LUMI AI Fac o y Se ice Cen e
Empowe ing Eu ope’s AI Ecosys em
D5.4 Connec o s o Common Eu opean Da a Spaces
2
D5.4
Connec o s o Common Eu opean Da a Spaces
D5.4 Connec o s o Common Eu opean Da a Spaces
3
P ojec Ti le
LUMI AI Fac o y Se ice Cen e
P ojec Ac onym
LUMI-AIF
P ojec Numbe
101234208
Type o Ac ion
HORIZON-JU-RIA
Topic
HORIZON-JU-EUROHPC-2025-AI-01-IBA-01
S a ing Da e o P ojec
01.03.2025
Ending Da e o P ojec
29.02.2028
Du a ion o he P ojec
36 mon hs
Websi e
lumi-ai- ac o y.eu
Wo k Package
WP5 – Da a access and in eg a ion
Task
T5.6: Connec ion o Common Eu opean Da a Spaces and o he
da a eposi o ies
Lead Au ho s
Heidi Laine (CSC)
Con ibu o s
Ja no Lai inen (CSC), Susanna Repo (CSC)
Pee Re iewe s
Ma in Ma hiesen (CSC), Tuuli Rand e (UT), Magnus Valg e
(UT)
Ve sion
1.0
Due Da e
31.10.2025
Submission Da e
31.10.2025
Dissemina ion le el
X
PU: Public
SEN: Sensi i e – limi ed unde he condi ions o he G an Ag eemen
EU-RES. Classi ied In o ma ion: RESTREINT UE (Commission Decision 2005/444/EC)
EU-CON. Classi ied In o ma ion: CONFIDENTIEL UE (Commission Decision 2005/444/EC)
EU-SEC. Classi ied In o ma ion: SECRET UE (Commission Decision 2005/444/EC)
D5.4 Connec o s o Common Eu opean Da a Spaces
4
Ve sion His o y
Re ision
Da e
Edi o s
Commen s
0.5
15.10.2025
Heidi Laine
Ready o in e nal e iew
0.2
30.10.2025
Heidi Laine
Re iewe s’ commen s inco po a ed
0.3
31.10.2025
Heidi Laine
Final e sion sen o he PMO o quali y
check.
1.0
31.10.2025
Anna Luoma
Final quali y check pe o med by he PMO,
sen o e iew
Decla a ion on he Use o AI Assis ance
This epo has been p epa ed wi h he suppo o GPT-5-enabled Mic oso Copilo , which was used o
assis in d a ing ex , checking language, ga he ing backg ound in o ma ion, and o c ea ing he
glossa y o e ms. All con en has been ho oughly e iewed, ac -checked, and edi ed by he au ho s
o ensu e accu acy and alignmen wi h he objec i es o he epo .
Glossa y o Te ms
Te m
De ini ion
AI-on-Demand
A Eu opean pla o m p o iding access o AI esou ces, ools, and se ices
o esea ch and indus y.
Audi abili y
The capabili y o ack and e i y da a ansac ions and policy compliance
h ough immu able logs and epo s.
Common
Eu opean Da a
Spaces
Fede a ed ecosys ems enabling us ed and in e ope able da a sha ing
ac oss sec o s in compliance wi h EU egula ions.
Connec o
A so wa e componen ha acili a es secu e, policy-complian da a
exchange be ween sys ems o da a spaces.
Da a
So e eign y
The p inciple ha da a owne s e ain con ol o e hei da a, including
how and by whom i is used.
Da aspace
P o ocol (DSP)
A s anda dized p o ocol o secu e da a exchange be ween connec o s,
sepa a ing con ol and da a planes.
DSSC (Da a
Spaces Suppo
Cen e)
An EU-backed o ganiza ion p o iding bluep in s, s anda ds, and guidance
o building and ope a ing da a spaces.
Eclipse Da a
Space
An open-sou ce amewo k implemen ing IDS and Gaia-X p inciples o
so e eign da a exchange.
D5.4 Connec o s o Common Eu opean Da a Spaces
5
Connec o
(EDC)
EOSC
(Eu opean Open
Science Cloud)
An EU ini ia i e p o iding a ede a ed en i onmen o sha ing esea ch
da a, ools, and se ices.
Eu opean
Heal h Da a
Space (EHDS)
An EU ini ia i e enabling secu e and complian use o heal h da a o
esea ch, inno a ion, and policymaking.
Eu opean
Language Da a
Space
A sec o -speci ic da a space ocused on mul ilingual language esou ces
o AI and language echnology de elopmen .
Fede a ed
Iden i y
A sys em ha allows use s o access mul iple se ices using a single se o
c eden ials ac oss us ed domains.
Gaia-X
A Eu opean ini ia i e p omo ing ede a ed cloud and da a in as uc u e
based on openness, anspa ency, and in e ope abili y.
IDS
(In e na ional
Da a Spaces)
A e e ence a chi ec u e and s anda d o secu e and so e eign da a
exchange de eloped by he In e na ional Da a Spaces Associa ion.
In e ope abili y
The abili y o di e en sys ems, o ganiza ions, o da a spaces o wo k
oge he seamlessly h ough s anda dized p o ocols.
LDS
The Eu opean Language Da a Space
LUMI AIF
LUMI AI Fac o y
Policy
En o cemen
Mechanisms embedded in connec o s o ensu e ha da a usage complies
wi h ag eed-upon e ms and egula ions.
REMS
(Resou ce
En i lemen
Managemen
Sys em)
A sys em o managing access o sensi i e da ase s h ough en i lemen
wo k lows and license en o cemen .
SPE (Secu e
P ocessing
En i onmen )
A con olled in as uc u e o p ocessing sensi i e da a, ensu ing
compliance wi h p i acy and secu i y egula ions.
Simpl
Middlewa e
Eu opean Commission’s open-sou ce middlewa e o in e ope abili y and
us ac oss Common Eu opean Da a Spaces.
D5.4 Connec o s o Common Eu opean Da a Spaces
6
Execu i e Summa y
This deli e able p esen s he ini ial s a egy and echnical ounda ion o in eg a ing he LUMI AI Fac o y
(LUMI AIF) wi h Common Eu opean Da a Spaces h ough s anda dized connec o componen s. These
connec o s a e essen ial o enabling secu e, policy-complian , and in e ope able da a exchange
be ween LUMI AIF and ex e nal da a ecosys ems, in alignmen wi h he Eu opean Da a S a egy.
The epo ou lines he a chi ec u al p inciples, componen selec ion c i e ia, and deploymen oadmap
o connec o s ha suppo da a so e eign y, us , and in e ope abili y. I emphasizes he ole o
connec o s as ope a ional enable s o go e nance logic, capable o en o cing usage policies, managing
iden i y, and acili a ing dynamic con ac nego ia ion ac oss he e ogeneous sys ems.
Th ee key connec o echnologies a e e alua ed:
• REMS, o en i lemen wo k lows and access go e nance in sensi i e domains like heal h and
genomics.
• Eclipse Da a Space Connec o (EDC), o s anda dized, so e eign da a exchange ac oss esea ch
and language da a spaces.
• Simpl middlewa e, o scalable in eg a ion wi h mul iple Eu opean da a spaces and suppo o
seman ic in e ope abili y and policy en o cemen .
The epo iden i ies high- alue da a spaces o ini ial in eg a ion, including he Eu opean Open Science
Cloud (EOSC), Eu opean Language Da a Space (LDS), and he Eu opean Heal h Da a Space (EHDS). I
also highligh s he s a egic ele ance o REMS in EHDS and i s ope a ional use in he Genome Da a
In as uc u e.
A phased deploymen plan is p oposed, s a ing wi h sandbox en i onmen s and MVP connec o s o
EOSC and LDS, ollowed by egula ed in eg a ions using Secu e P ocessing En i onmen s and REMS.
The plan includes echnical baselines, go e nance mechanisms, and ope a ional models o ensu e
compliance wi h e ol ing Eu opean s anda ds such as IDS-RAM, Gaia-X, DSSC bluep in s, and he Da a
Ac .
Looking o wa d, he epo ecommends expanding connec o co e age, enhancing compliance
ooling, and aligning wi h s a egic ini ia i es like Exa4Mind and he Eu oHPC Fede a ion Pla o m.
These e o s will posi ion LUMI AIF as a ede a ed, us ed node in he Eu opean AI and HPC ecosys em,
suppo ing scalable, esponsible, and inno a ion-d i en da a use.
D5.4 Connec o s o Common Eu opean Da a Spaces
7
Table o Con en s
1. In oduc ion ............................................................................................................... 9
1.1 Pu pose and scope o he epo 9
1.2 LUMI AIF objec i es o Eu opean da a spaces 9
2. Backg ound and Con ex ........................................................................................... 10
2.1 O e iew o he Common Eu opean Da a Spaces ini ia i e 10
2.2 Role o connec o s in da a sha ing and in e ope abili y 11
2.3 Rele an s anda ds and e e ence a chi ec u es 12
3. C i e ia o componen selec ion ............................................................................... 14
3.1 C i e ia o iden i ying high- alue da a spaces 14
3.2 Da a Spaces ele an o LUMI AIF 15
3.2.1 Eu opean Open Science Cloud 16
3.2.2 Eu opean Language Da a Space 17
3.2.3 Eu opean Heal h Da a Space 18
3.2.4 Da a spaces in he ields o manu ac u ing and communica ion echnologies 20
4. Connec o Componen s O e iew ............................................................................. 20
4.1 Gene al a chi ec u e o connec o s 20
4.2 A chi ec u al E alua ion: REMS, EDC, and Simpl 21
4.3 Technology s ack and dependencies 23
4.3.1 Co e Technology S ack Componen s 23
4.3.2 Dependencies and In eg a ion Poin s 24
5. Connec o deploymen plan...................................................................................... 24
5.1 Es ablishing P elimina y Ag eemen s o Da a Space Connec o s in LUMI AIF 24
5.1.1 Pu pose and Scope o P elimina y Ag eemen s 25
5.1.2 Requi emen s o Ag eemen Design 25
5.1.3 Implemen a ion S eps o LUMI AIF 25
5.1.4 Technical deploymen 26
D5.4 Connec o s o Common Eu opean Da a Spaces
8
6. Conclusions and Nex S eps ...................................................................................... 29
D5.4 Connec o s o Common Eu opean Da a Spaces
9
1. In oduc ion
1.1 Pu pose and scope o he epo
The pu pose o his deli e able epo is o p esen he ini ial i e a ion o da a space connec o solu ions
o LUMI AI Fac o y (LUMI AIF) designed o enable in e ope abili y and secu e da a exchange be ween
LUMI AIF and da a spaces. These connec o s o m he ounda ion o s anda dized, us ed, and e icien
da a sha ing ac oss domains and pla o ms in alignmen wi h he Eu opean Da a S a egy
1
. In he
s a egy, common and in e ope able da a spaces and he da a pools om key Eu opean sec o s ha hey
o m a e a way o implemen ing he single ma ke o da a in he EU.
In his epo we
• In oduce he i s LUMI AIF analysis on connec o componen s o high- alue Common
Eu opean Da a Spaces.
• Desc ibe he unde lying a chi ec u e, unc ionali ies, and in eg a ion app oach.
• Ou line he me hodology o selec ing, de eloping, and alida ing connec o s.
• De ine he s a egy o con inuous g ow h and e olu ion o he connec o ecosys em as p ojec
equi emen s and s akeholde needs ma u e.
• P esen how o ensu e alignmen wi h ele an s anda ds, secu i y, and da a p o ec ion
p inciples.
This deli e able ep esen s an ea ly s age o de elopmen . The componen s and p ocesses desc ibed
he e will be e ined, ex ended, and alida ed u he as he p ojec p og esses, inco po a ing eedback
om eal-wo ld use cases and e ol ing equi emen s.
1.2 LUMI AIF objec i es o Eu opean da a spaces
The o e a ching goal is o es ablish a obus and scalable mechanism o seamless da a exchange and
in eg a ion ac oss he e ogeneous sou ces, he eby enhancing he accessibili y and u ili y o AI esou ces
wi hin high pe o mance compu ing (HPC) wo k lows.
A he co e o his deli e able is he concep o he da a space connec o , a se ice componen ha
acili a es secu e, policy-complian da a sha ing be ween sys ems. In he con ex o Eu opean da a
spaces, connec o s a e essen ial o ope a ionalizing he p inciples o da a so e eign y, in e ope abili y,
and us . They allow he LUMI AIF use s and cus ome s o in e ac wi h ex e nal da a ecosys ems, and
ice e sa, wi hou comp omising con ol o e da a usage o iola ing egula o y cons ain s.
The deli e able aims o guide implemen a ion o connec o s ha a e compa ible wi h he a chi ec u al
and go e nance models de ined by Eu opean ini ia i es such as he Simpl, In e na ional Da a Spaces
Associa ion (IDSA), Gaia-X, and he Da a Spaces Suppo Cen e (DSSC). These connec o s mus suppo
1
h ps://commission.eu opa.eu/s a egy-and-policy/p io i ies-2019-2024/eu ope- i -digi al-age/eu opean-da a-
s a egy_en
D5.4 Connec o s o Common Eu opean Da a Spaces
16
Figu e 3, which is based on a IDSA igu e abou undamen al concep s o a da a space
9
, gi es a high-le el
desc ip ion o key oles and esponsibili ies o da a space connec o use cases in he LUMI AIF con ex .
LUMI AIF use s can use in eg a ed da a space connec o solu ions and amewo ks o ge ing da a om
da a space o he en i onmen . LUMI AIF will p o ide he necessa y echnical deploymen and
ag eemen p ocess pa hs wi h high- alue da a spaces, bu indi idual use s need o o m he binding
ag eemen s wi h he da a spaces / da a space pa icipan s (depending on he da a space go e nance in
place) based on he da a speci ic policies.
Figu e 3 Main LUMI AIF da a space connec o use case
In he nex sub-chap e s, we will examine some o he da a spaces ha ha e been ecognized as
especially ele an o LUMI AIF and ou use s and cus ome s. The lis is no de ini e no inal, as he
landscape o Eu opean da a spaces con inues o de elop dynamically.
3.2.1 Eu opean Open Science Cloud
The Eu opean Open Science Cloud (EOSC) is he Eu opean Union’s lagship ini ia i e o c ea e a
ede a ed, us ed, and mul idisciplina y en i onmen o sha ing esea ch da a, ools, and se ices.
Recognized as he common Eu opean da a space o science, esea ch, and inno a ion, EOSC is designed
o mobilize and align digi al esou ces ac oss Eu ope, enabling esea che s o publish, ind, and euse
da a in acco dance wi h he FAIR p inciples, Findabili y, Accessibili y, In e ope abili y, and Reusabili y.
EOSC is no a single pla o m bu a “sys em o sys ems,” ede a ing na ional and ins i u ional da a
eposi o ies, esea ch in as uc u es, and scien i ic se ice p o ide s in o a cohesi e ne wo k. The launch
o he EOSC EU Node in Oc obe 2024 ma ked a signi ican miles one, p o iding a e e ence
implemen a ion and a ga eway o esea che s o access in e ope able se ices ac oss Eu ope. EOSC’s
9
h ps://in e na ionalda aspaces.o g/wp-con en /uploads/dlm_uploads/IDSA-Da a-Space-Connec o -Repo -
1_Oc obe _2025-1.pd
D5.4 Connec o s o Common Eu opean Da a Spaces
17
a chi ec u e suppo s machine-ac ionable da a, c oss-disciplina y collabo a ion, and ep oducible
science, while i s ipa i e go e nance, comp ising he Eu opean Commission, Membe S a es, and he
EOSC Associa ion, ensu es s a egic coo dina ion and long- e m sus ainabili y.
Fo he LUMI AIF, EOSC ep esen s a high- alue da a space ha aligns wi h i s mission o in eg a e
wo ld-class compu ing, high-quali y da a, and op- ie AI expe ise. EOSC o e s access o a as a ay o
esea ch da ase s, including hose om li e sciences, en i onmen al sciences, social sciences, and
physics, domains whe e AI applica ions a e apidly e ol ing.
EOSC’s ede a ed a chi ec u e complemen s LUMI AIFs dis ibu ed compu ing model. The abili y o
access da a ac oss bo de s and disciplines wi hou cen alizing i aligns wi h LUMI’s emphasis on da a
so e eign y and compliance.
EOSC’s go e nance and policy amewo ks p o ide a s uc u ed en i onmen o esponsible da a use,
which is c i ical o LUMI AIF’s engagemen wi h sensi i e o egula ed da ase s.
Simpl
10
is he Eu opean Commission’s open-sou ce sma middlewa e designed o suppo
in e ope abili y and us ac oss Common Eu opean Da a Spaces, including EOSC. I p o ides modula
componen s o iden i y managemen , policy en o cemen , seman ic in e ope abili y, and cloud- o-
edge ede a ion. Simpl is being in eg a ed in o EOSC h ough easibili y s udies and pilo deploymen s,
wi h he goal o enabling seamless da a exchange be ween EOSC nodes and o he sec o al da a spaces.
Fo EOSC, Simpl o e s a echnical ounda ion o connec o de elopmen , se ice o ches a ion, and
c oss-space in e ope abili y. I suppo s he deploymen o EOSC se ices in he e ogeneous
en i onmen s, including HPC in as uc u es like LUMI. Simpl’s a chi ec u e allows EOSC o scale i s
ede a ion model, in eg a e wi h ex e nal da a spaces (e.g., heal h, language, public p ocu emen ), and
main ain compliance wi h Eu opean da a legisla ion.
In he con ex o LUMI AIF, Simpl enables he de elopmen o connec o componen s ha link LUMI o
EOSC in a secu e, policy-awa e, and scalable manne . This in eg a ion suppo s he b oade goal o
embedding LUMI in o he Eu opean esea ch da a ecosys em, acili a ing AI-d i en esea ch and
inno a ion ac oss disciplines.
3.2.2 Eu opean Language Da a Space
The Eu opean Language Da a Space (LDS)
11
is one o he sec o -speci ic ini ia i es unde he b oade
amewo k o Common Eu opean Da a Spaces, as ou lined in he Eu opean Da a S a egy. I s p ima y
objec i e is o acili a e he sha ing, euse, and alo isa ion o mul ilingual language esou ces ac oss
Eu ope, suppo ing bo h public and p i a e ac o s in de eloping language echnologies ha e lec he
linguis ic di e si y o he con inen . This da a space is pa icula ly ocused on enabling access o high-
quali y da ase s, models, and se ices ha suppo ansla ion, speech ecogni ion, na u al language
p ocessing (NLP), and o he language-based AI applica ions.
The ele ance o he LDS o he LUMI AI Fac o y is bo h s a egic and echnical. S a egically, i aligns
wi h he Eu opean Union’s commi men o digi al so e eign y and linguis ic inclusi i y. By pa icipa ing
10
h ps://digi al-s a egy.ec.eu opa.eu/en/policies/simpl
11
h ps://language-da a-space.ec.eu opa.eu/index_en
D5.4 Connec o s o Common Eu opean Da a Spaces
18
in his da a space, LUMI AIF can con ibu e o and bene i om a ede a ed in as uc u e ha suppo s
he de elopmen o AI models in all o icial EU languages, including less widely spoken ones such as
Finnish o Es onian. This is especially impo an o ensu ing ha AI sys ems deployed in Eu ope a e
cul u ally and linguis ically adap ed, a oiding biases ha a ise om aining on p edominan ly English-
language da ase s.
Technically, he Eu opean Language Da a Space p o ides a s uc u ed en i onmen o accessing
cu a ed language esou ces h ough s anda dized connec o s and go e nance amewo ks. Th ough
LDS, LUMI AIF can in eg a e wi h eposi o ies such as he Finnish Language Bank and o he na ional o
ins i u ional language a chi es.
LDS suppo s he de elopmen o open-sou ce ools and se ices ha can be deployed wi hin high-
pe o mance compu ing en i onmen s. This includes p e- ained models, anno a ion ools, and
seman ic esou ces ha a e essen ial o building obus NLP pipelines. By connec ing o his da a space,
LUMI AIF can accele a e he de elopmen o mul ilingual AI applica ions, suppo c oss-bo de esea ch
collabo a ions, and con ibu e o he Eu opean AI-on-Demand ecosys em.
The LDS Connec o
12
is he cen al echnical componen o his da a space. I is based on he Eclipse
Da aspace Connec o (EDC) and ex ended h ough communi y-d i en implemen a ions such as T ac us-
X and So i y. Key ea u es include:
• Pee - o-pee a chi ec u e: LDS connec o s ope a e in a decen alized manne , allowing
pa icipan s o publish, disco e , and exchange language asse s di ec ly.
• Me ada a and policy managemen : Each da a o e ing includes me ada a and usage policies,
which a e nego ia ed and en o ced h ough he connec o .
• Con ac nego ia ion and da a ans e : The connec o suppo s au oma ed con ac
conclusion and secu e da a exchange, including inancial ansac ions whe e applicable.
• Mul ilingual suppo : The LDS connec o includes speci ica ions o mul ilingual asse
desc ip ions, enabling seman ic in e ope abili y ac oss Eu opean languages.
3.2.3 Eu opean Heal h Da a Space
The Eu opean Heal h Da a Space (EHDS) is a lagship ini ia i e unde he Eu opean Da a S a egy,
designed o acili a e he secu e and e icien use o heal h da a ac oss he Eu opean Union. I aims o
empowe indi iduals wi h con ol o e hei pe sonal heal h da a while enabling esea che s,
policymake s, and inno a o s o access high-quali y da ase s o seconda y use in a legally complian and
echnically obus manne . EHDS is no a single pla o m, bu a ede a ed ecosys em composed o
na ional in as uc u es, common se ices, and ha monized go e nance amewo ks.
A he co e o EHDS is he concep o Secu e P ocessing En i onmen s (SPEs). These a e con olled
echnical in as uc u es whe e sensi i e heal h da a can be accessed and p ocessed wi hou
comp omising p i acy o iola ing legal cons ain s. SPEs mus mee s ingen equi emen s o da a
p o ec ion, iden i y managemen , audi abili y, and policy en o cemen . They a e designed o suppo
12
h ps://ec.eu opa.eu/news oom/lds/i ems/818143/en
D5.4 Connec o s o Common Eu opean Da a Spaces
19
pseudonymized o anonymized da a access, wi h mechanisms o dynamic consen , usage logging, and
secu e compu a ion.
Fo he LUMI AIF, EHDS ep esen s a s a egically impo an da a space o se e al easons. Fi s , heal h
da a is among he mos aluable and complex da ase s o AI de elopmen , pa icula ly in domains such
as medical imaging, genomics, epidemiology, and pe sonalized medicine. Access o EHDS da ase s
h ough SPEs would enable LUMI AIF o suppo high-impac esea ch and inno a ion in hese a eas,
le e aging i s high-pe o mance compu ing capabili ies o ain models on la ge-scale, he e ogeneous
heal h da a.
The EHDS is a sec o -speci ic da a space ocused on enabling bo h p ima y use (clinical ca e) and
seconda y use ( esea ch, inno a ion, policymaking) o elec onic heal h da a ac oss he EU. While he
EHDS egula ion de ines he legal and go e nance amewo k, i s echnical implemen a ion elies on
Secu e P ocessing En i onmen s (SPEs) and in e ope able connec o componen s.
Connec o solu ions o EHDS a e s ill eme ging, bu some a chi ec u al di ec ions can be assumed:
• SPE-in eg a ed connec o s: These connec o s a e embedded wi hin secu e en i onmen s ha
mee EHDS speci ica ions o p i acy, audi abili y, and policy en o cemen . They acili a e
con olled access o pseudonymized o anonymized heal h da a o seconda y use.
• Compliance wi h EHDS s anda ds: Connec o s mus suppo in e ope abili y wi h elec onic
heal h eco d (EHR) sys ems, en o ce usage condi ions, and in eg a e wi h he Heal hDa a@EU
in as uc u e. This includes suppo o iden i y ede a ion, consen managemen , and logging
mechanisms.
Whe he any o he exis ing a chi ec u es and solu ions can be adjus ed o EHDS is s ill di icul o know
o ce ain. Based on in o ma ion ecei ed h ough EHDS de elopmen p ojec s, such as TEHDAS2
13
, he
newly de eloped eDeli e y
14
will no be adequa e o EHDS, no he Simpl. The Commission is cu en ly
solici ing sugges ions om s akeholde s o p o ocols o use. X- oad
15
, de eloped by Es onia and Finland
has been sugges ed as one o he po en ial, cu en ly exis ing and unc ional op ions.
CSC’s REMS (Resea che ’s Access Managemen Sys em), while no a da a space connec o , plays a
s a egically impo an ole in he con ex o he Eu opean Heal h Da a Space (EHDS), p o iding a
ma u e and scalable solu ion o managing con olled access o sensi i e heal h da a, pa icula ly o
seconda y use in esea ch. REMS enables anspa en and audi able pe mi wo k lows, ensu ing ha
da a access complies wi h e hical, legal, and o ganiza ional equi emen s. This aligns closely wi h EHDS
objec i es o acili a e us wo hy da a sha ing ac oss bo de s and sec o s. REMS also suppo s
in e ope abili y ac oss echnical and go e nance laye s, making i a aluable e e ence model o EHDS
implemen a ion. I s ele ance is u he unde sco ed by i s use in he Genome Da a In as uc u e (GDI),
whe e REMS helps manage access o genomic da ase s ac oss Eu opean coun ies, demons a ing i s
capaci y o ope a e in complex, mul i-s akeholde en i onmen s wi h high da a p o ec ion s anda ds.
13
h ps:// ehdas.eu/
14
h ps://ec.eu opa.eu/digi al-building-blocks/si es/spaces/DIGITAL/pages/467110114/eDeli e TEHDAS
15
h ps://x- oad.global/
D5.4 Connec o s o Common Eu opean Da a Spaces
20
3.2.4 Da a spaces in he ields o manu ac u ing and communica ion
echnologies
As o la e 2025, manu ac u ing- ela ed da a spaces in Eu ope a e mo ing om planning o deploymen .
Suppo ed by he Digi al Eu ope P og amme and na ional ini ia i es, p ojec s like SM4RTENANCE and
UNDERPIN a e pilo ing eal-wo ld applica ions such as p edic i e main enance and dynamic asse
managemen .
16
These e o s a e guided by Gaia-X and In e na ional Da a Spaces (IDS), and a leas
SM4RTENANCE is elying on he Eclipse Da a Space Connec o (EDC) o da a in e ac ions. Key
challenges o manu ac u ing da a spaces include agmen ed da a landscapes, s anda d ha moniza ion,
and SME onboa ding.
Eu opean commission doesn’ iden i y communica ion echnologies as a speci ic da a space domain, no
does he IDSA Da a Spaces ada
17
so i ’s mo e di icul o ecognize he ele an and ma u e da a spaces
o ha ield. Howe e , i is likely ha hey oo will ely on some o he widely used s anda ds and
a chi ec u es, see chap e 2.3. I is also likely ha as he LUMI AIF indus y engagemen expands and
e ol es, mo e insigh in o he communica ion echnologies da a space landscape will be accumula ed,
and LUMI AIF da a space connec o plans and solu ions can be adjus ed acco dingly.
4. Connec o Componen s O e iew
4.1 Gene al a chi ec u e o connec o s
In gene al, he a chi ec u e o da a space connec o s e lec s he need o balance da a so e eign y,
in e ope abili y, and us in decen alized da a ecosys ems. These connec o s a e no monoli hic
sys ems, bu modula componen s designed o media e secu e and policy-complian da a exchange
be ween au onomous en i ies, whe he o ganiza ions, pla o ms, o se ices, wi hin and ac oss da a
spaces.
A he co e o mos da a space connec o a chi ec u es is a dual-plane design: he con ol plane and he
da a plane. The con ol plane handles me ada a exchange, iden i y e i ica ion, policy nego ia ion, and
con ac en o cemen . I ensu es ha da a sha ing ag eemen s a e espec ed and ha access is g an ed
only unde p ede ined condi ions. The da a plane, by con as , is esponsible o he ac ual ansmission
o da a, o en using secu e and audi able channels.
A chi ec u es such as he Eclipse Da a Space Connec o (EDC) and Simpl middlewa e ollow his
sepa a ion explici ly, enabling lexible in eg a ion wi h a ious iden i y p o ide s, policy engines, and
da a ca alogs. These connec o s o en suppo ede a ed disco e y mechanisms, allowing pa icipan s
o loca e and e alua e da a o e ings ac oss dis ibu ed en i onmen s. They also implemen usage
con ol amewo ks, which go beyond access con ol by en o cing how da a can be used a e i has been
sha ed.
A key a chi ec u al ea u e is ex ensibili y. Connec o s a e ypically buil o be modula , allowing
in eg a ion wi h di e en p o ocols (e.g. HTTP, MQTT, S3), s o age sys ems, and seman ic models. This
16
h ps://digi al-s a egy.ec.eu opa.eu/en/policies/da a-spaces
17
h ps://in e na ionalda aspaces.o g/wp-con en /uploads/dlm_uploads/The-Da a-Spaces-Rada -Ve sion-4.pd
D5.4 Connec o s o Common Eu opean Da a Spaces
21
modula i y is essen ial o adap ing o sec o -speci ic equi emen s and o e ol ing alongside eme ging
s anda ds such as he Da aspace P o ocol (DSP) and Gaia-X compliance amewo ks.
Ano he a chi ec u al conside a ion is compliance and audi abili y. Connec o s mus suppo logging,
moni o ing, and policy en o cemen mechanisms ha align wi h Eu opean egula ions such as he
GDPR, Da a Go e nance Ac , and Da a Ac . This o en in ol es embedding us se ices, such as
ce i ica ion au ho i ies and egis ies, in o he connec o a chi ec u e.
4.2 A chi ec u al E alua ion: REMS, EDC, and Simpl
The ele ance o REMS, Eclipse Da a Space Connec o (EDC), and Simpl middlewa e o he LUMI AIF
s ems om hei complemen a y oles in enabling secu e, in e ope able, and policy-complian access o
da a—each add essing di e en laye s o he da a space ecosys em and aligning wi h bo h echnical and
s a egic goals o LUMI AIF and he Eu opean Da a S a egy.
REMS is pa icula ly aluable o LUMI AIF in con ex s whe e access o sensi i e o es ic ed da ase s
mus be go e ned h ough o mal wo k lows. I s a chi ec u e suppo s en i lemen managemen ,
ede a ed iden i y, and license en o cemen , which a e essen ial o esea ch in as uc u es dealing wi h
con olled da a. Fo LUMI AIF, which may need o in eg a e da ase s om esea ch domains such as
genomics, social sciences, o heal h, REMS p o ides a ma u e and audi able mechanism o managing
who can access wha da a, unde which condi ions. While REMS is no a da a space connec o , i
complemen s connec o -based a chi ec u es by handling he go e nance and au ho iza ion laye ha
connec o s mus espec .
Eclipse Da a Space Connec o is di ec ly aligned wi h he Eu opean da a space a chi ec u e and is being
es ed in he Eu opean Language Da a Space and he Finnish Language Bank. This makes i highly
ele an o LUMI AIF, especially in mul ilingual AI model de elopmen and na u al language p ocessing
(NLP) asks. EDC suppo s ede a ed da a exchange, policy en o cemen , and iden i y managemen , all
o which a e c i ical o in eg a ing LUMI in o b oade Eu opean da a ecosys ems. I s modula
a chi ec u e allows LUMI o connec wi h o he da a spaces using s anda dized p o ocols, acili a ing
access o language esou ces, anno a ed co po a, and p e- ained models while ensu ing compliance
wi h usage policies.
Simpl middlewa e, de eloped unde he Da a Spaces Suppo Cen e (DSSC), p o ides he ounda ional
in as uc u e o connec ing o Common Eu opean Da a Spaces. I is designed o suppo
in e ope abili y, us , and alue c ea ion ac oss sec o s. Fo LUMI AIF, Simpl o e s a scalable and
s anda ds-complian way o in eg a e wi h mul iple da a spaces, including hose in domains such as
mobili y, ene gy, and public adminis a ion. I s a chi ec u e suppo s seman ic in e ope abili y, iden i y
ede a ion, and policy nego ia ion, making i sui able o high-pe o mance AI wo k lows ha equi e
access o di e se and dis ibu ed da a sou ces.
Toge he , hese h ee solu ions o m a laye ed and complemen a y s ack o LUMI AIF:
• REMS go e ns access o sensi i e da ase s.
• EDC enables s anda dized, so e eign da a exchange ac oss language and esea ch domains.
• Simpl p o ides he middlewa e backbone o connec ing o mul iple Eu opean da a spaces in a
scalable and policy-awa e manne .
D5.4 Connec o s o Common Eu opean Da a Spaces
22
Thei combined use suppo s LUMI AIF’s mission o become a ede a ed, in e ope able, and legally
complian AI in as uc u e embedded wi hin he Eu opean da a space ecosys em.
REMS (Resou ce En i lemen Managemen Sys em)
REMS is a domain-speci ic access managemen sys em de eloped by CSC, p ima ily used o
managing access o esea ch da ase s. I s a chi ec u e is laye ed and modula , consis ing o h ee main
ie s:
• API Laye : Handles HTTP eques s, pe o ms coa se-g ained access con ol, and ans o ms
eques s in o se ice laye calls. I includes Swagge -based documen a ion and li es in he
ems.api.* namespaces.
• Se ice Laye : Encapsula es business logic and o ches a es ope a ions ac oss mul iple
da abase namespaces. I manages wo k lows, licensing, use se ings, and asynch onous
command p ocessing. This laye a oids ci cula dependencies and is designed o modula
expansion.
• DB/Ex e nal Laye : Manages da a pe sis ence and ex e nal in eg a ions. I includes domain-
speci ic logic o se ializa ion, schema coe cion, and da abase que ies. Ex e nal se ices (e.g.,
EGA) a e accessed ia dedica ed namespaces ( ems.ex .*). [gi hub.com]
REMS is igh ly coupled o i s in e nal da a model and access wo k lows. While i suppo s ede a ed
login and API-based in eg a ion, i is no designed as a gene al-pu pose da a space connec o . I s
a chi ec u e p io i izes consis ency, audi abili y, and ine-g ained en i lemen managemen o e
c oss-domain in e ope abili y.
Eclipse Da a Space Connec o (EDC)
EDC is a gene al-pu pose, open-sou ce amewo k o so e eign, in e -o ganiza ional da a exchange.
I is designed o implemen he In e na ional Da a Spaces (IDS) and Gaia-X e e ence a chi ec u es.
I s a chi ec u e includes:
• Connec o Co e: Ac s as he endpoin o da a exchange, en o cing usage policies and
managing da a lows.
• Fede a ed Ca alog: Enables disco e y o da a o e ings ac oss o ganiza ions.
• Iden i y Hub: Manages au hen ica ion and au ho iza ion using s anda ds like OAu h2 and
decen alized iden i y sys ems.
• Policy En o cemen and Moni o ing: Ensu es compliance wi h da a usage ag eemen s and
p o ides audi capabili ies.
• Ex ensibili y Laye : Suppo s in eg a ion wi h a ious da a ans e p o ocols, s o age
sys ems, and cloud en i onmen s. [p ojec s.eclipse.o g], [news oom.eclipse.o g]
EDC is modula and designed o euse ac oss sec o s. I suppo s bo h con ol and da a planes,
enabling dynamic nego ia ion o da a con ac s and secu e da a ansmission. I s a chi ec u e is
aligned wi h Eu opean da a space p inciples, emphasizing in e ope abili y, da a so e eign y, and
us .
Simpl Middlewa e
Simpl is a Eu opean Commission-backed middlewa e solu ion de eloped unde he Da a Spaces
Suppo Cen e (DSSC). I p o ides ounda ional componen s o building and connec ing da a spaces.
I s a chi ec u e is s uc u ed a ound h ee pilla s:
D5.4 Connec o s o Common Eu opean Da a Spaces
23
• Da a In e ope abili y: Includes seman ic models, da a o ma s, and APIs o enabling c oss-
domain da a exchange.
• Da a So e eign y and T us : P o ides iden i y managemen , policy en o cemen , and
aceabili y mechanisms.
• Da a Value C ea ion: Suppo s da a disco e y, ma ke place unc ionali y, and mone iza ion
ea u es. [dssc.eu]
Simpl connec o s se e as endpoin s o da a space pa icipa ion, in eg a ing wi h sha ed egis ies
and se ices such as us egis ies, da a ca alogs, and obse abili y ools. The a chi ec u e
dis inguishes be ween con ol and da a planes, and is designed o be ex ensible and complian wi h
e ol ing Eu opean s anda ds (e.g., Da a Ac , Gaia-X, IDS).
4.3 Technology s ack and dependencies
The echnology s ack and dependencies o da a space connec o s in he Eu opean con ex a e shaped by
a con e gence o open s anda ds, modula so wa e componen s, and go e nance amewo ks designed
o ensu e in e ope abili y, da a so e eign y, and us ac oss ede a ed ecosys ems. These connec o s
a e no s andalone applica ions, bu in eg a ed sui es o se ices deployed wi hin con olled
en i onmen s, such as cloud-na i e in as uc u es o on-p emises sys ems.
4.3.1 Co e Technology S ack Componen s
Connec o F amewo ks
The mos p ominen implemen a ions include he Eclipse Da aspace Connec o (EDC) and he FIWARE
Da a Space Connec o , bo h o which ollow he IDS Re e ence A chi ec u e Model (IDS-RAM). These
amewo ks a e modula and ex ensible, allowing o ganiza ions o in eg a e hem in o exis ing IT
landscapes. They suppo he Da aspace P o ocol (DSP), which de ines schemas and communica ion
p o ocols o publishing da a, nego ia ing usage ag eemen s, and accessing da a wi hin ede a ed
sys ems.
Iden i y and T us Se ices
Connec o s will on iden i y e i ica ion mechanisms such as OAu h2, OpenID Connec , and
decen alized iden i y amewo ks. Many o hese amewo ks ha e no ye been olled ou , bu a e
being de eloped. T us se ices include emo e a es a ion, legal iden i y e i ica ion, and compliance
checks. These a e essen ial o en o cing pa icipa ion ules and ensu ing ha da a sha ing occu s wi hin
a us ed en i onmen .
Policy En o cemen and Con ac Managemen
Usage con ol is implemen ed h ough policy engines ha in e p e and en o ce da a usage ag eemen s.
These may be based on s anda ds like XACML o cus om ule-based sys ems. Con ac nego ia ion and
en o cemen a e co e o he con ol plane o he connec o a chi ec u e.
Me ada a and Seman ic In e ope abili y
D5.4 Connec o s o Common Eu opean Da a Spaces
24
Me ada a managemen is suppo ed h ough APIs and ocabula ies ha enable seman ic alignmen .
Ini ia i es like Sma Da a Models, NGSI-LD, and SAREF p o ide domain-speci ic on ologies ha help
connec o s in e p e and ans o m da a consis en ly ac oss sec o s.
Da a Exchange APIs and P o ocols
Connec o s expose endpoin s o da a access, o en using REST ul APIs, G aphQL, o MQTT o
s eaming da a. The anspo laye mus suppo secu e communica ion (e.g., TLS) and may include
adap e s o cloud s o age (e.g., S3, Azu e Blob) o edge de ices.
Deploymen and Run ime En i onmen s
Mos connec o s a e designed o un in con aine ized en i onmen s (e.g., Docke , Kube ne es) o
suppo scalabili y and o ches a ion. This allows o ganiza ions o deploy connec o s in hyb id cloud
se ups o edge compu ing scena ios.
Logging and Obse abili y
Connec o s include logging mechanisms o audi abili y and moni o ing. These a e essen ial o
compliance wi h Eu opean egula ions such as he GDPR and he Da a Ac , and o ensu ing
anspa ency in da a ansac ions.
4.3.2 Dependencies and In eg a ion Poin s
Ex e nal Regis ies and Ca alogs: Connec o s o en in eg a e wi h ede a ed da a ca alogs and us
egis ies o disco e da a o e ings and e i y pa icipan c eden ials.
Seman ic Mapping Tools: Dependencies include ools o mapping be ween di e en da a models and
se ializa ion o ma s (e.g., JSON-LD, RDF).
Go e nance F amewo ks: Connec o s mus align wi h go e nance models de ined by ini ia i es like
IDSA, Gaia-X, and DSSC, which speci y pa icipa ion ules, ce i ica ion c i e ia, and compliance
mechanisms.
5. Connec o deploymen plan
The goal is o deploy and ope a e a se o s anda ds-aligned da a space connec o s ha le LUMI AIF
use s and cus ome s disco e , nego ia e, and access ex e nal da a and AI asse s in a policy-complian
way, wi h clea go e nance, obse abili y, and li ecycle managemen . The plan builds on D5.4’s in en o
align as much as possible wi h ele an e e ence a chi ec u es and s anda ds such as IDS/GaiaX/DSSC,
o p io i ize high- alue Eu opean da a spaces, and o main ain connec o s as i s -class componen s o
he LUMI AIF pla o m.
5.1 Es ablishing P elimina y Ag eemen s o Da a Space
Connec o s in LUMI AIF
A ounda ional equi emen o deploying da a space connec o s wi hin he LUMI AI Fac o y is he
es ablishmen o p elimina y ag eemen s ha de ine he p inciples o access and use. These ag eemen s
se e as he legal, e hical, and echnical sca olding upon which us ed da a exchange can occu .
D5.4 Connec o s o Common Eu opean Da a Spaces
25
Wi hou hem, connec o s isk becoming me e echnical condui s, lacking he go e nance mechanisms
necessa y o ensu e esponsible and complian da a and model usage.
5.1.1 Pu pose and Scope o P elimina y Ag eemen s
P elimina y ag eemen s a e no me ely o mali ies; hey a e ope a ional ins umen s ha :
• De ine licensing e ms o da ase s and models, including pe missible use cases, edis ibu ion
igh s, and a ibu ion equi emen s.
• Embed e hical guidelines, such as es ic ions on sensi i e da a p ocessing, bias mi iga ion
obliga ions, and anspa ency expec a ions.
• Speci y echnical cons ain s, including da a o ma s anda ds, access p o ocols, and esou ce
usage limi s.
These ag eemen s mus be adap able o he e ol ing na u e o AI de elopmen and da a go e nance,
pa icula ly in a ede a ed and mul i-s akeholde en i onmen like LUMI AIF.
5.1.2 Requi emen s o Ag eemen Design
To be e ec i e, p elimina y ag eemen s should mee he ollowing c i e ia:
Requi emen
Desc ip ion
Modula i y
Ag eemen s should be composed o eusable clauses ha can be ailo ed o
speci ic connec o s o use cases.
Machine-
eadabili y
Te ms mus be encoded in a o ma ha connec o s can in e p e and en o ce
au oma ically.
Policy e sioning
Ag eemen s should suppo e sion con ol o ack changes and ensu e
backwa d compa ibili y.
In e ope abili y
Ag eemen s mus align wi h Eu opean and in e na ional s anda ds (e.g., Gaia-X,
EOSC, IDSA).
Audi abili y
Usage condi ions and en o cemen ac ions should be logged o compliance
e i ica ion.
5.1.3 Implemen a ion S eps o LUMI AIF
To es ablish hese ag eemen s wi hin LUMI AI Fac o y, he ollowing p ocess is ecommended:
1. S akeholde Mapping
Iden i y da a p o ide s, model de elope s, se ice ope a o s, and end-use s. Cla i y hei oles
and esponsibili ies.
2. Templa e De elopmen
C ea e ag eemen empla es based on exis ing amewo ks (e.g., IDSA Usage Policies, EOSC
Rules o Pa icipa ion).
3. Legal and E hical Re iew
Engage legal and e hics expe s o alida e he empla es agains applicable egula ions (e.g.,
GDPR, AI Ac ).