That makes Te sto e t’s Netwo ked AI a chitectu e a diffe e t ki d of solutio – ope , ge e al-pu pose, flexible fo hete oge eous o sta d alo e deployme ts, a d backed by AI expe ts – that ca withsta d the co sta t cha ge i the AI i dust y.
Co ti ui g to build o p evious pe fo ma ce, Te sto e t sha ed ew LLM a d video with lip-sy c a d audio be chma ks. O the latest models e te p ises a e deployi g ight ow, Te sto e t Galaxy Blackhole supe cluste s post:
Kimi K2.6: 900 toke s/seco d/use , 3x faste tha GPUsDeepSeek-R1-0528 671B: 400+ toke s/seco d/use , up f om 350+ at TT-Deploy SFLTX 2.3 Fast: oughly 6-seco d video ge e atio at 144 f ames, 1080p, with audio a d lip-sy c, 4x faste tha GPUs
Diffe e t model families, o e a chitectu e, with capacity that g ows ea -li ea ly as Galaxies a e added. Te sto e t’s pe fo ma ce e ables e te p ises to scale p emium i fe e ce wo kloads efficie tly.
Lau chi g today at TT-Deploy JP, Te sto e t a ou ced TT-Ascalo S, a compute-de se RISC-V CPU fo age tic AI. Age tic AI lea s o the CPU i a ew way, gated less by aw compute tha by o chest atio , I/O, a d late cy, a d TT-Ascalo S is built fo it:
De sity: Built o the fou datio of TT-Ascalo X, TT-Ascalo S is pu pose built fo a footp i t ~50% the size, delive i g ~140% pe fo ma ce pe mm².Efficie cy: A compact, powe -efficie t desig fo high-th oughput executio laye s.Late cy: Ascalo S is tu ed fo the mixed, b a ch-heavy, tool-co ected executio patte s typical of age t u times, helpi g e able mo e p edictable executio .
I additio to age tic AI, TT-Ascalo S is applicable to high-efficie cy se ve s, etwo ki g a d sto age SoCs, a d data-ce te edge deployme ts. As lice sable RISC-V IP, it e ables custome s to exte d the same a chitectu al fou datio i to custom silico desig s.
Co ecti g the accele ato s a d the CPU alike is Te sto e t’s Netwo ked AI. The a chitectu e u ifies compute, memo y, a d etwo ki g ove sta da d Ethe et, with a ope -sou ce softwa e stack. Te sto e t Galaxies a d supe cluste s sta d alo e o d op i to a existi g GPU fleet without eplaci g existi g i f ast uctu e. Custome s ca add capacity without betti g o a si gle model, wo kload, o ve do ; the systems keep pe fo mi g as models cha ge; a d the i f ast uctu e stays i the custome ‘s ha ds. Fo the e te p ises a d atio s buildi g ext-ge p ivate AI i f ast uctu e, that i depe de ce a d flexibility is key.
“We built o e a chitectu e that u s eve ythi g, d ops i ext to whateve you al eady ow , a d scales f om a lice sable co e to a supe cluste . That’s what lets a compa y, o a cou t y, ow its AI a d adapt to the o ly co sta t i AI – cha ge. We’ e excited to pa t e so widely i Japa with key pa t e s like ai&, Rapidus, P efe ed Netwo ks, Socio ext a d Tu i g.” said Jim Kelle , CEO of Te sto e t.
Ac oss Japa , Te sto e t e ables cloud se vice p ovide s, data ce te s, a d p ivate i f ast uctu e with eve ythi g f om lice sable IP to la ge-scale supe cluste deployme ts.
Its la gest deployme t to date u s with ai&, Japa ‘s ve tically i teg ated f o tie AI platfo m. ai& lau ched its hete oge eous i fe e ce platfo m to expa d sove eig AI i Japa . Te sto e t Galaxies a d supe cluste s suppo t key i fe e ce a d age tic wo kloads, a d e ables custome use cases such as: chat, RAG, visio , a d post t ai i g, se ved e ti ely f om withi the cou t y. They’ve deployed mo e tha 120 Te sto e t Galaxy systems ep ese ti g the la gest scale of sove eig AI compute i the egio .
“ai& is built o the co victio that the best i fe e ce i f ast uctu e outes eve y wo kload to the silico best suited to it. Te sto e t’s Galaxy supe cluste s have p ove themselves o exactly the wo kloads ou e te p ise custome s ca e about, a d this deployme t ep ese ts the la gest sove eig AI compute footp i t i Japa . We’ e excited fo what the pa t e ship u locks,” said David Be ett, CEO & co-fou de at ai&
That each ests o a deep base i the cou t y. Tu i g, the Tokyo-based auto omous-d ivi g compa y, has demo st ated Te sto e t Blackhole u i g i side a auto omous vehicle as a p oof-of-co cept. A d th ough the atio al 2 m p og am with Rapidus, adopted by NEDO a d led by LSTC, Te sto e t co t ibutes the RISC-V CPU chiplet. Te sto e t Japa has ope ated i Tokyo si ce 2023, u s a AI data ce te i Osaka today, a d is b i gi g up to 200 Japa ese silico e gi ee s i to its desig teams.
TT-Deploy JP b i gs the ecosystem o stage, with pa t e s i cludi g ai&, Rapidus, P efe ed Netwo ks, Socio ext a d Tu i g, plus live demos f om ai&, BOS a d Tu i g.
Te sto e t is a AI compute compa y led by CEO Jim Kelle – a chitect of Apple A4/A5, AMD Ze , a d Tesla’s Full Self-D ivi g chip. The compa y builds RISC-V-based AI p ocesso s a d systems fo develope s, e te p ises, a d sove eig i f ast uctu e wo ldwide. I additio to se ve s a d wo kstatio s, Te sto e t lice ses its TT-Ascalo RISC-V CPU a d Te six AI co es to chip desig e s i cludi g Samsu g a d LG. Backed by Bezos Expeditio s, Samsu g, LG Elect o ics, Hyu dai Moto G oup, Fidelity, a d othe s, Te sto e t has aised ove $1B+ a d ope ates f om Sa ta Cla a, Austi , To o to, Belg ade, Tokyo, a d Ba galo e. Lea mo e at te sto e t.com.
Justi Mauldi Salie t PR
