Copy igh © 2025 1
Eu opean Con e ence on EDGE AI
Technologies and Applica ions - EEAI
20-22 Oc obe 2025, Naples, I aly
The in e sec ion o imagina ion and
execu ion, whe e edge AI lea ns o
c ea e, eason, and ac .
Copy igh © 2025
Eu opean Con e ence on EDGE AI
Technologies and Applica ions - EEAI
Milan, I a20-22 Oc obe 2025 Naples, I aly
2
Copy igh © 2025
O idiu Ve mesan, SINTEF AS, No way
F om Mic o o Me a Edge
A chi ec ing Gene a i e and Agen ic Edge AI Technologies and Au onomous Sys ems
Copy igh © 2025
P esen a ion Ou line
•Edge AI G anula i y
•E olu ion o AI Models
•Edge AI and Au omo i e Indus y
•Edge AI Dynamics
•Edge AI T ans o ma ion
•GenAI Enginee ing Tools
•Edge AI E olu ion
3
Copy igh © 2025
Edge AI G anula i y
4
Applica ions
DSPs, FGPAs, CPUs, GPUs, ASICs Ne wo k
P ocessing Uni (NPU), In elligen P ocessing
Uni (IPU). Tenso P ocessing Uni (TPU),
Reduced Ins . Se Compu e RISC-V,
Neu omo phic.
Compu ing uni s (indus ial p ocessing, panel
uni s, e c.) , ne wo k compu ing uni s
(in elligen ou e s, swi ches, ga eways and
o he communica ions ha dwa e), in elligen
con olle s (PLCs, RTUs, DCS).
Mic o and clus e ed se e s o handle
compu e in ensi e asks / wo kloads
(e.g., high-end CPUs, GPUs, FGPAs,
e c.), on p emises edge compu ing,
local edge.
Cloud In as uc u e. Local,
egional and na ional da a
cen es. Fede a ion o clouds
and da a cen es.
Mul i-access edge
compu ing (MEC)
in as uc u e. Fog
p ocessing and
pla o ms
Deep
Edge Me a
Edge
Da a
Cen e s
Mic o
Edge
Mic o Deep Me a MEC Cloud
Fusion
Fusion
Con inuum-X Con inuum-X Con inuum-X Con inuum-X
HW/SW/AI
Da a
HW/SW/AI
Da a
HW/SW/AI
Da a
HW/SW/AI
Da a
Senso s
Ac ua o s
GPU
ASIC
TPU
IPU
Real- ime
P ocesses
Fog
MEC
Compu ing Con inuum
Copy igh © 2025
E olu ion o AI Models – F om Gian s o E iciency
•2025 con inues he end owa ds model minia u isa ion.
•P ocesso s o AI - Focus on aining
5
Nex -Gen AI needs mo e chips, as e , and sma e sys ems
eleased: F om gene a i e AI o agen ic AI and physical AI.
Sou ce: YOLE G oup 2025
Adap ing la ge models o speci ic uses accele a es demand o
lexible, gene al-pu pose GPUs and AI accele a o s ha can
handle smalle -scale aining and une close o he edge,
expanding semiconduc o demand beyond hype scale s.
Copy igh © 2025
NVIDIA – F om Gaming o Playing he Game
•NVIDIA is no longe jus a p ocesso designe bu now p o ides a comple e da a cen e
solu ion wi h i s DGX Pod / Supe Pod pla o m. In el and AMD y o ollow his end.
•AI ASIC designe s a e seeking o weaken he AI monopoly o NVIDIA and AMD wi h
highly ene gy-e icien p ocesso s. La ge companies o whom p ocesso de elopmen
is no hei co e business ely on ex e nal i ms o co-de elop hei AI ASICs.
• S a ups and es ablished companies will pa ly ace di ec compe i ion in he ma ke .
•AI hype scale s a e looking o con ol a la ge pa o he supply chain, s a ing wi h AI
accele a o s, o massi ely educe hei CAPEX.
•Hype scale expansion is also aking place in ARM-based CPUs, which a e mo e ene gy-
e icien han x86.
•The slowdown o In el and he mo e powe -hung y x86 p ocesso s a e making way o
mo e ene gy-e icien ARM p ocesso s.
6
NVIDIA a he cen e o chess game - Da a cen e alue chain
F om cloud o edge, he pa h o wa d demands a
uni ied e olu ion o he compu e s ack o ma ch
AI’s con inuous mo emen , balancing h oughpu ,
la ency, ene gy e iciency, and scalabili y. The
u u e o inno a ion depends on his.
Sou ce: YOLE G oup 2025
Nex -Gen sys ems demand a mo e om HW
han ea lie models. Agen ic AI, which in ol es
au onomous decision-making, equi es con inual
in e ence and ad anced memo y a chi ec u es o
execu e asks. Physical AI, embedded in obo ic
sys ems, adds eal- ime senso y in eg a ion,
mobili y and p ocessing, u he inc easing
compu e equi emen s.
Copy igh © 2025
Edge AI and Au omo i e Indus y
•The e olu ion o semiconduc o p ocess nodes om 180
nanome e (nm) o below 5nm ma ks a d ama ic leap in
pe o mance, powe e iciency and in eg a ion densi y.
•Leading semiconduc o companies a e aking dis inc
s a egic app oaches o cap u e alue in he au omo i e
ma ke .
•Newe en an s and compu e ocused i ms like Qualcomm
and N idia a e posi ioning hemsel es as end- o-end
pla o m p o ide s, so wa e-de ined capabili ies o e ing
cen alised compu e, AI accele a ion, and scalable
a chi ec u es.
•E o s a e unde way wi hin g oups like he Au omo i e Edge
Compu ing Conso ium and he b oade RISC-V ecosys em
o de ine chiple -based e e ence a chi ec u es o ehicle
applica ions.
7
Sou ce: YOLE G oup 2025
Copy igh © 2025
Edge AI Dynamics
•Edge AI e ol es wi h mul imodal models, agen ic AI and
indus ial-g ade deploymen s. Edge AI is en e ing he
nex phase, mo ing beyond TinyML models o unning
mul imodal la ge language models (LLMs), small
language models (SLMs) and ision-language models
(VLMs) di ec ly on indus ial-g ade edge de ices.
•Reasoning models o e a signi ican ad ancemen due o hei
cha ac e is ics o sol e complex p oblems wi h high accu acy.
•Agen ic AI c ea e new oppo uni ies o imp o e p oduc i i y,
and decision making, and equi e us and p o en applica ions.
•Mul imodal GenAI is ans o ming applica ions by enabling he
addi ion o new in e aces, ea u es and unc ionali ies.
•Syn he ic da a help mul iple indus ies and use cases as i lowe
he ba ie s o eliance on eal-wo ld da a and enables
oppo uni ies o simula ions and da a-gene a ion echniques.
8
05
04
03
02
01
Imp o ed
C ea i i y
E hical
Conside a ions
Real-Time
Applica ions
Mul i-
Modal
Gene a i e
AI
Hyb id
Models
GENERATIVE
AI
Copy igh © 2025
Edge AI T ans o ma ion
•The ans o ma ion is accele a ed by:
9
•AI accele a o s e iciency and pe o mance ha suppo eal- ime
in e ence o models wi h billions o pa ame e s mee ing indus ial la ency
and he mal cons ain s.
•Edge AI solu ions a e powe ed o eal- ime, local p ocessing. Real-wo ld
edge deploymen s ocus on compac 3B–8B models o op imise la ency,
cos , and ene gy e iciency.
•LLMs, SLMs and VLMs a e becoming mo e edge-op imised. A chi ec u al
inno a ions and ligh weigh mul imodal models a e enabling in e ence a
lowe compu a ional equi emen s. Ze o-sho VLMs (abili y o a model o
unde s and and pe o m asks wi hou ha ing been speci ically ained on
hose asks) a e changing edge applica ions.
•The edge AI ecosys em is ma u ing on he HW, SW and AI side wi h new
amewo ks o e ings ha manage he ull li ecycle o edge AI, om
model aining and op imisa ion o deploymen , o e - he-ai (OTA)
upda es, and un ime managemen .
Copy igh © 2025
E en O ganise s
16
The objec i es o LoLiPoP IoT (Long Li e Powe Pla o ms o In e ne o Things) a e o
de elop ene gy ha es ing-based inno a i e Long Li e Powe Pla o ms ha enable
e o i o wi eless senso ne wo k edge de ices o asse acking, condi ion and
pe o mance moni o ing. www.lolipop-io .eu
LoLiPoP IoT
EdgeAI-T us
AIMS5.0
SC4EU
SC4EU is a unique Chips JU "Inno a ion Ac ion" p ojec o ake he supply chain
managemen o semiconduc o p oduc ion in Eu ope o a new le el. A ue demand
pla o m, along wi h i s on ology as a o mal desc ip ion o all in o ma ion wi hin he
chain, acili a es close in e ac ion and smoo h, anspa en collabo a ion, making e en
highly complex supply chains esilien , lexible, and agile. h ps://sc4.eu/
EdgeAI-T us add esses an ad anced, us wo hy edge AI ecosys em h ough cu ing-
edge ha dwa e, so wa e, and ools. The p ojec aims o enhance decen alized EdgeAI
ope a ions ha a e secu e, eliable, and sus ainable. By in eg a ing AI-based
algo i hms, de ices, and APIs, EdgeAI-T us os e s in e ope abili y and secu e da a
exchange ac oss di e se pla o ms. om senso -ac ua ed de ices o cloud sys ems, all
wi hin a dynamic ze o us en i onmen . h ps://www.edgeai- us .eu/
AIMS5.0 aims o boos he economy by adop ing, ex ending, and implemen ing AI-
enabled HW and SW componen s and sys ems ac oss he en i e indus ial alue chain.
New echnologies om IoT and based on Seman ic Web on ologies, ML and AI help
Eu opean manu ac u e s o shi om Indus y 4.0 o Indus y 5.0, c ea ing human-
cen ic wo kplace condi ions and a clima e- iendly p oduc ion. h ps://aims50.eu/
Copy igh © 2025
Suppo ing O ganiza ions
17
The Eu opean Technology Pla o m on Sma Sys ems In eg a ion is an indus y-
d i en policy ini ia i e, de ining esea ch, de elopmen and inno a ion needs as
well as policy equi emen s ela ed o Sma Sys ems In eg a ion and in eg a ed
Mic o- and Nanosys ems. The main objec i e is o de elop a ision and o se up a
S a egic Resea ch Agenda. www.sma -sys ems-in eg a ion.o g
Inside Indus y Associa ion is he Eu opean Technology Pla o m o esea ch,
design and inno a ion on In elligen Digi al Sys ems and hei applica ions. The
Associa ion is a membe ship o ganisa ion o he Eu opean esea ch and
inno a ion ac o s wi h mo e han 200 membe s and associa es om all o e
Eu ope. www.inside-associa ion.eu
Chips Join Unde aking suppo s esea ch, de elopmen , inno a ion, and u u e
manu ac u ing capaci ies in he Eu opean semiconduc o ecosys em. Launched as
pa o he Chips o Eu ope Ini ia i e, i con on s semiconduc o sho ages and
s eng hens Eu ope's digi al au onomy, engaging a signi ican EU,
na ional/ egional and p i a e indus y unding o nea ly €11 billion.
h ps://po al.chips-ju.eu opa.eu/
EU
AENEAS
EPoSS
INSIDE
Chips JU
AENEAS s anding o Associa ion o Eu opean NanoElec onics Ac i i ieS, is an
indus ial Associa ion, es ablished in 2006, p o iding unpa alleled ne wo king
oppo uni ies, policy in luence & suppo ed access o unding o all ypes RD&I
pa icipan s in he ield o mic o and nanoelec onics enabled componen s and
sys ems. h ps://aeneas-o ice.o g/
