What an AI Engineer earns in Spain
476 job ads for the role posted in Spain between July and September. What they pay, who posts them, and what they ask for in return.
For the same AI Engineer title, a product company pays €69,000 and an IT services firm pays €49,400. That is €19,600 of difference for the same title.
A quick note on that second category, because it does not map cleanly onto English. What Spain calls a consultora here is not a McKinsey-style advisory: it is an IT services firm or body shop that employs you and bills you out to a client. That distinction runs through this whole piece.
It is a big gap, though there is still little data behind it (n=31). Of the 476 job ads for the role posted in Spain between 27 July and 18 September, only 31 publish a full salary band. That is 7%.
This is the second piece of the series, between what an AI Engineer actually does and who holds those jobs today. Here we look at the job ads; in the third one, at the people.
What does it pay?
For the same title, a product company pays €19,600 more than an IT services firm
Midpoint of the published band, across the 31 of 476 job ads that state a salary.
Product company n=15
Traditional employer n=3
IT services / staffing n=13
Difference
The full range runs from 30 to 130K a year. The median midpoint lands at €52,500, and the middle half of the ads sits between 44,250 and 63,750.
One in five goes above 80K, and between 70 and 90 there is almost nothing
Job ads by salary band. More than half fall between 40 and 60K.
ads per band · thousands of € gross per year
5
30-40
7
40-50
9
50-60
3
60-70
1
70-80
0
80-90
2
90-100
3
100-110
0
110-120
1
120-130
Before blaming the role, we went and measured the same gap across the whole market. We took the 475 technology ads that publish a band in the same window — out of the 19,940 in the market, not out of the 476 for the role — split IT services and staffing from the rest, and the difference in medians comes out at €11,000 (44,000 against 55,000).
The €19,600 for this role is almost double that gap. Role by role the difference varies a lot: in backend there is barely any, and in design or in AI and ML it is double the market average. These are cells of three to five ads a side, so it is enough to rank the roles and not enough to put a number on each one.
Two biases in these numbers. First, whoever publishes a band is not a representative sample: they tend to be foreign companies bound by pay-transparency rules, or product companies comfortable with what they pay. Second, with 13 bands from IT services firms, 15 from product companies and 3 from traditional employers, a single odd ad moves the midpoint by thousands of euros.
Who posts these ads?
42.6% of the 427 ads that state a sector are posted by an IT services, staffing or consulting firm, not by the company you would end up at. The heaviest posters are T-Systems, NTT Data, Logicalis, Keyrus, Capitole, GFT and Accenture: the top seven are all IT services firms, without exception.
The full split, grouped by the sector the poster declares:
| Sector of the poster | Ads | % |
|---|---|---|
| Consulting, staffing and IT services | 182 | 42.6% |
| Software and digital product | 122 | 28.6% |
| Industry, energy and engineering | 32 | 7.5% |
| Banking, insurance and finance | 25 | 5.9% |
| Health, pharma and research | 19 | 4.4% |
| Telecommunications | 9 | 2.1% |
| Other sectors | 38 | 8.9% |
427 unique ads carry a sector. Three in ten ads come from a company selling its own product and four in ten from someone who will place you at a client. The rest is split between industry, banking, pharma and telecoms in small pieces.
And almost nothing junior. Only 16 of the 476 (3.4%) carry junior, trainee or intern in the title.
What do the ads ask for?
An AI Engineer ad names RAG eleven times more often than a typical tech ad
Mentions in the ad text, narrowing the focus from the whole market down to a single title.
All technology ads n = 19,940AI and ML roles only n = 1,136AI Engineer only n = 476
Mention RAG
Mention agents
Mention LangChain or LangGraph
From 5.1% to 56.7% is eleven times. The frameworks multiply by fourteen and agents by five and a half as the focus narrows.
What does not follow is that searching for those words will lead you to the role. In Spain there are 1,017 technology ads that name RAG and only 270 carry AI Engineer in the title: three in four sit under some other job title. The vocabulary has already outgrown the role.
To compare against the international dataset we had to switch cohorts, because theirs is not only jobs with AI Engineer in the title: it also includes platform and data. So we took the wide set, every AI and machine learning ad, and placed it alongside:
Spanish AI job ads ask for practically the same things as international ones
Difference in percentage points between what Spanish ads ask for and what international ones ask for.
Agents
+5.5 60.9% vs 55.4%
Python
+4.9 75.7% vs 70.8%
RAG
+0.1 39.9% vs 39.8%
MCP
−2.7 11.4% vs 14.1%
Figures: Spain, all AI and ML ads vs international, 6,964 ads.
The Spanish side is the 1,136 AI and machine learning ads from the same window, not just the 476 titled AI Engineer, because the international dataset also includes adjacent platform and data roles. The two columns are not counted the same way: we search for words in the ad text and the international dataset extracts skills with a model that also groups synonyms. For what gets written by name — RAG, Python, MCP — the two counts converge. For what depends on synonyms they do not, which is why evaluation is handled separately, just below.
None of the four separates much. RAG comes out level, MCP three points below, and agents and Python five points above. What an AI job ad asks for in Spain looks a lot like what one asks for elsewhere.
Evaluation we cannot call. It was the signal we expected to come in below, and we cannot treat the result as solid. It depends entirely on where you draw the line on what counts as evaluation.
The international figure falls inside our margin of measurement
How many ads ask for evaluation, depending on where you draw the line on what counts as evaluation.
each band runs from counting the word alone (left) to counting the whole group (right)
Ads for the title n = 476
All AI and ML ads n = 1,136
International, AI-First roles
With what we know how to measure there is no gap to report. The interval runs from 45.6% to 68.9% depending on where you cut, and the international 59.7% lives inside it.
On either count, evaluation shows up in more than half the ads for the role. It is not marginal.
That comparison is against other countries. There is another one that says more, and it is against the Spanish market itself: what does an AI Engineer ad ask for that no other technology ad posted here asks for?
LLM frameworks are what most sets this role apart: they appear 21 times more often than in the rest of the market
Mentions in the ad text. Same counting method and same window in both columns.
AI Engineer rest of tech
LLM frameworks
21×
RAG
15×
LLM
8.7×
Python
2.7×
SQL
0.8×
n = 476 ads with AI Engineer in the title and 19,464 from the rest of technology, Spain, 27 July to 18 September 2026.
The real differentiator is the frameworks, not RAG, which is what gets the headlines. A LangChain or a LlamaIndex appears 21 times more often than in the rest of the market; RAG, 15.
An LLM framework is the plumbing between your code and the model: it chains calls, decides what context goes into each one, connects the tools the model can use, and holds the state of the conversation. Without one you end up writing that layer by hand. The three we group here are LangChain, the most widespread, LangGraph for agent flows, and LlamaIndex, more oriented to RAG.
SQL is asked for less in an AI Engineer ad (23.3%) than in a typical tech ad (27.8%). Of every signal that separates this role from the rest of the market, the only one that goes down is the data one.
In our own searches that turns into something clients find hard to accept: the team’s data engineer is almost never the right candidate. On paper they look closest. In practice the job is building and maintaining a production service against your data, and a senior backend engineer with cloud experience fits that better than someone who has lived in pipelines.
Back to evaluation: we do not know whether it is asked for less here, but we do know that whoever is hiring cannot read it off the ad. One in two names it, and naming it says nothing about whether the candidate can do it. If you are hiring, one question in the first technical interview covers it: when your system started failing, how did you measure it, over how many cases, and what changed the number? Why that question and not another one is in the first piece of this series.
How much competition is there for these jobs?
LinkedIn does not give the exact figure: below 25 applications per ad it says 25, and above 200 it says 200. Across the whole tech market — the 19,940, including the ones the table below leaves unclassified — 33.0% of ads read exactly 25 and 11.1% exactly 200, with not a single value below or above. Almost half the data is pinned to a cap. In AI Engineer ads the split is different: 23.3% at the floor and 16.6% at the ceiling.
With that, the mean does not mean much. So instead of one figure per role we show the whole split and the median.
Frontend408
105
Design552
100
Full-stack978
73
AI Engineer476
67
Mobile291
61
Product1227
56
Backend1662
55
QA750
54
Data1931
54
AI+ML*672
47
Generalist software2034
41
Security1199
39
DevOps and platform1588
28
* AI+ML counts the AI and machine-learning area without the AI Engineer title, so the two rows do not overlap. n = 19,940 technology ads posted in Spain from 27 July to 18 September 2026. The market median is 45.
The median AI Engineer ad draws 67 applications, against 45 for the market. It is the fourth most applied-to role of all, ahead of backend (55) and data (54), and behind only full-stack (73), design (100) and frontend (105).
The top end confirms it. 37.4% of AI Engineer ads go past 100 applications, and only 23.3% stay at the floor, against 47.1% for DevOps.
Two things worth separating, because they get confused. AI and ML ads excluding this title have a median of 47, two points above the market. With the title counted in, the whole area rises to 52. And AI Engineer alone, to 67. The pull does not come from AI; it comes from this title.
Those applications do not mean the role is easy to fill. In a sample of more than 1,800 AI engineers analysed in Spain, only 2.3% show up as open to work. Plenty of applications and almost nobody genuinely available, both at once. We take that contrast apart in the third piece of the series.
Recommendations
If you are hiring, three things follow from these numbers.
Four in ten of the ads you are competing against are posted by an IT services firm, and their midpoint is €49,400 against €69,000 for a product company. If you are a product company, that difference is your best argument, because from the outside the two ads look alike. Make it clear that you are hiring for your own product and not for an end client.
Publishing a salary band sets you apart from the 93% who do not. With this sample we cannot tell you how many bad-fit applications it saves you, but we can tell you it differentiates you today.
And above all, do not post and pray. An ad for this role gathers 67 applications at the median, more than backend or data, but only 2.3% of the AI engineers we have analysed show up as open to work. The volume arrives; the people you want to hire do not, because they are not looking at ads. Those you have to go and find one by one, open the conversation yourself and keep it going. It is how we work at nothiring, and it is also what the profile of the people in these jobs says: where they are, where they come from and what state you will find them in.
If you are looking for the job, look at who posts the ad before you look at the title. Product companies and IT services firms pay differently for the same job, and 42.6% of the ones that declare a sector are not the company you would end up at.
Data and method
| What it is | LinkedIn Spain postings with the full description |
| Window | 27 Jul – 18 Sep 2026 |
| Base | 26,532 postings, 19,940 unique technology ads |
| Cohort | 476 unique ads with AI Engineer or a variant in the title |
| Salary bands | 31 |
How we measured it and what we cannot claim
The salary bands are done by hand. Twenty-one ads fill in the structured field; the rest come from reading the ad and separating the salary from the meal vouchers, a standard Spanish benefit-in-kind that otherwise inflates the figure. We discarded one band with a scale error (“€90/year”), two freelance rates (hourly and daily), one ad from Lisbon, one figure that was not from Spain, and five that gave only one end (“from 30,000”, “up to 80,000”). That leaves 31 out of 476 ads, and it is a self-selected sample. Three of those 31 are the same Factorial role — Factorial is a Barcelona HR-software company — posted in Madrid, Barcelona and A Coruña with the same band: they survive deduplication because the ad changes city. They are 10% of the sample and three of the six that go above 80K, so the top end rests on fewer companies than it looks. Publishing a band correlates with being a foreign company bound by pay-transparency rules, or a product company comfortable with what it pays.
How we count the signals. It is keyword matching over the ad text. It measures what gets named, not what gets done. The international dataset does not work that way — it extracts skills with a model and groups synonyms — so the two counts are not interchangeable signal by signal. An ad saying “experience with RAG a plus” counts the same as one saying “you will build the retrieval pipeline”.
We count rag as a word, plus retrieval augmented. As a substring it would come out at 72.4% instead of 56.7%, and that excess is mostly leveraging.
The role cohort and the AI area are cut differently. The 476 come from the title; the 1,136 for “AI and ML” come from the taxonomy of our monthly market report, which classifies by priority and sends any title naming data to Data before checking whether it names AI. That is why adding the 672 for AI and ML without this title to the 476 for the role gives 1,148 and not 1,136: a dozen ads for the role live outside the area under that classification. Both figures are correct; they are not addable.
What we cannot claim. The absolute values are a floor. The application counter is LinkedIn’s public number at the moment of scraping, not the final total, and the table comparing roles leaves out the 31% that is unclassified. The ordering holds, the totals are partial. Work mode and company size come back empty in this cohort. The classification of employers into product company, traditional employer or IT services firm we did by hand over the 31 ads with a band.
Other people’s data. The international percentages are from Alexey Grigorev’s dataset: 6,964 descriptions from builtin.com across Los Angeles, New York, London, Amsterdam, Berlin and India, between February and August 2026 (our reading). We match neither in sample nor in method: theirs is builtin.com across six markets with skills extracted by a model that groups synonyms, ours is LinkedIn Spain counting words. Their evaluation and fine-tuning figures are over AI-First roles, not over the 6,964, so we have left them out of the chart. The author himself warns that his figures point in a direction, not to decimals.
This article gets updated. When there is new data it gets rewritten at this same URL, rather than published again.