How to fund an AI project in France: CIR, CII, Diag Data IA, France 2030
CIR, CII, Diag Data IA, Osez l'IA: a practical map of French public schemes to fund an AI project (agents, automation, product), with official rates and the MediaGen / AFP France 2030 case.
An AI project costs money before it pays off: salaries, cloud, integration, business time. In France, several public schemes can cover part of that — tax credits, a subsidised diagnostic, France 2030 calls. They are not interchangeable, and they do not stack in arbitrary ways.
Here is a practical map for a mid-sized to large company that wants to fund an agent, an automation, or an AI product. Rates and amounts below come from official sources (Service-Public Entreprendre, Bpifrance, France Num). Always check your case with your accountant or tax adviser: this is not legal advice.
At The Vibe Company, an AI agency founded in Paris in 2024 and backed by Y Combinator (W24), we have seen these schemes from the company side (our own CIR) and from the project side (MediaGen with AFP under France 2030). We build products and agents; we do not replace a tax firm filing a dossier on your behalf.
Four levers, four logics
| Scheme | What it is for | Who it targets (roughly) | What it actually funds |
|---|---|---|---|
| CIR | R&D tax credit | Companies running research work | A share of eligible R&D spend (not a plain off-the-shelf tool rollout) |
| CII | Innovation tax credit | SMEs (EU definition) | Design of prototypes / pilot plants for a new product, for spend until 31 Dec 2027 |
| Diag Data IA (Bpifrance) | Subsidised data / AI diagnostic | SMEs and mid-caps (often 10–2,000 employees per Bpifrance) | About 8 expert days to frame use cases — not the build |
| France 2030 / Osez l'IA | National AI adoption and innovation plan | SMEs, mid-caps, sometimes consortia (depends on the call) | Diagnostics, accelerators, selective project calls (e.g. MediaGen) |
The sequence we see most often: clarify the business problem first (Diag or internal audit), ship something useful, then structure CIR / CII around work that is genuinely R&D or product innovation — not the other way round.
CIR: the best-known lever, and the most misunderstood
France’s research tax credit (CIR) supports research and development. Per Service-Public Entreprendre, in mainland France the rate is 30% on eligible spend up to €100m, and 5% above that. In overseas territories, the first-tier rate is 50%.
It is not a cheque for “doing AI”. A standard SaaS rollout, a ChatGPT training course, or an integration with no technical uncertainty rarely qualifies. Work that creates new scientific or technical knowledge — models, evaluation pipelines, architectures, documented feasibility — may, under conditions.
We wrote our own CIR file with AI help, recovered €150,000, and published what worked (and what you must not let a model invent) in two articles (French):
For an AI project inside your company: document as you go (hypotheses, failures, technical choices, time spent). A CIR is built over time, not the night before the tax closing.
CII: for SMEs designing a new product
The innovation tax credit (CII) extends the CIR for SMEs only (under 250 employees, and turnover under €50m or balance-sheet under €43m, as recalled on Service-Public). It covers design of prototypes or pilot plants of a new product, for expenses incurred until 31 December 2027.
Mainland rate per Service-Public: 20%. Eligible expenses taken into account are capped at €400,000 per year. Overseas territories: 60% is indicated; in Corsica, 40% for small enterprises and 35% for medium ones.
CII does not cover the production phase of the prototype. It is also not a catch-all for any internal automation: it targets a new product meant to go to market later, not a purely internal tool with no novelty.
For an SME building an AI agent or product to sell (or a differentiating module), CII can sit next to CIR — as long as you do not double-count the same spend. Your tax adviser decides.
Diag Data IA: frame before you spend
Bpifrance’s Diag Data IA is a data / AI diagnostic, not a build budget. Per Bpifrance and diag.bpifrance.fr:
- max quote €10,000 excl. VAT;
- 40% subsidy (funded under France 2030 / Osez l'IA);
- €6,000 excl. VAT left for eligible SMEs and mid-caps under 2,000 FTEs;
- about 8 expert days to assess data/AI maturity and prioritise use cases.
France Num states that 2,000 Diag Data IA slots are co-funded for 2026–2027, subject to availability. Typical eligibility cited: SMEs / mid-caps (often from 10 employees). Apply via the Bpifrance platform.
What the Diag does not do: code your agent, wire your APIs, or guarantee ROI. It helps you avoid starting the wrong project. After that, you still have to build — in-house, with an agency, or both.
France 2030: from diagnostics to project calls
The Osez l'IA – France 2030 plan (Bpifrance, DGE, SGPI) bundles several offers for SMEs and mid-caps: self-assessment, use-case platform, Diag Data IA, AI accelerators, and more. It is not one counter with one cheque: each offer has its own rules.
Separately, France 2030 project calls fund heavier R&D or consortia. We have a public example: MediaGen, led by AFP with Quivr / The Vibe Company, Kairntech and Dawex — a France 2030 generative-AI project for newsrooms, backed by Bpifrance, listed at €4m on our site. Goals: augmented monitoring, multimodal archive search, production assistance under editorial control, with some components open-sourced.
That kind of call is not the standard path for a mid-cap that wants a support agent in six weeks. It is selective, slow, often consortium-based. Worth knowing if your AI project has a breakthrough or public-interest angle (health, media, industry…). For a classic business deployment, look first at Diag + build + CIR/CII.
How to choose for your project
You do not yet know where AI creates value. Start with framing (Diag Data IA or an internal audit). Prove one use case on your real data before buying a licence or launching a six-month build.
You have a clear case (support, ops, product) and need something that runs. Fund the build and the run first. CIR / CII do not replace a product budget; they can lower the net cost if the work is eligible and well documented. See also our checklist from AI agent POC to production.
You are an SME building a new AI product. Review CII + CIR with your adviser. The official CII cap (€400,000 of expenses) and 20% mainland rate give an order of magnitude — not a recovery promise.
You are aiming at a structuring France 2030 grant. Expect a call calendar, partners, and real admin work. MediaGen / AFP shows it is possible; it is not the shortest path to a first agent in production.
In every case: do not dress a non-R&D project as CIR. An audit costs more than an inflated file.
What we do (and what we do not)
The Vibe Company was founded in 2024 by Stan Girard and Antoine Dewez (ex-Theodo, ex-Stockly), YC W24, $1.7M raised. We have shipped dozens of projects: LocService (support), Monka (healthcare), MediaGen / AFP (France 2030).
We help build the agent or product, ship it to production, and document the technical work in a usable way. We ran CIR on our own activity. We are not a public-funding consultancy: for the tax filing, a ruling, or a project-call montage, work with the right professionals (accountant, CIR/CII adviser, Bpifrance).
If you are still choosing between agency, IT services firm, or an in-house team, read our guide on how to choose an AI agency.
FAQ
Does CIR fund any AI project?
No. CIR targets R&D. Many adoption projects (wiring an LLM to your CRM, automating tickets) are not, or not fully, research. Start from the official Service-Public text and your tax adviser.
Can Diag Data IA and CIR be combined?
Different logics: the Diag subsidises a diagnostic; CIR is a tax credit on R&D spend. Subsidies received for the same project can reduce the tax-credit base. Have your accountant confirm stacking rules.
How much do you get back with CII?
In mainland France, Service-Public states 20% of eligible expenses, within a €400,000 expense cap. Outside eligible spend and above the cap, nothing. Cash recovery also depends on your tax position.
Should you wait for a France 2030 call before starting?
Rarely. If the business need is clear, ship a first deliverable. France 2030 calls are selective and calendar-bound; Diag Data IA and CIR/CII often fit a “classic” company project better.
Does The Vibe Company file CIR dossiers for clients?
We build the projects and document the technical work. Tax filing sits with your finance team / adviser. We published our experience on our CIR; that is not a tax-representation offer.
Let’s talk about your project (and how to fund it)
You have an AI use case in mind and want to know what actually gets funded — build, Diag, CIR/CII, or a France 2030-style call? Bring the scope, company size and timeline. We will say what we can build, what fits a Bpifrance diagnostic, and what to ask your accountant.
