NVIDIA and Cisco Introduce New Pluggable Optics Form Factor to Scale AI Factories

New T a sceive s E able AI Builde s to Scale Compute Withi the Same Netwo k Footp i t While Reduci g Netwo ki g’s Sha e of the Powe Budget

News Summa y:

• Ult a-De se Small Fo m Facto Pluggable (USFP) is desig ed to be the fi st module to suppo t 400Gb/s-pe -la e elect ical sig als, e abli g capacity up to 12.8Tb/s so AI facto y builde s ca scale Ethe et AI fab ics, while exte di g a p ove deployme t model ac oss multiple ge e atio s of data ce te etwo ks.
• USFP delive s 4x the ba dwidth de sity a d 1.3x the powe efficie cy of 200Gb/s-pe -la e optics so AI facto ies ca co ect mo e computi g systems without addi g etwo k tie s, while usi g less powe to move each bit of data.
• NVIDIA a d Cisco a e wo ki g with TE Co ectivity a d Amphe ol to build a multi-sou ce ha dwa e ecosystem a ou d USFP.

SAN FRANCISCO, Oct. 12, 2026 (GLOBE NEWSWIRE) — OCP — NVIDIA a d Cisco today a ou ced pla s fo Ult a-De se Small Fo m Facto Pluggable (USFP), a ew high-de sity 400Gb/s-pe -la e optical module desig ed to equip the ha dwa e ecosystem fo the ext ge e atio of AI facto ies usi g pluggable t a sceive s.

As AI models g ow a d i fe e ce dema d expa ds, AI facto ies must co ect mo e computi g systems while ma agi g etwo k complexity, powe a d costs. Movi g mo e data pe watt a d pe switch helps ope ato s make bette use of thei i f ast uctu e i vestme ts.

Scali g AI facto ies equi es mo e ba dwidth pe co ectio a d mo e co ectio s pe switch. Whe a switch eaches its capacity, fu the etwo k expa sio ca equi e additio al switch laye s a d optical li ks, i c easi g late cy, powe co sumptio a d costs.

USFP is desig ed to e able AI facto y builde s to scale f om 200 Gb/s pe la e to 400Gb/s-pe -la e Ethe et AI fab ics, while exte di g a deployme t model p ove at scale ac oss multiple ge e atio s of data ce te etwo ks. With USFP, custome s ca emb ace ext-ge e atio optical tech ology a d meet the pe fo ma ce dema d of AI wo kloads without ove hauli g thei data ce te i f ast uctu e.

The fi st pla ed USFP-based t a sceive s will pack 16 full-duplex la es i a octal small fo m facto (OSFP)-sized footp i t suppo ti g both switches a d Supe NICs, with the ability to expa d to 32 a d mo e la es i the futu e. By doubli g both the la e cou t a d ba dwidth pe la e, USFP e ables AI fab ics to co ect mo e 1.6Tb/s e dpoi ts without addi g etwo k tie s. Custome s will be able to achieve 12.8Tb/s of ove all capacity with USFP.

Compa able with a OSFP-sized module, it delive s 4x the ba dwidth a d f o t-pa el ba dwidth de sity, with app oximately 1.3x the powe efficie cy, ve sus 200Gb/s-pe -la e pluggable modules. This mea s AI builde s ca i c ease thei compute scale withi the same etwo k footp i t, while dec easi g etwo ki g’s sha e of the ove all AI facto y powe budget.

“Buildi g the wo ld’s la gest AI facto ies sta ts with ethi ki g the smallest of co ectio s,” said Gilad Shai e , se io vice p eside t NVIDIA. “With USFP, NVIDIA a d Cisco a e buildi g a ecosystem to ei ve t pluggable optics — co ecti g mo e computi g with g eate efficie cy to powe the ext ge e atio of AI facto ies.”

“AI builde s eed a i te co ect a chitectu e that ca scale pe fo ma ce without locki g them i to a mo e complex ope ati g model,” said Bill Ga t e , se io vice p eside t a d ge e al ma age of optical systems a d optics at Cisco. “Cisco a d NVIDIA ecog ized that scali g pluggable optics to 400Gb/s pe la e would equi e mo e tha a othe i c eme tal exte sio of today’s module desig s. Th ough ou collabo atio , USFP is desig ed to delive a ew level of ba dwidth de sity while p ese vi g the se viceability, deployme t flexibility a d ope atio al simplicity of pluggable optics. By taki g a full-stack app oach — f om silico a d systems to optics a d the etwo k ope ati g system — we ca optimize eve y laye to delive the pe fo ma ce a d scale the ext ge e atio of AI etwo ks will equi e.”

Mo e AI I f ast uctu e F om Eve y SwitchDesig ed fo elect ical sig als of 400Gb/s pe la e, USFP uses a ew two-dime sio al a ay co ecto with a two-step i se tio desig that e gages high-speed co tacts afte the module is i se ted, educi g co tact wea , imp ovi g sig al i teg ity a d maki g i se tio fo ce mo e p edictable. The co ecto tech ology allows fo fu the scali g of la e cou t to 32 full duplex la es a d mo e i the futu e. Because the co ecto is desig ed fo 400Gb/s pe fo ma ce, it also offe s imp oved pe fo ma ce at 200Gb/s pe la e.

USFP suppo ts powe dissipatio of up to 100 watts without eedi g a sepa ate, dedicated embedded cooli g ci cuit withi the t a sceive , educi g switch desig complexity a d imp ovi g eliability. Togethe , these capabilities e able highe -ba dwidth fab ics while p ese vi g the matu e deployme t, se viceability a d ma ufactu ability of existi g pluggable t a sceive s.

All USFP co figu atio s suppo t optical desig s that educe late cy a d powe co sumptio by simplifyi g sig al p ocessi g. Suppo t fo a b oad a ge of optical tech ologies a d co ectio dista ces gives system desig e s the flexibility to meet u ique AI facto y deployme t eeds.

A Ha dwa e Ecosystem fo Next-Ge e atio AI Facto iesNVIDIA a d Cisco a e collabo ati g with TE Co ectivity a d Amphe ol to build USFP cages, co ecto s a d modules, with additio al pa t e s expected to joi . The sha ed fo m facto gives supplie s a commo fou datio fo developi g high-de sity pluggable i f ast uctu e fo 400Gb/s-pe -la e fab ics to build a obust ecosystem fo AI facto y builde s.

The collabo atio builds o NVIDIA a d Cisco’s existi g wo k to e su e NVIDIA Spect um-X™ Ethe et etwo ki g a d Cisco Silico O e-based Ethe et etwo ks wo k togethe seamlessly, simplifyi g AI i f ast uctu e deployme t fo e te p ises, eoclouds a d sove eig clouds.

NVIDIA a d Cisco pla to c eate a ope i dust y specificatio fo USFP th ough a upcomi g commu ity-led multi-sou ce p oject.

About Cisco

Cisco (NASDAQ: CSCO) is the wo ldwide tech ology leade that is evolutio izi g the way o ga izatio s co ect a d p otect i the AI e a. Fo mo e tha 40 yea s, Cisco has secu ely co ected the wo ld. With its i dust y leadi g AI-powe ed solutio s a d se vices, Cisco e ables its custome s, pa t e s a d commu ities to u lock i ovatio , e ha ce p oductivity a d st e gthe digital esilie ce. With pu pose at its co e, Cisco emai s committed to c eati g a mo e co ected a d i clusive futu e fo all. Discove mo e o The News oom a d follow us o X at @Cisco.

Cisco a d the Cisco logo a e t adema ks o egiste ed t adema ks of Cisco a d/o its affiliates i the U.S. a d othe cou t ies. A listi g of Cisco’s t adema ks ca be fou d at www.cisco.com/go/t adema ks. Thi d-pa ty t adema ks me tio ed a e the p ope ty of thei espective ow e s. The use of the wo d pa t e does ot imply a pa t e ship elatio ship betwee Cisco a d a y othe compa y.

About NVIDIANVIDIA (NASDAQ: NVDA) is the wo ld leade i AI a d accele ated computi g.

Fo fu the i fo matio , co tact:Alex Shapi oCo po ate Commu icatio sNVIDIA Co po atio p ess@ vidia.com 

Taylo Hassma Cisco Thassma @cisco.com

Cisco Fo wa d-Looki g Stateme tsThis p ess elease may be deemed to co tai fo wa d-looki g stateme ts, which a e subject to the safe ha bo p ovisio s of the P ivate Secu ities Litigatio Refo m Act of 1995. These fo wa d-looki g stateme ts i clude, amo g othe thi gs, stateme ts ega di g establishme t of a USFP MSA a d expected be efits the eof.

Reade s a e cautio ed that these fo wa d-looki g stateme ts a e o ly p edictio s a d may diffe mate ially f om actual futu e eve ts o esults due to a va iety of facto s, i cludi g: busi ess a d eco omic co ditio s a d g owth t e ds i the etwo ki g i dust y, ou custome ma kets a d va ious geog aphic egio s; global eco omic co ditio s a d u ce tai ties i the geopolitical e vi o me t; ove all i fo matio tech ology spe di g; the g owth a d evolutio of the I te et a d levels of capital spe di g o I te et-based systems; elia ce o ew p oduct i t oductio s a d acceptable ma ket accepta ce of those ew p oducts; a d othe facto s listed i Cisco’s most ece t epo ts o Fo m 10-K a d Fo m 10-Q filed with the SEC. A y fo wa d-looki g stateme ts i this elease a e based o limited i fo matio cu e tly available to Cisco, which is subject to cha ge, a d Cisco will ot ecessa ily update the i fo matio .

NVIDIA Fo wa d-Looki g Stateme tsCe tai stateme ts i this p ess elease i cludi g, but ot limited to, stateme ts as to: the be efits, pe fo ma ce a d capabilities of USFP a d NVIDIA Spect um-X Ethe et etwo ki g; NVIDIA’s collabo atio s with Cisco, TE Co ectivity a d Amphe ol; pla s fo a commu ity-led multi-sou ce p oject a d additio al pa t e s; a d the impact of AI o data ce te dema d a e fo wa d-looki g stateme ts withi the mea i g of Sectio 27A of the Secu ities Act of 1933, as ame ded, a d Sectio 21E of the Secu ities Excha ge Act of 1934, as ame ded, which a e subject to the “safe ha bo ” c eated by those sectio s based o ma ageme t’s beliefs a d assumptio s a d o i fo matio cu e tly available to ma ageme t a d a e subject to isks a d u ce tai ties that could cause esults to be mate ially diffe e t tha expectatio s. Impo ta t facto s that could cause actual esults to diffe mate ially i clude: global eco omic a d political co ditio s; NVIDIA’s elia ce o thi d pa ties to ma ufactu e, assemble, package a d test NVIDIA’s p oducts; the impact of tech ological developme t a d competitio ; developme t of ew p oducts a d tech ologies o e ha ceme ts to NVIDIA’s existi g p oduct a d tech ologies; ma ket accepta ce of NVIDIA’s p oducts o NVIDIA’s pa t e s’ p oducts; desig , ma ufactu i g o softwa e defects; cha ges i co sume p efe e ces o dema ds; cha ges i i dust y sta da ds a d i te faces; u expected loss of pe fo ma ce of NVIDIA’s p oducts o tech ologies whe i teg ated i to systems; a d cha ges i applicable laws a d egulatio s, as well as othe facto s detailed f om time to time i the most ece t epo ts NVIDIA files with the Secu ities a d Excha ge Commissio , o SEC, i cludi g, but ot limited to, its a ual epo t o Fo m 10-K a d qua te ly epo ts o Fo m 10-Q. Copies of epo ts filed with the SEC a e posted o the compa y’s website a d a e available f om NVIDIA without cha ge. These fo wa d-looki g stateme ts a e ot gua a tees of futu e pe fo ma ce a d speak o ly as of the date he eof, a d, except as equi ed by law, NVIDIA disclaims a y obligatio to update these fo wa d-looki g stateme ts to eflect futu e eve ts o ci cumsta ces.

© 2026 NVIDIA Co po atio . All ights ese ved. NVIDIA, the NVIDIA logo, a d othe NVIDIA p oduct a d se vice ames a e t adema ks a d/o egiste ed t adema ks of NVIDIA Co po atio i the U.S. a d othe cou t ies a d egio s. Othe compa y, p oduct, a d se vice ames may be t adema ks of the espective compa ies with which they a e associated.

A photo accompa yi g this a ou ceme t is available at https://www.globe ewswi e.com/NewsRoom/Attachme tNg/371a55a0-abdb-4755-a80b-f745caa4c38c