Test AI with your own data before deploying it
The Plan IA360's Red NEURONA (NEURONA Network) describes a proof of concept: see, test, decide. Run that same process today, without waiting for a centre.
Status as of 22 September 2026. The Red NEURONA (NEURONA Network) is a project announced inside the Plan IA360, presented on 21 September 2026. No centre is open yet: the first milestone is a pilot with 100 SMEs in two autonomous communities before the end of 2026, and the plan doesn't yet say which ones. There's no way to sign up or reserve access.
Among the fourteen flagship projects in the Plan IA360, one describes, without using these exact words, the precise process of a well-run proof of concept. The plan says that in these centres a business «ve la inteligencia artificial funcionar en su sector, la prueba con sus datos y la despliega por suscripción» ("sees artificial intelligence working in its sector, tests it with its own data, and deploys it by subscription"). See, test, decide. That's the right order for evaluating any AI project, whether or not there's a physical centre with that name behind it.
What the plan announces: test centres within 50 km (still not open)
It's the tenth flagship project in the plan: a public-private, territorial network of AI demonstration and deployment centres, backed by advanced supercomputing and designed to bring testing to within 50 kilometres of the business, drawing on capabilities that already exist, such as AI Factories, the Spanish Supercomputing Network, RETECH, data spaces, technology centres and Acelera Pyme ("Accelerate SME"). The stated goal is to go from 21.1% to 55% of businesses using AI by 2030, with a first pilot of 100 SMEs in two autonomous communities before the end of 2026 and a rollout to 10 communities and 5,000 businesses in 2027.
The plan says that process «puede ser particularmente relevante para la pyme, que tiene una barrera de entrada más alta y, al mismo tiempo, una ganancia relativa mayor» ("can be particularly relevant for SMEs, which have a higher barrier to entry and, at the same time, a greater relative gain") from adopting this technology. That makes sense: a large company can afford a pilot that doesn't go well and absorb it into the year's budget; for an SME, every failed attempt weighs more, often because it puts the only person responsible for technology on the line, and that's why being able to see something work before committing matters more the smaller the business is.
What still doesn't exist is an open centre, a list of locations, or a procedure to sign up.
It's worth separating two questions that tend to get mixed up: when a given business will be able to walk into a NEURONA centre, which no one can answer today because there's no territory-by-territory opening schedule, and what process that business will follow once it does, which the plan describes fairly precisely. The second question is the one that matters for deciding what to do in the meantime, because that process — see, test, decide — doesn't have to wait for the first.
What it describes is, literally, a proof of concept
Strip away the building and the subsidy, and what the plan describes is the process any serious technical team recommends before committing budget to an AI project: you don't sign a contract based on a demo run on someone else's data. You first check how the system behaves with your own data, under your own conditions, and only then decide whether to deploy it.
That process doesn't depend on a NEURONA centre existing to be applied. A maturity diagnosis starts exactly there: identifying which use case is most likely to work before building anything at scale, and designing the first pilot as an evaluation, not a launch. If the pilot doesn't demonstrate the expected result, it gets dropped without the whole project having been committed; if it does, it scales with real data behind it, not a hunch.
How to run a pilot risking as little as possible
Before anything else, it's worth knowing whether the data that will feed the test can actually support the project: what state it's in, which process it will support and who will maintain it during the pilot. Reviewing that and flagging the blockers before anyone writes a line of code is exactly the first step of the maturity diagnosis described above.
On that basis, a well-designed pilot is evaluated against the business's real dataset, not against some generic benchmark that says nothing about the specific case. It's the approach used to build an agent that runs a process with reasoning: first an evaluation phase against the client's real data and cases, then a time-boxed pilot with monitoring, and only if the evaluation justifies it, the move to production. Any action that costs money or exposes the client still gets signed off by a person until the system proves it's stable enough, not from day one.
What «a menos de 50 kilómetros» ("within 50 kilometres") means for those who don't want to wait
The geographic closeness the Red NEURONA proposes solves a real problem: testing something before buying it, without having to trust a demo built on someone else's data. But that same logic, seeing a similar case working and testing it with your own data before deciding, doesn't depend on the distance to a centre that doesn't exist yet. It can be applied today, remotely, with any provider willing to evaluate against the business's real dataset before proposing a full deployment. Waiting for a physical centre to open can carry an opportunity cost compared with a narrow pilot done now.
What physical closeness does add, once it exists, is something a remote pilot doesn't fully replace: seeing a case from a similar sector running in production, not in a presentation. That reference helps calibrate expectations (how long it really takes, which part of the process still needs a person) and it's reasonable to wait for a nearby centre to open if that sector reference is what you're after. What doesn't pay off is waiting just to start the diagnosis and the choice of use case, neither of which depends on seeing anything first.
How it relates to the voucher
The plan explicitly links the two projects: the business voucher is tied to the Red NEURONA «al subvencionar las soluciones ofrecidas por estos centros a las pymes» ("by subsidising the solutions these centres offer to SMEs"), so that public investment drives use of the installed infrastructure and feeds back into home-grown technology. In practice, that means the Red NEURONA is the way in for testing and the voucher is what can fund the next step, once the use case has already been seen working. What kind of project literally fits that voucher, with concrete process examples, is in which AI projects fit the IA360 voucher (and which don't). The rest of the project phases the plan describes — diagnosis, training, process integration, measurement — is in what AI project the Plan IA360 will fund, and how to build it.
Frequently asked questions.
Do I need to be near a NEURONA centre to test AI with my data?
No. The principle the Red NEURONA proposes (see, test with your own data, decide) can be applied today with any provider willing to evaluate against your real dataset before proposing a deployment, without depending on the location of a centre that doesn't exist yet.
What happens to my data during a pilot test?
It depends on how the pilot is designed, and it's one of the first things that needs to be set out in writing: what data leaves the business, where it's processed and who has access. A serious pilot starts by inventorying which sources go into the test and which stay out, not by connecting everything at once.
How long does a pilot last before you can decide whether to scale it?
There's no single timeframe: it depends on the process and how much data is available to evaluate. What is fixed is the criterion: deciding on data, not impressions, whether the use case moves to production.
Do I need a defined use case already to start testing?
It helps, but it isn't essential from day one: part of the work of an upfront diagnosis is precisely identifying and prioritising candidate use cases before deciding which one to test first.
Does a negative pilot count as the project failing?
It shouldn't. A pilot designed as an evaluation exists for exactly that purpose: separating the use cases that work from the ones that don't, before the business has invested in scaling any of them. Dropping a use case after a well-measured pilot is exactly what's supposed to happen with some of them; the expensive mistake isn't dropping one, it's scaling without having evaluated first, and presenting a dropped pilot as a failure only discourages honest measurement next time.