- AMD a d Ce eb as a e collabo ati g to adva ce a wo kload-optimized app oach to ult a-low-late cy AI i fe e ce i f ast uctu e.
- AMD Helios™ a d the Ce eb as Wafe -Scale E gi e will ope ate as a si gle disagg egated i fe e ce wo kflow, combi i g ult a-high-th oughput f om AMD I sti ct™ GPUs, with ult a-fast toke ge e atio of Ce eb as Wafe -Scale E gi e.
- Ce eb as pla s to deploy AMD Helios i its data ce te s, with the joi t solutio expected to be available fi st th ough Ce eb as Cloud i the seco d half of 2026.
SAN FRANCISCO a d SUNNYVALE, Calif., July 23, 2026 (GLOBE NEWSWIRE) — AMD (NASDAQ: AMD) a d Ce eb as Systems (NASDAQ: CBRS) a ou ced a tech ical pa t e ship to delive a ew disagg egated AI i fe e ce solutio that combi es AMD Helios™ ackscale solutio s with the Ce eb as Wafe -Scale E gi e. U veiled at Adva ci g AI 2026, the solutio is desig ed to delive the ult a-low late cy equi ed fo the most adva ced AI applicatio s while d amatically i c easi g the th oughput a d efficie cy.
The joi t AMD a d Ce eb as solutio will deploy AMD Helios alo gside Ce eb as Wafe -Scale E gi e tech ology i teg ated i a si gle i fe e ce wo kflow fo maximum pe fo ma ce a d efficie cy. AMD Helios will p ovide a high-pe fo ma ce, scalable th oughput e gi e. Ce eb as Wafe -Scale E gi e tech ology will p ovide ult a-fast, ult a-low late cy decode a d toke ge e atio . Togethe , the two compute e gi es a e expected to delive up to 5x highe toke s pe seco d pe watt (T/s/W) i.
AI i fe e ce wo kloads i c easi gly have diffe e t equi eme ts ac oss late cy, th oughput, toke capacity, cost a d scale. High-volume wo kloads p io itize maximizi g toke ge e atio , while codi g, eal-time copilots, live age ts a d age tic wo kflows dema d faste espo se times. These diffe e ces a e d ivi g dema d fo hete oge eous i f ast uctu e that matches compute tech ologies to specific wo kload equi eme ts.
The AMD a d Ce eb as solutio add esses this challe ge th ough disagg egated i fe e ce, optimizi g the two p ima y stages of the wo kflow i depe de tly. AMD Helios p ovides ult a-high th oughput, p ocessi g p ompts a d la ge co text wi dows. The Ce eb as Wafe -Scale E gi e accele ates the memo y-ba dwidth-i te sive toke ge e atio , with ult a-low late cy. By co ecti g these best-i -class e gi es th ough o e i teg ated wo kflow, the compa ies a e c eati g a diffe e tiated platfo m fo ult a-low-late cy i fe e ce without sac ifici g th oughput o scale.
“AI i fe e ce is becomi g o e of the la gest i f ast uctu e oppo tu ities i AI, a d its g owi g dive sity equi es a mo e flexible app oach,” said D . Lisa Su, chai a d CEO, AMD. “AMD Helios delive s leade ship pe fo ma ce a d scale fo the b oadest a ge of i fe e ce wo kloads. Togethe with Ce eb as, we a e exte di g that leade ship i to the most late cy-se sitive applicatio s a d c eati g a powe ful ew platfo m fo eal-time age tic AI.”
“The dema d fo ult a-fast i fe e ce is g owi g at a u p ecede ted pace. Ce eb as delive s the wo ld’s fastest, ult a-low-late cy i fe e ce,” said A d ew Feldma , CEO a d co-fou de , Ce eb as. “Pa t e i g with AMD gives us a i c edible oppo tu ity to b i g that pe fo ma ce to eve mo e custome s.”
Fast toke ge e atio is becomi g i c easi gly impo ta t as AI moves i to softwa e developme t, auto omous age ts, obotics, scie tific discove y a d othe applicatio s whe e espo se time di ectly shapes the use expe ie ce a d the useful ess of the system. The joi t solutio b i gs togethe compleme ta y a chitectu es pu pose-built fo these dema ds.
AMD Helios p ovides the high-th oughput p ompt e gi e, ack-scale efficie cy a d deployme t scale equi ed to p ocess la ge umbe s of complex equests. Ce eb as Wafe -Scale E gi e tech ology p ovides the ult a-low-late cy a d decode pe fo ma ce eeded to etu toke s i eal time. The esult is a solutio desig ed specifically fo the ult a-low-late cy segme t of the i fe e ce ma ket, with AMD Helios as the fou datio fo high-th oughput a d bala ced i fe e ce wo kloads ac oss the data ce te .
Ce eb as pla s to deploy AMD Helios systems i its data ce te s, with the joi t solutio expected to become available i itially th ough Ce eb as Cloud i the seco d half of 2026.
Ce eb as Systems (NASDAQ: CBRS) builds the wo ld’s fastest AI i f ast uctu e. The Ce eb as team of pio ee i g compute a chitects, compute scie tists, AI esea che s, a d e gi ee s of all types came togethe to make AI bliste i gly fast th ough i ovatio a d i ve tio . We believe that whe AI is fast, it will cha ge the wo ld. Leadi g global co po atio s, esea ch i stitutes, a d gove me ts choose Ce eb as to u thei AI wo kloads. Ce eb as solutio s a e available o p emises a d i the cloud. Visit ce eb as.ai fo mo e.
This p ess elease co tai s fo wa d-looki g stateme ts co ce i g Adva ced Mic o Devices, I c. (AMD) such as the featu es, fu ctio ality, pe fo ma ce, availability, scalability, deployme t, timi g a d expected be efits of AMD’s collabo atio a d joi t solutio with Ce eb as, which a e made pu sua t to the Safe Ha bo p ovisio s of the P ivate Secu ities Litigatio Refo m Act of 1995. Fo wa d-looki g stateme ts a e commo ly ide tified by wo ds such as “would,” “may,” “expects,” “believes,” “pla s,” “i te ds,” “p ojects” a d othe te ms with simila mea i g. I vesto s a e cautio ed that the fo wa d-looki g stateme ts i this p ess elease a e based o cu e t beliefs, assumptio s a d expectatio s, speak o ly as of the date of this p ess elease a d i volve isks a d u ce tai ties that could cause actual esults to diffe mate ially f om cu e t expectatio s. Such stateme ts a e subject to ce tai k ow a d u k ow isks a d u ce tai ties, ma y of which a e difficult to p edict a d a e ge e ally beyo d AMD’s co t ol, that could cause actual esults a d othe futu e eve ts to diffe mate ially f om those exp essed i , o implied o p ojected by, the fo wa d-looki g i fo matio a d stateme ts. Mate ial facto s that could cause actual esults to diffe mate ially f om cu e t expectatio s i clude, without limitatio , the followi g: impact of gove me t actio s a d egulatio s such as expo t egulatio s, impo t ta iffs, t ade p otectio measu es, a d lice si g equi eme ts; competitive ma kets i which AMD’s p oducts a e sold; the cyclical atu e of the semico ducto i dust y; ma ket co ditio s of the i dust ies i which AMD p oducts a e sold; AMD’s ability to i t oduce p oducts o a timely basis with expected featu es a d pe fo ma ce levels; loss of a sig ifica t custome ; eco omic a d ma ket u ce tai ty; qua te ly a d seaso al sales patte s; AMD’s ability to adequately p otect its tech ology o othe i tellectual p ope ty; u favo able cu e cy excha ge ate fluctuatio s; ability of thi d pa ty ma ufactu e s to ma ufactu e AMD’s p oducts o a timely basis i sufficie t qua tities a d usi g competitive tech ologies; availability of esse tial equipme t, mate ials, compo e ts (such as memo y supply), subst ates o ma ufactu i g p ocesses; ability to achieve expected ma ufactu i g yields fo AMD’s p oducts; AMD’s ability to ge e ate eve ue f om its semi-custom SoC p oducts; pote tial secu ity vul e abilities; pote tial secu ity i cide ts i cludi g IT outages, data loss, data b eaches a d cybe attacks; u ce tai ties i volvi g the o de i g a d shipme t of AMD’s p oducts; AMD’s elia ce o thi d-pa ty i tellectual p ope ty to desig a d i t oduce ew p oducts; AMD’s elia ce o thi d-pa ty compa ies fo desig , ma ufactu e a d supply of mothe boa ds, softwa e, memo y a d othe compute platfo m compo e ts; AMD’s elia ce o Mic osoft a d othe softwa e ve do s’ suppo t to desig a d develop softwa e to u o AMD’s p oducts; AMD’s elia ce o thi d-pa ty dist ibuto s a d add-i -boa d pa t e s; impact of modificatio o i te uptio of AMD’s i te al busi ess p ocesses a d i fo matio systems; compatibility of AMD’s p oducts with some o all i dust y-sta da d softwa e a d ha dwa e; costs elated to defective p oducts; failu e to mai tai a efficie t supply chai as custome dema d cha ges; AMD’s ability to ely o thi d pa ty supply-chai logistics fu ctio s; AMD’s ability to effectively co t ol sales of its p oducts o the g ay ma ket; impact of climate cha ge o AMD’s busi ess; AMD’s ability to ealize its defe ed tax assets; pote tial tax liabilities; cu e t a d futu e claims a d litigatio ; impact of e vi o me tal laws, co flict mi e als elated p ovisio s a d othe laws o egulatio s; evolvi g expectatio s f om gove me ts, i vesto s, custome s a d othe stakeholde s ega di g co po ate espo sibility matte s; issues elated to the espo sible use of AI; est ictio s imposed by ag eeme ts gove i g AMD’s otes, the gua a tees of Xili x’s otes a d the evolvi g c edit ag eeme t; AMD’s ability to satisfy fi a cial obligatio s u de gua a tees, leases a d othe comme cial commitme ts; impact of acquisitio s, joi t ve tu es a d/o i vestme ts o AMD’s busi ess a d AMD’s ability to i teg ate acqui ed busi esses; impact of a y impai me t of the combi ed compa y’s assets; political, legal a d eco omic isks a d atu al disaste s; futu e impai me ts of tech ology lice se pu chases; AMD’s ability to att act a d etai key employees; a d AMD’s stock p ice volatility. I vesto s a e u ged to eview i detail the isks a d u ce tai ties i AMD’s Secu ities a d Excha ge Commissio fili gs, i cludi g but ot limited to AMD’s most ece t epo ts o Fo ms 10-K a d 10-Q.
Ce eb as uses its i vesto elatio s page (i vesto s.ce eb as.ai), its X accou t (@ce eb as), a d its Li kedI page (li kedi .com/compa y/ce eb as-systems/) to disclose mate ial o -public i fo matio a d fo complyi g with its disclosu e obligatio s u de Regulatio FD. Acco di gly, i vesto s should mo ito these cha els, i additio to followi g Ce eb as’ p ess eleases, Secu ities a d Excha ge Commissio (SEC) fili gs, public co fe e ce calls a d public webcasts.
This p ess elease co tai s “fo wa d-looki g stateme ts” withi the mea i g of applicable secu ities laws. All stateme ts othe tha stateme ts of histo ical fact could be deemed to be fo wa d-looki g, i cludi g, but ot limited to, stateme ts ega di g the featu es, capacity, scalability, pe fo ma ce, timi g, data ce te deployme t a d impleme tatio , costs a d expected be efits a d oppo tu ities associated with Ce eb as’ collabo atio a d joi t solutio with AMD, a d a y assumptio s elati g to the fo egoi g. The wo ds “may,” “will,” “shall,” “should,” “expects,” “pla s,” “a ticipates,” “could,” “i te ds,” “ta get,” “p ojects,” “co templates,” “believes,” “estimates,” “p edicts,” “pote tial,” “objective,” o “co ti ue,” o the egative of these wo ds o othe simila te ms o exp essio s that co ce ou expectatio s, st ategy, pla s, o i te tio s a e i te ded to ide tify fo wa d-looki g stateme ts, although ot all fo wa d-looki g stateme ts co tai these ide tifyi g wo ds. These fo wa d-looki g stateme ts a e subject to a umbe of isks a d u ce tai ties, ma y of which i volve facto s o ci cumsta ces that a e beyo d Ce eb as’ co t ol. These isks a d u ce tai ties i clude, but a e ot limited to: Ce eb as’ ability to sustai a d ma age its g owth, access bo owi gs a d othe sou ces of capital o acceptable te ms, a d deploy available capital to suppo t g owth; its histo y of et losses a d ability to achieve a d mai tai p ofitability; its limited ope ati g histo y at its cu e t scale a d ability to accu ately fo ecast eve ue a d app op iately budget a d ma age expe ses; its depe de ce o a limited umbe of sig ifica t custome s, i cludi g Ope AI, G oup 42 Holdi g Ltd, Mohamed bi Zayed U ive sity of A tificial I tellige ce, a d AWS, a d the pote tial impact of a y eductio i dema d f om, mate ial adve se developme t i its elatio ships with, o failu e to meet its obligatio s to, such custome s, i cludi g u de its Maste Relatio ship Ag eeme t with Ope AI; the timi g, executio a d expected be efits of its st ategic custome , pa t e a d fi a ci g a a geme ts; its histo ical elia ce o sales of ha dwa e systems a d the ea ly-stage, apidly evolvi g ma ket fo its cloud-based offe i gs a d AI i f ast uctu e; its ability to secu e sufficie t data ce te capacity a d capital to suppo t its cloud-based offe i gs; its ability to lau ch ew offe i gs a d add ew p oduct capabilities; a d its ability to compete effectively i the apidly evolvi g a d competitive ma ket fo AI computi g solutio s.
Ce eb as’ actual esults could diffe mate ially f om those stated o implied i fo wa d-looki g stateme ts due to a umbe of facto s. Acco di gly, u due elia ce should ot be placed o such stateme ts. These fo wa d-looki g stateme ts a e made as of the date they we e fi st issued a d a e based o i fo matio available to Ce eb as togethe with Ce eb as’ expectatio s, estimates, fo ecasts, p ojectio s, beliefs, a d assumptio s as of such date. These fo wa d-looki g stateme ts should ot be elied upo as ep ese ti g Ce eb as’ views as of a y date subseque t to the date of this p ess elease. Past pe fo ma ce is ot ecessa ily i dicative of futu e esults. Ce eb as u de takes o i te tio o obligatio to update o evise a y fo wa d-looki g stateme ts, whethe as a esult of ew i fo matio , futu e eve ts, o othe wise, except as equi ed by law.
Fu the i fo matio o pote tial isks that could affect actual esults is i cluded i Ce eb as’ most ece t fili gs with the Secu ities a d Excha ge Commissio (the “SEC”), i cludi g i Ce eb as’ most ece t Qua te ly Repo t o Fo m 10-Q, copies of which may be obtai ed by visiti g Ce eb as’ I vesto Relatio s website at i vesto s.ce eb as.ai o the SEC’s website at www.sec.gov.
Liz Sti eAMD I vesto Relatio s720-652-3965liz.sti e@amd.com
K iselle La a Ce eb asp @ce eb as.ai
Sea Do seyCe eb as I vesto Relatio si vesto s@ce eb as.ai
_______________i Based o modelli g by AMD Pe fo ma ce Labs a d Ce eb as i July 2026 to dete mi e toke s pe seco d pe kilowatt (TPS/kW) at a compa able i te activity poi t with Kimi 2.6 1T Model compa i g a AMD Helios ackscale solutio with Ce eb as WSE to a Ce eb as WSE-o ly co figu atio . System ma ufactu e s may va y co figu atio s, yieldi g diffe e t esults. MI400-021
