Luxembourg's R&D project aid covers a share of eligible project costs. Two things set the ceiling on that share: the category the work falls into, and the size of the company.
A small enterprise can be co-funded up to 60% of eligible costs on industrial research and up to 40% on experimental development; a large enterprise, 40% and 20%. Most founders meet that classification for the first time inside the application form. It is the largest single lever on the money.
Those rates are the maxima published on guichet.lu's R&D project aid page, read in August 2026; that page carries a last-modified date of August 2025. They are ceilings, not entitlements. The published wording is that the maximum aid may not exceed these thresholds, calculated on the project's eligible costs.
The rates, by category and company size
| Project category | Small enterprise | Medium enterprise | Large enterprise |
|---|---|---|---|
| Industrial research (base rate) | 60% | 50% | 40% |
| Experimental development (base rate) | 40% | 30% | 20% |
The size difference is not something you add on top. Under its enterprise-size category, the published grid of majorations attributes 20% to a small enterprise and 10% to a medium one, and that uplift is already the gap between the columns above.
Apply it a second time to the small-enterprise figure and you produce a number that exists nowhere in the scheme. Read off your row and your column; the cell where they meet is the ceiling for the project.
Notice how much the rows and the columns are each worth. A small enterprise doing experimental development lands on the same ceiling as a large enterprise doing industrial research. Those cells happen to coincide, they are published separately, and they can move apart. Either way, a step down the category ladder can cost a small enterprise as much as being a large one would.
The ladder of uncertainty
Public R&D funding across the EU runs on a shared logic: the further from the market and the deeper the uncertainty, the larger the share the state will carry.
Industrial research is the upper rung companies commonly reach: planned work to acquire new knowledge for products or processes, where the question is still whether the approach can work at all. The 60% small-enterprise ceiling belongs to that rung.
Experimental development is the workhorse rung: combining what is already known into new or improved products, prototypes and pilots. The uncertainty sits in the engineering rather than in the science, and for the same small enterprise the ceiling falls to 40%.
The through-line is technical uncertainty: work whose outcome a competent professional could not simply look up or routinely execute. Building your app's settings page is not R&D. Making an algorithm do something the literature says is hard — that might be.
The floor, and the ceiling that triggers a notification
The rate table has boundaries on either side of it that it does not show.
At the bottom sits a minimum on the aid itself. The gross aid on an R&D project cannot come out below €1,000 for a small or medium enterprise, or €100,000 for a large one.
For an SME that floor is low enough to be irrelevant most of the time, but it does rule out filing for a fragment of work: if the eligible costs are small enough that the aid would land underneath it, there is no application worth building.
At the top, the general conditions applying to all RDI aid, the set published under the loi du 6 juin 2025, carry notification thresholds per undertaking and per project: €55,000,000 where the project is principally fundamental research, €35,000,000 where it is principally industrial research, €25,000,000 where it is principally experimental development.
Those figures sit far above anything a young company files. They are here for a different reason: this is the one place the published framework treats fundamental research as a category with its own number attached. The rate table above covers industrial research and experimental development, and a company planning genuinely fundamental work cannot read its rate off the neighbouring rows.
Why founders misclassify, in both directions
The optimists call everything research and get corrected downward, spending credibility in the file to do it. The modest ones call genuine research "just development" and leave the gap between the rows sitting on the table. On identical eligible costs, that gap is the distance between 60% and 40% for a small enterprise.
The fix is the same in both directions: describe the uncertainty itself. What did you not know? What could have failed? What would a competent engineer have been unable to routinely deliver? Projects written around those questions classify themselves, and they read as work rather than as a pitch to the person who assesses technical files for a living.
What the file has to make legible
| Part of the file | What it has to make legible | Where it goes thin |
|---|---|---|
| The technical problem | why the outcome was genuinely in doubt | a product roadmap restated as if novelty were self-evident |
| The state of the art | what already exists, and why it is not enough | a literature paragraph that never compares itself to your approach |
| Work packages | which tasks carry the uncertainty, which are routine | an undivided package mixing both, assessed as the lower category |
| Costs | which hours and purchases belong to which package | headcount split by feel once the year is over |
The cost line is where files most often thin out, because it is the only part that cannot be written from memory. Hours recorded against work packages while the work happens survive a request for evidence; hours reconstructed afterwards read as reconstruction. The bookkeeping habits that keep an accountant calm are the same ones that keep this file defensible, applied to engineering time instead of invoices.
Structuring projects so the classification holds
Classification is assessed per project, which makes the project boundary a design decision rather than an administrative one.
- Separate the uncertain from the routine. A single project containing real research plus ordinary build work is assessed as one thing, and the routine half pulls the whole toward the lower category. Scoping them as separate projects holds each at its own level.
- Settle the sequencing before you scope, not after. Whether work already under way can still sit inside the project is a question to answer at the start; the traps around timing are covered separately in the funding application mistakes guide.
- Allocate costs while the work happens. Time tracking against work packages, and purchases tagged to a package when they are made, is what turns a technical story into an auditable cost base.
- Think in chains. On the growth side, RDI work pairs with the Young Innovative Enterprise scheme, where co-funding runs up to 70% of the projected financing need and the aid cannot exceed €1,000,000 per company. It also feeds the IP Box, where 80% of net eligible income is exempt under article 50ter L.I.R. There is a catch: the 80% exemption applies only to the share of net eligible income produced by the modified nexus ratio in article 50ter, alinéa 6 L.I.R. So the share that qualifies depends on who did the development, and outsourcing the R&D to a group company shrinks it. The same technical narrative, written once and carefully, serves all of them.
Common questions
These are the questions founders ask once the rates stop being abstract and the scoping starts.
Can software development qualify as R&D?
It can, but rarely the whole product. The qualifying part is the component whose outcome was in doubt: the model that might not have converged, the protocol that might not have held under load. The screens and integrations built around it are not that component. Where the component is the project, the ceiling is normally the experimental-development one, 40% for a small enterprise, unless the work genuinely sits at the industrial-research rung.
Does the SME bonus come on top of the small-enterprise rate?
No. The 20% attributed to small enterprises in the size category of the majorations grid is what makes the small-enterprise column higher than the large-enterprise one. The cell in the table is already the answer. Adding the uplift again overstates the ceiling.
Is there a minimum project size?
The published minimum sits on the aid rather than on the project: gross aid of at least €1,000 for a small or medium enterprise, and €100,000 for a large one, on each R&D project.
Does the same work feed the IP Box later?
Income from software or patents that come out of the project can fall under the IP Box, but only part of it is exempt: the 80% exemption applies only to the share of net eligible income produced by the modified nexus ratio in article 50ter, alinéa 6 L.I.R. The cost records built for the aid file are the kind of evidence the tax side asks for as well, which is a reason to build them once and build them properly.
What to do with this
Write the uncertainty down in plain language before you touch a form: what you did not know, what could have failed, why a competent engineer could not have delivered it routinely. Then decide which rung the work stands on, split the routine build out into its own scope, put time tracking against the work packages, and check the arithmetic against the ceiling for your row and column and against the €1,000 floor.
Turning that into a submission is a different job from doing the research: the project has to be scoped, the eligible costs allocated to it, and the application file and its annexes built around the same technical story. No file can make an uncertain project certain. The ones that read best say precisely where the doubt lies.