportunity sits.

What STEP actually changed

STEP — the Strategic Technologies for Europe Platform — is the EU's admission that it was funding everything and prioritizing nothing. Rather than inventing yet another programme, STEP re-labels and re-points existing funding streams toward a short list of strategic technologies: digital and deep tech (AI explicitly among them), clean tech, and biotech. Projects that advance those technologies get priority treatment inside the machinery that already exists.

For a company in Portugal, that abstraction becomes concrete in the national incentive systems, where STEP-labelled calls have appeared in two flavours that matter here:

• STEP R&D&I — for projects that develop technology: applied research, experimental development, building systems that don't yet exist. If your AI project involves genuine technical uncertainty — new architectures, new integrations, methods you'll have to discover rather than assemble — this is its home.
• STEP Productive Innovation — for projects that deploy technology into operations: modernizing production, adopting advanced digital capability, changing how the company actually works. The agent that transforms your order flow may qualify even if the underlying science is settled, because the innovation is in your operation.

That pairing matters strategically: it covers both ends of the build — inventing the capability and industrializing it. Plenty of ambitious roadmaps are, structurally, one of each. (The mechanics of how we structure projects to fit these instruments live on our funding page; the philosophy is in Funded innovation in practice.)

One honesty note before the math: calls open and close, rates and eligibility vary by instrument, region and company size, and nothing here is a promise about a specific programme. The durable fact is the direction: Europe has decided that companies building AI capability are a funding priority. Directions like that outlive individual calls.

Funded structuring changes the math

Here's the part that should interest whoever owns your P&L. Public co-funding for company projects typically arrives as non-dilutive support — often a substantial share of eligible costs, sometimes non-refundable, sometimes as favourable instruments. No equity leaves the cap table; no board seat appears.

To make the effect visible, purely illustrative numbers: suppose an ambitious AI build costs €60k — the realistic shape of an agentic system with retrieval over company knowledge, which typically runs 6–12 weeks and €30k–70k. Self-funded, the decision is "spend €60k for uncertain returns" — a decision boards postpone, shrink, or split into half-measures. With co-financing for eligible projects commonly landing between 45% and 75%, take 60% for the arithmetic: the decision becomes "spend ≈€24k net for the same system". Postponement gets harder to defend. But the deeper shift is in scope: with co-funding, the rational move is often the more ambitious project — the data platform plus the agent layer, rather than the quick integration that solves this quarter and nothing else. Funding doesn't just discount the project you had. It changes which project is worth having.

The self-funded roadmap and the funded roadmap are rarely the same roadmap. The second one is usually the one you'd actually want.

And there is a second dividend nobody advertises: the application is a forcing function. To ask for money you must write down objectives, workplan, budget, risks, and measurable outcomes. Most internal AI initiatives never survive that exercise — which is precisely the point. The discipline that funding demands is the same discipline that separates production systems from eternal pilots (we dissected that failure mode in The demo-to-production chasm). Companies that structure fundable projects end up with better projects even before the money lands.

How to structure a project that deserves funding

Patterns from projects that get funded — and, more importantly, that deserve to be:

1. Lead with the innovation, not the shopping list. "Buy licenses and hire a consultant" is procurement. "Develop an agent system that automates X under constraints Y, with technical risk Z" is a project. Evaluators fund the second; so should you.
2. Be honest about which track you're on. Genuine technical uncertainty → R&D&I framing. Proven technology, transformative deployment → productive innovation. Forcing one into the other's costume is the classic self-inflicted rejection.
3. Design milestones you'd want anyway. Working increments, measurable indicators, evaluation gates. If a milestone exists only for the funder, it's fake; funded projects go best when the funder's calendar and the engineering calendar are the same calendar.
4. Keep the IP and the capability. Structure so that what's built — models, data assets, systems, know-how — accumulates inside your company. Funding that leaves you with a vendor's invoice and no capability is subsidised dependency.
5. Start before the call opens. Calls have windows; strategy doesn't. Companies that treat funding as a standing capability — a pipeline of structured, fundable projects waiting for the right instrument — beat companies that improvise an application the week a call closes.

What this looks like in practice

We practice this ourselves: building ambitious systems from Portugal with funded structuring as a first-class tool — it's part of why we build here, where engineering quality and instrument access combine unusually well (the broader case is in Why we build ambitious technology in Portugal). When we scope client systems — agent platforms, automation layers, the kind of work on our AI systems page — "is there a fundable structure here?" is a standard early question, not an afterthought.

And an illustrative archetype: picture a 120-person food producer. Quality control runs on spreadsheets and the heroics of two supervisors. The timid plan: buy a dashboard. The fundable plan: an agentic quality system — sensor data ingestion, an anomaly-detection layer tuned to their lines (genuine R&D: their processes, their failure modes, no off-the-shelf answer), and agents that draft corrective actions for supervisor approval. Structured well, that's a credible R&D&I candidate — and the funded version includes the data platform that makes the next five projects cheap. The unfunded version was a dashboard.

Ask.

The quiet scandal of European tech funding isn't misallocation. It's non-collection: instruments designed for exactly the companies that never open the door. The EU has already decided AI-building companies deserve co-investment — that decision was made, published, and pointed at you. The only step it cannot take on your behalf is the asking.

If you're weighing an AI or automation investment and want an honest read on whether a funded structure fits — including "no, self-fund this one, it's faster and cleaner" when that's the truth — talk to us. Structuring is part of how we work at Gravitnomad, and the first conversation costs exactly nothing.