AI funding in Poland 2026: up to 83% for training (BUR, KFS)

A conversation about an AI implementation ends with “we don’t have the budget for this this year” more often than it ends with a substantive doubt. And usually this is not the truth about money - it is the truth about the calendar. Next year’s budget is already closed, and the AI topic appeared after it closed.

At the same time the opposite is happening. In the pilot call of the Dig.IT program, Polish companies submitted 146 applications worth over PLN 57 million against an allocation of PLN 20 million - almost three times the available pool. Public money for digitalization and AI exists, it just does not lie where most companies look for it, and it does not work the way most companies assume.

This guide is a map of funding for AI implementation and for team competencies, written from the perspective of a contractor, not a grant advisor. WeAreFuture is a Polish AI advisory and implementation firm based in Warsaw, listed in Poland’s Development Services Database (BUR) run by PARP, the Polish Agency for Enterprise Development. We deliver projects financed from these mechanisms, so we know them from the settlement side, not only from the rules. Here you will find four types of funding and how they differ, four paths - each for a different type of company - a decision table, a worked calculation for a specific project, and the situations in which funding makes no sense.

All figures are as of August 4, 2026. We deliberately do not give call dates - they change every quarter, while the mechanisms and eligibility conditions stay. For each instrument we link to the source where you can check the current status.

1. Three wrong assumptions that close off access to funding

Three patterns of thinking that repeat in our conversations regardless of industry and organization size.

Pattern 1: “We’ll wait for an AI grant.” No such program exists - with one exception, covered below. AI is not a separate budget category, it is an eligible cost in a dozen or so different instruments: once as software, once as an intangible asset, once as a development service, once as the payroll cost of an R&D team. A company that waits for “a call for AI” misses four calls in which its project qualifies.

Pattern 2: “Funding is a discount.” It is not. A discount lowers the price and ends at the invoice. Funding is a separate project with its own schedule, documentation, indicators and risk of repayment. It has its own order of steps that cannot be reversed: in Poland’s National Training Fund (KFS), expenses incurred before the contract with the labor office is signed are not eligible for funding, and there is no room for negotiation. A company that first orders the training and then looks for funding is out of the game from the start.

Pattern 3: “We’ll fit the project to the call.” This is the costliest of the three because it seems rational. A company finds an open competition and trims its scope to fit - it adds a component it does not need in order to meet a criterion. It gets the grant and is left with a project that does not solve its problem, yet has to be settled and its indicators maintained. The order should be the reverse: first the business decision, then the search for an instrument that finances it. If none fits, that does not mean the project is bad - it means you will finance it differently.

2. Four types of funding and how they differ

There are several dozen instruments, but only four types of funding. They differ in two things that decide everything: when the money is in your account and what you risk.

The order in the table is not accidental - it goes from the easiest to access to the most demanding. We describe the four paths further on in the same order.

The practical consequence: a grant is the most desired and the least certain, and the two most accessible types - reimbursement of a development service and tax relief - are at the same time the least glamorous. A company that wants to finance an AI implementation this year has a better chance with one of them than with a grant competition whose results it learns after eight months.

3. Path A - team competencies: the highest rates on the market

We start here because it is the simplest entry: there is no competition, calls are continuous, and practically every employer qualifies. If your bottleneck is not technology but the fact that people do not use the tools, you are in addition in the best-funded area of the whole list. Rates reach 83% here, and the entry condition is that the service provider is listed in the Development Services Database (BUR). From January 1, 2026 this is a statutory requirement for KFS - the earlier Register of Training Institutions no longer suffices in this respect.

The first observation from this table: the top three rows are almost equal. So the selection criterion is not the percentage but who qualifies and what the amount limit is - and those differ drastically.

FERS “Automation Competencies for Digital Transformation” (FERS is the EU-funded European Funds for Social Development program) has the highest rate and the best-matched scope: AI, automation, data analysis, IoT and cloud. Reimbursement reaches 83% of eligible costs, and the own contribution is at least 17% of the gross value of services. The maximum eligible costs per enterprise are PLN 111,425 for micro and small companies and PLN 222,850 for medium-sized ones. The program is served by six macro-regional operators - it is worth confirming the exact reimbursement level and limit with the operator responsible for your macro-region. Two important things: there is no industry (PKD) filter here - the only criterion is SME status - and that is exactly why it is the most widely available path for companies that do not find their code in the sector councils. Large enterprises, however, are excluded.

The Key Competencies for Sectors program is run by sector operators - one per sector, 26 in total - and finances up to 80% of the costs of training and development-oriented consulting. Large companies are admitted, but their employees may make up at most 40% of project participants, which means no guarantee of a place. Qualification starts with checking whether your company’s PKD code (Polish Classification of Activities) falls into one of the sectors - “industry” is not enough, the specific code counts. For companies from sectors without their own council a separate mechanism is provided.

HR Academy is one of the few instruments in which a large enterprise gets the full 80% - with a budget of PLN 108.2 million, for training and consulting in HR, including digital tools, analytics and AI in HR processes.

Regional PSF is a mechanism in which a voivodeship-level operator finances a service chosen by the company from the Development Services Database. The level depends on the voivodeship and company size - the range is 50-80%, with increases for a participant aged 55+ or for a service leading to a qualification in the Integrated Qualifications System. In some regions large companies are excluded from PSF entirely. An important operational detail: the payment model depends on the operator - in some projects the company finances the service and gets a reimbursement, in others the operator pays the provider directly. For cash flow that is a fundamental difference and a question worth asking in the first conversation with the operator.

We run our AI training for companies in this mode: the service description, the program with learning outcomes and the documentation on the provider’s side are produced by us, because the BUR procedure requires it.

KFS - the path best described in the regulations

KFS deserves separate treatment, because it is both the simplest operationally and the most formally treacherous. Everything below follows directly from the Act on the Labor Market and Employment Services (Art. 125-133), not from interpretation.

On top of that comes one condition you will not find in most guides, and which affects the decision: the company must keep the trained person employed for at least three months after the training ends. Violating this condition means no access to KFS funds for a year. If you plan training in a team that is undergoing reorganization, this is a risk to calculate before you submit the application.

Artificial intelligence is named explicitly in KFS 2026. National priority number 3 reads: “Support for continuing education in connection with the use of new processes, technologies and work tools in companies, with particular emphasis on digital skills, AI and so-called green skills”. The KFS budget for 2026 is about PLN 420 million, of which the predominant part remains at the disposal of county labor offices - and these run calls on a rotating basis and some close them within a few days of announcement.

The cost catalog is closed and covers courses, postgraduate studies, exams and certifications, as well as medical examinations and insurance required for participation. It does not cover accommodation, meals, travel or deductible VAT. The practical consequence for pricing: catering at a workshop has to be a separate item outside the application, and the price is quoted net.

A shared limit that is easy to forget

Most of the competency mechanisms described are de minimis aid, and it has one limit for the whole company: EUR 300,000 over a rolling three years, counted 36 months back from the day the aid is granted. “One enterprise” in the EU sense also includes linked entities, so a company in a group does not count the limit separately. Usage is checked by tax ID (NIP) in the SUDOP system run by UOKiK (the Polish Office of Competition and Consumer Protection) - and it is worth doing before the conversation about the project scope, not while assembling documents. An exhausted limit can close the whole path.

4. Path B - manufacturing SMEs: investment grants

This is the best-served segment. If your company is a small or medium-sized enterprise in industrial manufacturing, you have instruments available that nobody else has.

Dig.IT (the Industrial Development Agency, ARP) is the closest fit for a project of the type “AI implementation plus training people”. The grant is from PLN 150k to PLN 850k at funding of up to 50% of eligible costs, in the de minimis formula. A technology component is mandatory - implementing a ready digital technology or commissioning software development, including solutions based on AI and automation. Employee training in using the implemented technology is optional and may make up to 40% of eligible costs - one of the few grant instruments in which the competency layer has such a high limit. On our side this corresponds to combining an AI implementation with a training layer in one project, settled as two items in the same application.

Three conditions that eliminate more companies than the industry profile itself: at least five closed financial years, a free de minimis limit, and a grant not exceeding 30% of average annual net sales revenue from the last three years. The last one can be a surprise - it means a smaller company will not reach for the upper end of the range even if the project would justify it. Payment comes in one lump sum, as a reimbursement, so the project has to be pre-financed in full.

Inno_Lab AI Pilot (PARP) is the only Polish instrument that says outright, in its name and purpose, that it is about implementing artificial intelligence solutions. A grant of up to PLN 300k at funding of up to 70% of the project value, for SMEs from priority industries - among others industrial manufacturing, construction, and transport and warehousing. It is a pilot program within the Inno_LAB project running until 2029, so it is worth keeping on the radar regardless of the status of the current call.

Regional programs finance SME digitalization under the European Funds programs of individual voivodeships. There is no single rule here - parameters differ between regions and between calls. A fair generalization is funding up to 70%, with some calls having entry thresholds in the form of a minimum grant value of around PLN 500k. Eligible are usually intangible assets, fixed assets and implementation, in some calls training too. For example, the Silesian action “Digital Innovations in SMEs” names AI and big data explicitly in its goals.

SMART Path is an instrument with a completely different profile and the one most often confused with the others. It does not finance digital transformation - it finances the implementation of the results of completed R&D work. Funding reaches PLN 50 million at an intensity of 25-70%, with a minimum eligible cost value of PLN 3 million. From 2026 PARP abolished the modular construction and runs separate calls for R&D work and for implementation - which means that in the implementation call you must have research completed and documented before submitting the application. Upskilling may make up at most 15% of implementation costs. If your AI project does not result from your own earlier research work, this is not your path.

Two traps that spoil a well-planned budget

Subscriptions. A subscription for software in the SaaS model is an operating cost, not an intangible asset subject to depreciation. Investment grants finance the latter. In FEPW (European Funds for Eastern Poland) the provision was entirely unambiguous: “a monthly subscription for using SaaS software is not an eligible expense”. In Dig.IT long-term and perpetual licenses are preferred. The practical consequence: if the project is based on subscription tools, the cost structure has to be planned before the application is submitted, not after it is decided. This is not a universal rule and every set of rules has to be read separately - but it is the first thing to check.

Location. The Warsaw metropolitan region is excluded from regional investment aid and from the European Funds for Eastern Poland program. This does not mean a company headquartered in Warsaw has access to nothing - what counts is the place where the investment is carried out, not the registered address. This difference can be worth several hundred thousand zlotys and it is worth checking at an early stage, not at the stage of assembling the application.

5. Path C - large companies: preferential loan and tax reliefs

Here begins a problem that grant guides stay silent about: most digitalization instruments are limited to SMEs. A large enterprise - that is, the typical client of a serious AI project - drops out on formal criteria before anyone looks at the substance. Two options remain, both available.

The loan fund for digital and green transformation (PARP, FENG). This is the strongest argument in the whole list, because it covers micro, small, medium-sized and large enterprises from all over Poland. Parameters: a loan from PLN 2.5 million to PLN 15 million, interest of 0.5% per year, a repayment period of up to 15 years, a grace period on capital repayment of up to 24 months. The digital component directly finances the process of transforming the enterprise toward Industry 4.0. In addition, part of the capital may be forgiven - in the range of 5-15%, depending on the component type and the location of the investment, with higher rates applying to a specified list of cities and municipalities. Loans are granted by Financing Partners selected by PARP in a tender; their list is published after the procedure is decided.

It is worth setting this against the cost of money on the market. A 15-year loan at 0.5% with a two-year grace period is cheaper than inflation. For a large company that finances investments from credit anyway, it is an instrument cheaper than its own cost of capital - and available without a competition, on a continuous basis. A preferential loan is not a worse version of a grant. It is a different instrument, which works where a grant does not apply at all. For large organizations we build the AI Engine, a shared company context used by people and agents, in stages, process by process (collaboration model). The financing of such work usually comes together from a loan and tax reliefs, not from grants.

Tax reliefs are the second of them and the most underrated item in the whole list, because there is no call, competition or application - there is only record-keeping discipline.

• The R&D relief lets you deduct from the tax base 200% of payroll costs of employees and contractors carrying out research and development work, together with contributions, and 100% of other eligible costs. The 200% limit on employee costs is available to all taxpayers who conduct R&D, not only formal research and development centers.

• IP Box is a rate of 5% on income from a qualified intellectual property right - including the copyright to a computer program, without needing a patent. It requires separate records that make it possible to determine income from the qualified IP and calculation of the nexus ratio. It can be used together with the R&D relief, but without double deduction of the same costs.

• The innovative employee relief solves the problem of companies that are entitled to the R&D relief but have too low an income to deduct it. It allows reducing the personal income tax (PIT) advances of employees engaged in R&D for at least half of their working time - it turns an undeducted relief into a current cash benefit.

The key distinction that decides access to the reliefs: is your AI project the purchase of a tool or development work. Deploying a ready-made assistant under a license is a purchase. Building your own agent, integrating a model with an ERP system, developing your own data layer - that is usually development activity. The line is thin and disputed, which is why in this field the standard is to request an individual tax ruling rather than assess it yourself.

What the reliefs will not cover: the robotization relief requires the acquisition of a brand-new industrial robot, so AI software alone does not meet the basic condition - and in addition the instrument expires with the tax year that begins in 2026. The prototype relief concerns trial production of a new product, not internal implementations.

6. Path D - before you spend the first zloty

If you are not yet sure that your AI use case makes sense, there is a path that costs nothing. European Digital Innovation Hubs (EDIH) offer small and medium-sized companies free services in the de minimis formula: a digital maturity audit, consulting, training and “test before invest” - the ability to test a solution in the hub’s environment before incurring the costs of a commercial implementation. The second phase of the EDIH program is focused directly on AI.

We write about this knowingly, even though it is a substitute for part of our services. A company that does not know whether a problem is worth implementing should first check it - best of all for free. A conversation about a paid project makes sense only when it is known what is to be solved.

7. Decision table - which path for whom

Four questions that set your place in this table: what is your size status (this is not obvious - the EU definition takes into account linked and partner enterprises, and companies regularly classify themselves wrongly), what is your PKD code, where do you carry out the investment and how much free de minimis limit do you have. Only then does a conversation about a specific instrument make sense.

8. How much it covers on a concrete bill

Percentages in tables impress until you set them against a bill. Take an item from our training price list: a closed workshop for a team, tailored to the organization’s processes, in a range of PLN 7,500 - 25,000 net per group.

Two remarks that complicate this calculation. First, amount limits run out faster than percentages: with a limit of PLN 222,850 of eligible costs for a medium-sized company in the automation program you finance several workshop groups, not one - and it is worth planning full use of the limit right away instead of going back to the operator for a top-up. Second, with the reimbursement model you need the full amount at the start, so the question is not “how much will I get” but “do I have the money to pay the invoice before the reimbursement”.

For implementations the calculation works differently, because the scale is different. For a project in the range of several hundred thousand zlotys, a Dig.IT grant of up to 50% changes the investment decision, but it also introduces an eight- or ten-month delay and the risk of losing the competition. A 0.5% loan does not lower the project cost, but does not delay it by a day. This is a choice nobody can make for you: cheaper and later, or the same and now.

9. When funding makes NO sense

Four situations in which the honest answer is “finance it from your own funds”.

The call schedule diverges from the business need. If the problem costs you money every month and the nearest call will be decided in eight months, the cost of waiting exceeds the value of the grant. Calculate it instead of assuming funding is always beneficial.

The cost of administration exceeds the benefit. The application, documentation, settlement and indicators are the time of your people plus often the fee of a grant advisor. With funding of a few thousand zlotys the calculation simply does not close.

No ability to pre-finance. Most of these mechanisms work as reimbursement. A company that does not have the money to pay the invoice before the refund will not deliver the project - regardless of how well it qualifies.

The project exists in order to use the funding. This is the most expensive scenario and it is easiest to recognize by one symptom: nobody in the company can say which business indicator is to improve. You are then left with a reporting obligation, indicators to maintain and an implementation nobody uses.

We add three things that are worth knowing about, because they circulate as money for the taking and are not. The InvestAI initiative with EUR 200 billion is a target for mobilizing public-private capital at the level of the Union, not a grant pool for companies; the EUR 20 billion carved out of it concerns large-scale infrastructure and consortia, not applications from individual enterprises. The National Recovery Plan (KPO) is in its settlement phase and does not open new calls for company digitalization. The BGK technology loan (BGK is the Polish national development bank) has no confirmed call in 2026. If someone offers you participation in any of these three as an available source of funding, it is worth asking for a link to the rules.

10. How we approach this at WeAreFuture

We will be precise, because this distinction matters for what you can expect from us.

What we do NOT do:

• We do not write grant applications and we are not a grant advisory firm. That is a separate profession and a separate responsibility.

• We do not promise that funding “will turn up”. Eligibility depends on your PKD, company size, location and de minimis limit - and this is checked before the conversation about scope, not after the agreement is signed.

• We do not build the project scope to fit the call criteria. If an instrument requires a component you do not need, we say so plainly.

What we do:

• We are listed in the Development Services Database, with the SUS 3.0 quality certificate - and the BUR listing is a necessary condition for our training to be financed from KFS and PSF.

• We are a partner of the technology providers we implement and teach: Microsoft Solutions Partner, OpenAI Select Partner in the OpenAI Partner Network and the Claude Partner Network. With funding this has one practical meaning: we teach specific tools that we implement ourselves, not generic “AI” - and it is the scope and learning outcomes that decide whether the service complies with the priority.

• We prepare the documentation on the provider’s side: the service description, the program with learning outcomes, the justification of compliance with the priority - that is, those elements of the application that would have to be produced at our end anyway.

• We indicate which path fits your situation, together with the conditions that may exclude you from it.

• We work with a grant advisor on your side if the project justifies their involvement.

The filter we apply to every project is the same regardless of the source of funding: does this project lead to a lasting change in how the organization works, or is it a one-off event. Funding lowers the cost. It does not change the answer to this question.

11. What next

The simplest next step is to establish four parameters of your company: size status, PKD code, place of carrying out the investment and free de minimis limit. These four figures narrow several dozen instruments down to two or three - and only then is it worth reading the rules.

If you are looking for a picture of the implementation itself - architecture, stages and costs - we described it in the guide to AI implementations. If the decision concerns the competency layer, program levels and price ranges are in the guide to AI training for companies. The broader context in which both of these elements come together is in the article on AI transformation.

And if you have a specific project and want to check whether it can be financed - write to us. We will also tell you if in your case the answer is “it can’t, and it’s better not to wait”.

FAQ