Modelpla e gives platfo m a d AI teams o e co t ol pla e fo ma agi g models ac oss cha gi g GPU i f ast uctu e, se vi g tech ologies a d p ovide s without tyi g applicatio s to the i f ast uctu e u de eath them
SALT LAKE CITY, Nov. 10, 2026 (GLOBE NEWSWIRE) — Upbou d, the compa y behi d C osspla e, today submitted Modelpla e, its ope sou ce co t ol pla e fo AI i fe e ce, to the Cloud Native Computi g Fou datio (CNCF) Sa dbox. If accepted, Modelpla e will b i g a fleet-wide co t ol pla e fo ma agi g models a d GPU i f ast uctu e i to a eut al commu ity whe e p ovide s, mai tai e s a d e te p ise use s ca help shape its developme t.
Teams a e i c easi gly u i g a d expe ime ti g with diffe e t models, se vi g tech ologies a d GPU i f ast uctu e ac oss multiple cluste s a d p ovide s. Platfo m e gi ee s have to make those pieces wo k togethe , fi d e ough GPU capacity a d keep applicatio s u i g whe a model, se vi g e gi e o p ovide cha ges.
Modelpla e applies the ope , decla ative co t ol pla e app oach Upbou d b ought to cloud i f ast uctu e with C osspla e. Platfo m a d AI teams specify which models to u a d what i f ast uctu e those models eed. Modelpla e coo di ates deployme ts, se vi g e gi es a d GPU capacity ac oss a fleet of Kube etes cluste s a d cloud p ovide s. Develope s get a co siste t way to access i fe e ce without havi g to t ack whe e a model u s o how it’s se ved.
“Getti g a model u i g is o ly the begi i g. Keepi g it eliable as wo kloads, models a d GPU i f ast uctu e cha ge is whe e platfo m teams a e b eaki g ew g ou d,” said
AI i fe e ce is evolvi g i to a ho izo tal ecosystem, with i ovatio happe i g i depe de tly ac oss models, se vi g e gi es, accele ato s a d i f ast uctu e p ovide s. No si gle ve do will ow eve y laye , a d teams i c easi gly eed to b i g these tech ologies togethe athe tha build a ou d a si gle stack. Modelpla e p ovides a ope co t ol pla e ac oss that ecosystem, allowi g o ga izatio s to take adva tage of adva ces at each laye while mai tai i g a co siste t way to deploy a d ope ate i fe e ce ac oss thei i f ast uctu e.
At KubeCo + CloudNativeCo No th Ame ica, Upbou d fou de a d CEO Bassam Tabba a demo st ated that app oach usi g Modelpla e ac oss i f ast uctu e f om AWS a d Vult . The demo st atio showed how a applicatio ca mai tai the same i fe e ce e dpoi t while the models, se vi g tech ologies a d i f ast uctu e u de eath it cha ge. It’s a p actical example of why po tability matte s as o ga izatio s ma age i c easi gly dive se GPU e vi o me ts.
“Modelpla e makes u i g i fe e ce o Vult a co figu atio choice, ot a sepa ate deployme t model. Teams ca use the same model ma ifests ac oss p ovide s a d choose GPU i f ast uctu e based o what each wo kload eeds,” said
That i te ope ability exte ds beyo d i f ast uctu e p ovide s to the tech ologies that make i fe e ce possible. Modelpla e is desig ed to wo k alo gside the specialized se vi g, scheduli g a d accele atio p ojects adva ci g i depe de tly ac oss the ecosystem. NVIDIA Dy amo is o e example of how those capabilities ca be i teg ated i to a commo fleet wide ope ati g model.
Modelpla e i teg ates with NVIDIA Dy amo to b i g dist ibuted i fe e ce capabilities to fleets of GPU cluste s. The i teg atio uses G ove a d the KAI Schedule to coo di ate multi- ode wo kloads, a d ModelExp ess to t a sfe model weights di ectly betwee GPUs, educi g epeated eads f om sto age as eplicas scale.
“Scali g AI i fe e ce equi es coo di atio ac oss se vi g e gi es, GPU esou ces a d the softwa e co ecti g them. NVIDIA Dy amo p ovides a ope , modula fou datio fo dist ibuted i fe e ce. Modelpla e builds o that ecosystem at the fleet level, helpi g platfo m teams deploy models ac oss e vi o me ts while co ti ui g to be efit f om adva ces i accele ated computi g,” said
These i teg atio s demo st ate Modelpla e’s app oach to b i gi g dive se models, se vi g tech ologies a d GPU i f ast uctu e togethe as o e hete oge eous i fe e ce fleet. By submitti g Modelpla e to CNCF, Upbou d is i viti g the b oade cloud ative a d AI i f ast uctu e commu ities to help shape this ope , fleet-wide co t ol pla e.
E te p ise i f ast uctu e teams a e al eady movi g towa d a u ified, API-fi st, cloud- ative co t ol pla e model: o e platfo m to gove i f ast uctu e ac oss e vi o me ts, with co siste t i te faces fo e gi ee s, softwa e a d i c easi gly AI age ts. I fe e ce is a atu al exte sio of that model. Modelpla e b i gs the same decla ative co t ol pla e patte to models a d GPU i f ast uctu e, while its ope fou datio gives the ecosystem a eut al place to build that ext laye togethe .
“We’ e seei g la ge global e te p ises i cludi g some of the wo ld’s la gest fi a cial i stitutio s futu e-p oof thei i f ast uctu e fo a AI-d ive ope ati g model,” said
Upbou d uses Modelpla e to u its ow chat, codi g assista t a d secu ity age t wo kloads. That wo k has su faced p oblems that a e ’t appa e t whe a model fi st sta ts: GPU quota does ’t gua a tee available capacity, a d a appa e tly successful equest ca still etu a i complete espo se. Upbou d is sha i g those lesso s a d usi g them to guide imp oveme ts to the ope sou ce p oject.
The six-pa t D i ki g Ou Ow Champag e se ies docume ts how Upbou d evaluated models fo a build secu ity age t, diag osed failu es i the se vi g stack a d tested what it would take to expa d a self-hosted codi g assista t beyo d a pilot.
- D i ki g Ou Ow Champag e: What It Takes to Ow You I tellige ce
- D i ki g Ou Ow Champag e, Pa t 2: Fi di g the Co ks
- D i ki g Ou Ow Champag e, Pa t 3: Use Case #1, the Build Secu ity Age t
- D i ki g Ou Ow Champag e, Pa t 4: Ca dli g the Bottles
- D i ki g Ou Ow Champag e, Pa t 5: What Does It Take to G ow a Self-Hosted Codi g Model f om Pilot to Team?
- D i ki g Ou Ow Champag e, Pa t 6: The Model Was Ra ely the P oblem
Modelpla e is available u de the Apache 2.0 lice se at https://github.com/modelpla eai/modelpla e, with docume tatio at https://docs.modelpla e.ai.
Develope s a d platfo m teams ca explo e the p oject, sha e ope ati g expe ie ce a d co t ibute imp oveme ts.
Upbou d is pio ee i g i f ast uctu e platfo ms fo the age tic AI e a, se vi g Fo tu e 500 compa ies a d platfo m e gi ee s ac oss mo e tha 100 cou t ies. The compa y empowe s i f ast uctu e a d platfo m teams with i tellige t co t ol pla es, based o Kube etes a d C osspla e, that p ovisio , ope ate, a d adapt so platfo ms a e eady fo both huma s a d AI age ts. Upbou d is the c eato a d p ima y mai tai e of C osspla e, the popula ope -sou ce f amewo k fo buildi g cloud- ative co t ol pla es, with ove 100 millio dow loads a d adoptio by mo e tha 1,000 o ga izatio s i p oductio wo ldwide. Upbou d is also the c eato of Modelpla e, the ope -sou ce co t ol pla e fo AI i fe e ce built o C osspla e. Fou ded i 2017 a d headqua te ed i Sa Mateo, CA, a Se ies B sta tup backed by GV (fo me ly Google Ve tu es), Altimete Capital, a d I tel Capital, Upbou d has aised $69M to date. Fo mo e i fo matio , visit www.upbou d.io.
CONTACT: Media Co tact
upbou d@o og oup.com
