Bulgaria's €49M AI Factory: What Companies Can Use Now
On 7 November, Bulgaria's Ministry of Innovation and Digital Transformation published part of the draft 2027 budget. The most quoted number was €49 million for an AI factory. Behind the headline is something more specific: the money continues BRAIN++ / Discoverer++ at Sofia Tech Park, and according to the ministry the project should provide computing resources, expert help and training to companies, researchers and public institutions.
For most Bulgarian companies, the question is not whether the state is spending this money well. The more useful question is what a company can get from the factory today and what it needs to bring so it doesn't leave empty-handed. Our answer, after reading the access terms, is that the entry ticket is not a GPU budget. It is a well-scoped question.
What is actually in the draft budget
The facts from the ministry's post, reported by Plovdiv24 and Fakti.bg on 7 November:
- €49 million in 2027 to continue the European BRAIN++ / Discoverer++ project;
- an additional €62.1 million for science and applied research, and €10 million more for secondary vocational technical education;
- funding for the European digital identity wallet.
This is a draft. The numbers can change before Parliament adopts the budget.
BRAIN++ is not a new idea. On 12 March 2025 EuroHPC selected Bulgaria as one of six new AI factories in Europe. The Sofia Tech Park and INSAIT project is worth €90 million, and at the time the government committed to 50% national funding from 2026. The factory has two parts: the new Discoverer++ supercomputer and a service centre for government, education and private organisations.
What a company can get
Most news coverage stops here, but the most useful information is on the BRAIN++ website. Access is split into tiers.
Small Projects. This is the entry point. They are free, last from a few days to a week, and never carry equity, a SAFE or a success fee. According to the service description, a company can get limited GPU access, AI consultancy, problem scoping, an architecture review, data strategy consultation, training and co-working space at Sofia Tech Park. A minimum score of 60 points is required.
Big Projects for private companies. For startups, SMEs and scale-ups with more mature ideas. Here BRAIN++ applies a market-investor participation model: the factory provides a substantial support package and receives a proportionate, capped participation in future commercial success. The minimum score is 70 points.
Where proposals are of equal quality, preference goes to Bulgarian projects, then South-East European ones, then other EU Member States.
Why the question matters more than the compute
What follows is our interpretation, not a position of the ministry or BRAIN++.
A week of free expertise is a lot if you know what you are asking and almost nothing if you don't. The Small Projects description asks for a clearly scoped objective, a short execution window and modest use of compute and expert time. In other words, the factory is set up for teams that have already narrowed the problem to one process, one dataset and one success criterion.
Most Bulgarian companies start from a different place. According to figures cited by Capital on 7 November, about 8.5% of companies in Bulgaria used AI in 2025, against an EU average of 20%. For a company that has not yet picked its first process, a supercomputer is not the next step. The next step is deciding which process to test and how to know whether the test worked.
There is a second filter. If your use case is employees writing faster with Copilot or ChatGPT, you do not need a GPU cluster. The factory makes sense when you need to train or fine-tune your own model, run a model on data you don't want to send to a cloud provider, or check whether an idea is technically feasible at all.
What a well-prepared application looks like
BRAIN++ does not publish a template, so this is the practical list we would prepare before the first conversation:
- One process, described with numbers. For example: "We manually handle about 400 complaints a month and want to test whether a model can classify them by cause."
- A data sample you are allowed to share. Check what it contains: personal data, trade secrets, customer data. The data leaves the company, so processing terms need to be clear upfront.
- A success criterion before the start. What accuracy or saved effort would justify the next step.
- One person who decides. A week goes fast when every answer waits for three approvals.
- The next tier's terms, already read. If a Small Project shows promise, a private-company Big Project comes with the factory's participation in future commercial success. Better to know that before day one than after the last day.
What to watch
Two things will show whether the money works for business too: whether the €49 million survives into the final budget, and how many approved Small Projects go to companies rather than only research teams. If in 2027 the factory mostly serves researchers while companies walk past it, the reason will probably not be a lack of compute.
The same ministry is also preparing its own AI bill after the parliamentary committee rejected the previous one. Both tracks run in parallel, but for a company today the factory is the more practical news: the door is open, and the first project is free.
If you have not yet picked the process to start with, that is the work we do in AI Director as a Service: prioritising use cases, preparing the data and setting a success criterion before you spend time on compute.
