Skip to main content
Bluecoders

AI and deeptech recruitment in France in 2026: roles in short supply, key skills and salaries

Géraud MesnardNovember 22, 2026

2026 edition · Sources consulted in November 2026

In France, AI hires little but expensively, and always in the same places. Between 2019 and early 2025, just over one job posting in 1,000 (0.13%) was for an AI role, and 6.2% of large companies posted at least one, against 0.6% of SMEs (Aghion, Bergeaud, Bunel and Delbouve, July 2025). Deeptech accounts for 50,000 direct jobs, and the French state is targeting the creation of 100,000 direct jobs by 2030 (updated Deeptech plan, June 2026). At the same time, only 49% of companies struggled to hire managers in 2025, the lowest level since 2022 (Apec, the French management employment association). AI profiles remain the exception: in our 2026 barometer, the median pay of our deeptech, AI and data placements reaches €75,000 gross a year, against €65,000 across all sectors.

AI recruitment has changed in nature over two years. In 2024, a start-up was looking for "a data scientist". In 2026, it is looking for an AI engineer who can wire a language model into a product, an ML engineer who can train one on its own data, or a research engineer fresh from a lab. Three roles, three talent pools, often one single job title. The market for managers has eased; the market for AI and deeptech profiles much less so. This page brings together the reference public figures and what we see first-hand on our clients' AI roles, from the lab-born start-up to the group building its R&D team.

Key takeaways

  • A narrow, concentrated market: 176,392 AI job postings between 2019 and early 2025, or 0.13% of postings, published first by large companies (Aghion et al., 2025).
  • A measurable employment effect: four years after hiring AI profiles, companies have around 10% more jobs than comparable companies (same study).
  • A deeptech sector aiming to scale up: 50,000 direct jobs from the first plan, a target of 100,000 direct jobs created by 2030 (DGE, the French business directorate, June 2026).
  • AI skills expected everywhere: 53% of mid-caps and large companies will give more weight to AI skills in their management hires (Apec, May 2026).
  • Widely scattered salaries: a €54k median for big data managers, but 10% above €98k (Apec); €75,000 across our deeptech, AI and data placements.
  • A severe funnel: on our AI roles, 7 hired for every 100 candidates presented between November 2025 and October 2026.
  • A delayed regulatory timeline: the AI Act's rules on high-risk systems, including those used for hiring, apply from 2 December 2027.

How is this resource built?

What the page covers. Recruitment for AI roles (AI engineer, ML engineer, data scientist, research scientist, MLOps, AI leadership) and technical profiles at deeptech start-ups, in France, on permanent contracts.

What it does not cover. Training and certifications (covered in our resource on AI training for developers), and how recruiters use AI themselves.

The sources. Public bodies, sector observatories and research papers, listed at the end of the page with their date. The Aghion et al. study draws on Dares's (the French Ministry of Labour's statistics department) JOCAS database (postings collected from around fifteen job boards). The Stanford AI Index measures something else: the share of postings that ask for an AI skill, whatever the role.

Bluecoders data. It comes from our Marvin ATS and our 2026 salary barometer (69 permanent placements, August 2025 to August 2026). For the funnel, we tracked the 18 AI roles at our clients that received candidates between November 2025 and October 2026, identified by their title (AI, ML, data science, research, MLOps). This is internal data, not a market measure.

How many jobs and hires in AI and deeptech in 2026?

Which companies hire AI profiles?

The most precise study on the topic, Diffusion de l'intelligence artificielle en France (Spreading artificial intelligence in France, Philippe Aghion, Antonin Bergeaud, Simon Bunel and Paul Delbouve, July 2025), screened more than 120 million job postings. It identifies 176,392 for an AI role between January 2019 and January 2025, or 0.13% of postings. The share fell back after 2021, then rose sharply again from early 2023, driven by generative AI.

AI recruitment remains a large-company affair: 6.2% of large companies posted at least one AI role, against 2.9% of mid-caps and 0.6% of SMEs. The authors also measure what these hires produce: four years after hiring AI profiles, a company has around 10% more jobs than a comparable company, and its revenue grows in similar proportions. AI hired in-house creates jobs. Off-the-shelf AI has no measured significant effect.

Another measure circulates: according to the Stanford AI Index 2026, 0.99% of job postings published in France in 2025 asked for an AI skill (Lightcast data), against 4.69% in Singapore, which tops the ranking. This figure counts every posting that mentions AI, including for a salesperson or a project manager. Aghion et al.'s figure only counts roles that design or deploy AI.

How big is deeptech in jobs?

According to the Direction générale des entreprises (the French business directorate), the first Deeptech plan delivered 50,000 direct jobs, high value-added, at deeptech companies. Its update, published on 29 June 2026, sets a target of creating 100,000 direct jobs by 2030, and keeps the target of 500 deeptech start-up creations a year (close to 400 in 2024). Creating that many jobs in four years means hiring PhDs, research engineers and industrialisation profiles that labs and large groups are also chasing.

How many hires are expected?

No public source forecasts hiring for AI alone. Two benchmarks frame the market:

  • Apec forecasts 61,160 hires of IT managers in 2026 (+4% year on year) and 52,000 in studies and R&D, out of 305,800 management hires in total. The association cites the rise of AI among the drivers of demand, without putting a figure on it.
  • OPIIEC (the observatory of the digital, engineering and consulting sector) estimates close to 45,000 jobs to be created over three years under the effect of AI in the sector, including 31,600 in digital (study, April 2025). 28% of companies in the sector have already created AI-dedicated roles.

The general context points the other way: according to Apec, 49% of companies struggled to hire managers in 2025, 5 points fewer than in 2024 and 15 fewer than in 2022. The management job market is easing. On our AI roles, we do not see that easing: 7 hired for every 100 candidates presented (see below).

What are the most sought-after roles in AI?

The Aghion et al. study ranks roles by the number of AI postings they published. IT developers come far ahead, ahead of research and R&D roles.

Role (French ROME classification)AI postings, 2019 to early 2025
IT developer (M1805)24,931
Socio-economic studies officer (M1403)15,249
Industrial R&D engineer (H1206)6,104
IT systems and network expert (M1802)2,233
Scientific research engineer (K2402)2,194
Chief information officer (M1803)1,494
IT systems functional consultant (M1806)1,421

Source: Aghion, Bergeaud, Bunel and Delbouve, Diffusion de l'intelligence artificielle en France, July 2025, table 3. The authors note that automatic ROME code attribution carries errors.

The ROME classification has no code for "AI engineer" or "ML engineer": these roles spread across developers, studies officers (where data scientists fall) and research engineers. France Travail's (the French public employment service) 2026 Besoins en main-d'œuvre (labour needs) survey does not isolate them either, but it does measure the pressure on their family: 57.4% of hiring plans for IT and telecoms R&D engineers and managers are rated difficult, against 43.8% across all occupations, out of 18,330 plans.

Which profiles do our clients hire?

From November 2025 to October 2026, 18 AI roles at our clients received candidates. They break down as follows:

  • 7 AI engineer roles, often a company's first hire ("Founding AI Engineer") or at Staff level: integrating language models into a product, building agents, putting them into production.
  • 4 ML engineer or data scientist roles, to train and evaluate models on proprietary data: medical biostatistics, LLM safety, operations research.
  • 3 AI leadership or architecture roles, including a sovereign inference platform.
  • 4 other roles: research (AI scientist, computer vision), MLOps and product.

A few published placements, with our clients' agreement:

  • Alpha10X: a Staff AI Engineer.
  • White Circle (AI model safety): a Senior ML Engineer LLM.
  • PhysCade (medtech): a Lead AI R&D and a Senior Data Scientist / ML Engineer.
  • Diago.IA: an ML Engineer, in 6 weeks.
  • TheraPanacea (medtech): its first Head of Data (see the case study).
  • Streem Energy: a Product Builder.

These hires share one trait: our clients are not looking for a generic "AI" profile, but someone able to take a model from notebook to product, in their domain.

What skills are becoming strategic in AI?

OPIIEC distinguishes two blocks. Developing AI: data science, data governance, mathematics, infrastructure, monitoring and AI project management. Using AI: knowing the solutions, mastering the legal framework (GDPR, AI Act), knowing how to use it and supporting change. Companies are set to look more closely at these skills: according to Apec, 53% of mid-caps and large companies, 39% of SMEs and 31% of micro-businesses plan to give more weight to AI skills in their management hires. Today, 2 companies in 10 already factor it into selection.

On our AI roles, three skill families make the difference in 2026.

From model to product: what should an AI engineer know how to do?

Wiring a language model into an API is no longer enough. Teams are looking for engineers who know how to build agents, write evaluations (evals), measure the quality of a response and decide when a model cannot be trusted. Conversations with our clients' technical leadership point the same way: when agents write part of the code, review shifts toward specifications and architecture. The profile in demand is a product engineer with real architecture foundations, able to judge what an agent produces. An average engineer equipped with AI simply produces more average code.

AI applied to science: why does the dual skill set matter?

In deeptech and medtech, a model is only as good as the domain it serves. Our searches cover data scientists trained in biostatistics, computer vision engineers and AI scientists fresh from labs. The rare skill lies at the meeting point of machine learning with physics, biology or signal processing, and the validation constraints that come with it.

Inference and MLOps: who keeps the models running?

A model in production is expensive to run and must stay available. Our AI roles include MLOps engineers and AI platform architects, up to the technical leadership of a sovereign inference platform. These profiles often come from DevOps, high-performance computing or the cloud. They are rare because few companies have run models in production long enough to train them.

How much does an AI engineer earn in France in 2026?

No public source publishes a pay scale for AI roles alone. The most solid benchmark remains Apec, by job family, supplemented by our own placements.

What is the market benchmark?

According to Apec (Les rémunérations des cadres dans 111 familles de métiers, 2025 edition, published December 2025), the median pay of managers, fixed plus variable, is €55k gross a year. Data scientists are classed in the big data family. ML engineers and AI engineers do not appear under that name in any family.

Apec job familyMedian80% of managers earn between
R&D management€65k€43k and €104k
Big data (including data scientist, data engineer, computer science and mathematics researcher)€54k€40k and €98k
Industrial and technical IT€51k€39k and €71k
Software development€50k€38k and €74k
Business intelligence and analytics€50k€39k and €75k

Source: Apec, Les rémunérations des cadres dans 111 familles de métiers, 2025 edition. Gross annual pay, fixed plus variable.

In the big data family, 10% of managers earn more than €98k, almost as much as in R&D management (€104k). The median, however, stays close to that of all managers.

What do our own placements show?

In our tech and engineering salary barometer (69 permanent placements signed between August 2025 and August 2026), the median pay of our placements at deeptech, AI and data clients reaches €75,000 gross a year. That is the best-paid sector in the barometer, ahead of aerospace, defence and industry (€64,000) and business and solutions tech (€62,000). The median across all our placements is €65,000.

The Data, AI and ML family shows a lower median, €62,000, but a third quartile at €105,000 and a maximum at €150,000. Two lessons follow:

  • The job title does not set the salary. In the Data, AI and ML family, the gap between the median and the third quartile exceeds €40,000: integrating an existing model and designing one are not paid the same.
  • The company's sector weighs more than the role. Our deeptech, AI and data placements sit €10,000 above the median of all our placements.

On senior AI-oriented profiles, our clients compete with the best-funded AI labs. We flag this from the brief stage onward: a budget set at the market median will not attract these candidates.

How long does it take to hire an AI profile, and why?

According to our data, Bluecoders' median time-to-deal is 22 days in 2026, across all hires, between opening the role and the accepted offer. Rare AI profiles take longer: our published hire of an ML Engineer for Diago.IA closed in 6 weeks.

Why do AI hires take longer?

Three causes recur in our research:

  • A narrow, highly sought-after talent pool. Experienced AI engineers are constantly approached by start-ups, labs and large groups. Those who have already put a model into production are even fewer.
  • Blurry job titles. A poorly scoped "AI engineer" role attracts developers who use AI, not engineers who build it. The sorting then happens at interview stage, too late.
  • Processes that run too long. Candidates who have a choice drop out of five- or six-step processes. We recommend three steps at most, with a first technical conversation led by an engineer from the team.

What we observe on our AI roles

Across our clients' 18 AI roles between November 2025 and October 2026:

  1. Presented to the client104 · 100 per 100 presented
  2. Interviewed by the client74 · 71 per 100 presented
  3. Shortlisted16 · 15 per 100 presented
  4. Hired7 · 7 per 100 presented
StepCandidatesPer 100 presented
Presented to the client104100
Interviewed by the client7471
Shortlisted1615
Hired77

Source: Bluecoders data (Marvin ATS), AI roles (AI, ML, data science, research, MLOps titles), November 2025 to October 2026.

Every presented candidate has already been qualified by our consultants. Out of 100 presented candidates, 71 are interviewed by the client, but only 7 are hired. Pre-selection convinces; making the shortlist much less so: this is the stage where technical evaluation, pay level and competing offers play out. By comparison, our Defence, Industry & Hardware practice hires 6 candidates for every 100 presented (see our resource on defence recruitment).

What constraints weigh on AI recruitment?

Two topics recur in our clients' briefs: the AI Act's timeline and the place of the PhD.

What does the AI Act change for recruitment?

The European AI regulation applies in stages. The digital omnibus on AI, in force since 27 July 2026, pushed back its timeline: the rules applicable to high-risk AI systems under Annex III apply from 2 December 2027, and those for systems embedded in products from 2 August 2028, according to the European Commission. Annex III specifically targets employment: AI tools used to screen applications or assess candidates fall under it. The same text also simplified the AI literacy requirement imposed on companies.

For teams developing or deploying AI, this timeline creates a need for skills in model documentation, evaluation, risk management and compliance. OPIIEC already ranks mastery of the legal framework (GDPR, AI Act) among the key skills. On our roles, these skills add to the technical profile, rather than creating a separate role.

Do you need a PhD to work in AI?

On our roles, a PhD is often requested for research or an AI scientist role, rarely for a product-facing AI engineer or ML engineer. No public source measures the share of AI postings that require a PhD. Two schemes matter for companies hiring PhDs:

  • The "young PhD" scheme of the research tax credit was abolished by article 55 of the 2025 finance act. It allowed companies to count double, for 24 months, the staff costs of a PhD hired on their first permanent contract. Its reinstatement was not included in the 2026 finance act and is back under discussion in the 2027 budget review.
  • CIFRE agreements from the ANRT (the French national research-and-technology agency) remain open: the company employs a PhD student for three years, on a research project run with a lab.

For deeptech start-ups, abolishing the "young PhD" scheme makes the first PhD hire more expensive. The criterion we look at stays the same: the ability to carry a research effort through to a usable result, with or without a thesis. Career paths and training are detailed in our resource on AI training for developers.

Where is France's AI talent located?

AI postings are more geographically concentrated than job postings as a whole. According to Aghion et al., Paris, Saclay, Toulouse, Grenoble and Versailles-Saint-Quentin are among the employment areas that publish the most. Some areas stand out through specialisation: Toulouse with aerospace, Niort with insurance and financial services, Lannion with telecommunications.

The map of training follows the same logic. France 2030 funds nine IA-Clusters, centres of excellence in research and training, to the tune of €360m:

IA-ClusterLead universityFrance 2030 funding
PR[AI]RIE PSAIUniversité Paris Sciences et Lettres€75m
MIAI ClusterUniversité Grenoble Alpes€70m
Hi! PARIS Cluster 2030Institut Polytechnique de Paris€70m
PostGenAI@PARISSorbonne Université€35m
ENACTUniversité de Lorraine€30m
DATAIA-ClusterUniversité Paris-Saclay€20m
ANITI IA ClusterUniversité de Toulouse€20m
3IA Côte d'Azur 2030Université Côte d'Azur€20m
SequoIAUniversité de Rennes€20m

Source: Ministère de l'Enseignement supérieur et de la Recherche (the French Ministry of Higher Education and Research), announcement of IA-Clusters winners, May 2024.

For an employer, these clusters are also talent pools: PhD students nearing completion, research engineers, partner labs for a CIFRE thesis. The programme aims to train 100,000 people by 2030, according to the government. Some of our AI hires happen remotely or hybrid, which widens the talent pool beyond the client's local area.

Bluecoders' view

"An AI engineer role today covers three jobs: training models, integrating them into a product, running them in production. Before we search, we get the client to specify which one, because the talent pools, salaries and timelines have nothing in common. In interviews, we mainly assess the ability to judge what a model produces, not just the stack."

Géraud Mesnard, Account Executive, AI, Data & Science practice

Hiring an AI engineer, an ML engineer or a researcher?

Our AI, Data & Science practice recruits AI engineers, ML engineers, research scientists, MLOps profiles and AI leaders for deeptech start-ups, AI labs and corporate R&D teams. Discover our approach to deeptech, AI and science recruitment, or talk to our practice about your next hire. For a Head of AI or a deeptech CTO, see our executive search offering.

How to cite this resource

Reusing the figures and tables on this page is allowed, with a link back to the page. Citation format:

Bluecoders, AI and deeptech recruitment in France in 2026: roles in short supply, key skills and salaries, 2026 edition, Géraud Mesnard. https://www.bluecoders.com/en/ressources/recrutement-ia-deeptech

Journalists: key figures, visuals and contacts are in our press kit.

FAQ

What are the most sought-after AI roles in France?

AI engineers, ML engineers, data scientists and research engineers. According to Aghion et al. (July 2025), IT developers publish the most AI postings (24,931 between 2019 and early 2025), ahead of studies officers and R&D engineers. Among our clients' 18 AI roles between November 2025 and October 2026, 7 were AI engineer roles.

How much does an AI engineer earn in France in 2026?

Apec does not isolate AI. In its 2025 edition, the median pay of big data managers, the family that includes data scientists, is €54k gross a year, and 10% of them earn more than €98k. In our 2026 barometer (69 permanent placements), the median of our deeptech, AI and data placements reaches €75,000, and the Data, AI and ML family rises to €150,000.

How long does it take to hire an ML engineer?

According to our data, Bluecoders' median time-to-deal is 22 days in 2026, across all hires. Rare AI profiles take longer: our published hire of an ML Engineer for Diago.IA closed in 6 weeks. On our AI roles, 7 candidates are hired for every 100 presented.

Do you need a PhD to be hired in AI?

Not for most roles. In our hires, a PhD is often requested for research or an AI scientist role, rarely for a product-facing AI engineer or ML engineer. The "young PhD" scheme of the research tax credit, which favoured hiring PhDs on a first permanent contract, was abolished by the 2025 finance act.

What AI skills are companies looking for?

To develop AI: data science, mathematics, infrastructure, deployment and model evaluation. To use it: knowledge of the solutions and the legal framework (GDPR, AI Act), according to OPIIEC. According to Apec, 53% of mid-caps and large companies will give more weight to AI skills in their management hires.

Does the AI Act change AI recruitment?

It mainly changes the skills expected. Since the digital omnibus came into force on 27 July 2026, the rules applicable to high-risk AI systems under Annex III, including those used for hiring, apply from 2 December 2027. Teams are therefore looking for engineers able to document, evaluate and monitor their models.

Sources

A recruitment need?

Describe what you need. A recruiter from your sector will call you back within 24 business hours.