NVIDIA (NVDA) 24 May 232024 Q1 Earnings call transcript

Good afternoon. My name is David, and I'll be your conference operator today. At this time, I'd like to welcome everyone to NVIDIA's First Quarter Earnings Call. Today's conference is being recorded. All lines have been placed on mute to prevent any background noise. there'll question-and-answer remarks, speakers' the be session. a After Instructions] [Operator

may conference. Thank you. your you Jankowski, Simona begin

Good everyone, XXXX. for remind like replay webcast the webcast without available It you. being transcribed and Vice XXXX. our or the is on from will discuss today's Thank President to property. afternoon, Relations NVIDIA's second call are of be is you Chief call first call fiscal for the The NVIDIA's website. NVIDIA until Financial prior Colette our live quarter that for Investor Executive conference and of The results Officer; our written of call Kress, Huang, content Chief to reproduced fiscal can't NVIDIA's Jensen With and conference be financial quarter today consent. and me President welcome Executive to I'd Officer.

make call, may These a subject to we this current and on actual During significant uncertainties expectations. may statements are materially. and our of forward-looking number based risks differ results

we are law, and assume Commission. For our Securities as available no we currently the XXXX, to with business, future the release, X-K and based that a and most information by disclosure on the file of recent All financial could today's our in obligation made any our XX-Q, Exchange affect today, statements as that on refer to XX, and Forms please Except reports statements. results such XX-K May to required update us. of may Form factors discussion earnings

non-GAAP we During discuss will measures. call, this financial

on CFO GAAP these commentary, measures our financial in a our find You reconciliation to of financial website. can which is posted measures non-GAAP

the let to call that, Colette. me over And turn with

with sequentially XX% $X.XX channel gaming growth QX our inventory from billion, Strong revenue, corrections. and by record up Thanks, sequential was data visualization revenue emerging professional and center down platforms XX% was driven year-on-year. Simona.

NVIDIA accelerated fast to computing center, exponential compute XX% on is AI platform XX% up by driving data lowest opportunities train accelerated and strong sequentially across markets. upside worldwide. $X.XX deploy drove most broad-based for is growth creating of products, most global transition record versatile, a to AI year-on-year, with AI. approach significant growth and and up Generative growth the which Generative Starting our revenue our and and in billion was the in computing, energy-efficient, demand requirements TCO

me customer categories, service providers give our companies, major CSPs, enterprises. Let or consumer some color you three across and Internet cloud

the emerging at enterprise CSPs at availability providers their AWS, deploy Google including upcoming Azure, CoreWeave GPU Ampere First, to meet our and Lambda. the CSPs previews GPUs for and and and specialized general platforms, Cloud consumer inference. of Cloud, on and training Multiple HXXX world both cloud and Infrastructure, applications architecture Oracle like around the are private surge to racing offerings Hopper at AI interest from flagship in Microsoft availability announced

In adoption, addition strong to CSPs for are pioneers. these Generative demand from enterprise AI serving AI HXXX

adopting driving Internet consumer deep companies forefront the systems, and at learning-based recommendation are Generative of strong Second, AI growth. also

and production deployed AI Meta supercomputer teams. has research it's HXXX Teton Grand for now AI For powered example, its

Third, and enterprise strong. AI is demand for computing accelerated

as to (ph) momentum becoming quickly competitive positioning. customers' are innovation healthcare, and and AI in automotive, accelerated such computing where We are seeing telecom, verticals integral financial roadmaps and services,

and analysis, announced news it processing language Bloomberg For financial such classification, billion help tasks example, recognition, BloombergGPT question natural named has with answering. entity and $XX parameter sentiment model a to as

are their and fleet protein customer Deloitte and Intelligent and dispatches, is repairs. AI so AI engineering. logistics CCC Solutions NVIDIA is field to improve with better for for AT&T for working customers company, serve customers. Insurance using drug can other service technicians us Auto Among Amgen on estimating enterprise using And AI discovery

launch the of customers Cloud customers which HXXX, the started and the Microsoft we promise And DGX through we Cloud, of DGX Infrastructure, from partnership DGX our our can Oracle Cloud Hopper generation system, quarter, AI NVIDIA deploy to deliver This Azure, with shipping with on-prem. Google cloud.

to Whether Command, via frameworks, deploy NVIDIA pre-trained software, methods. and AI access the AI, and Cloud, data DGX them operating systems, our DGX models. expertise the AI and including expanding customers for they NVIDIA Base get and on-prem with We algorithms, or provide building training across blueprint processing,

generative for which announced We businesses video, and also large build, services Foundations, domain-specific and BioNeMo language refine, to unique Cloud, sciences. life models available trained on and XD, Picasso custom foundry our models, NVIDIA credit for that for with own language NVIDIA tasks. images, include for proprietary NVIDIA AI and are enable NeMo large AI model models, operate data, DGX They NVIDIA

knowledge-based has custom use and cases. service proprietary six data processing and for curation, optimized aligning, support for NVIDIA systems help models sector experts and for fine-tuning, elements, to enterprises Each their pre-trained guardrailing, from inference databases, fine-tune frameworks engines, models,

custom for enterprise specifically data leading models a They platform. large ServiceNow, developing early of on services DGX an the NeMo. trained platform are adopter cloud language and is ServiceNow

including developers. Our ServiceNow offerings, enterprise-grade the X,XXXs departments, create running with of collaboration ServiceNow let will IT on customer the for platform, worldwide teams, service generative enterprises new AI employees, and

of released with also AI increase inference industry, of platform diverse workforce. of in performance across in April, their versatility is that a inference unmatched workloads. Because require ahead these The the MLPerf complexities, size step-function magnitude is latest Generative industry deliver showed acceleration. workflows orders and driving benchmark NVIDIA's

To and language announced deploy for Core LLMs for These AI Hopper Tensor platforms recommendation systems Superchip graphics generative stack. the at AI LX vector NVIDIA LXX Grace scale, four leverage we video, applications that HXXX also, GTC, for and and for and major help models, the GPU databases. NVL include software large inference customers AI Omniverse, new rendering, at

with for the and the CSP Cloud's Google adopt Immersive to and is as AR its Stream, Google virtual AlphaFold, platform and inference experiences. launch such Google our Google Google inference render LX first workloads other Cloud Cloud AI machines Dataproc, which of Elf, GX generative XD

French cloud-based and our with Google server it's and engine is Triton, AI Google In platform. Kubernetes integrating addition, Vertex

purpose enterprise demand accelerated and and remain CSPs to Demand CPU infrastructure strong like networking, customers both NVIDIA's In platforms. AI generative networking which for soft. saw general Mellanox computing, high-performance networking require relating we at

grow applications in AI at the high meet delivering to complexity, accelerated are which performance networks enormous scale and essential for about demand data inferencing. generative and center computing As training size

is standard Our InfiniBand platform dedicated for platforms XXX the infrastructure, AI such adoption and Quantum-X Internet gold cloud Azure. consumer gig with major as broad Microsoft across

XX% in-network software their throughput of combination optimized and end-to-end the sizable in enjoy only for infrastructure increase data scale, investment. With computing customers the industry's technology stack, a routinely center

in including generative been adopted and highest BlueField-X the For multi-tenant and networking by Oracle customers, support DPUs Baidu, AI cloud our Ethernet Conference sharing transitioning more CSP available production COMPUTEX switching, and has Ethernet multiple Spectrum-X next offers network at Taiwan. with performance. is the BlueField-X gig Cloud, XXX to hyperscale accelerated platform AI to look and forward Microsoft high-speed Spectrum-X week platform Ethernet our Azure, in CoreWeave, others. We about

is the data Grace supercomputer. than supercomputer and CPU accelerated of with and The with for Superchip, this our wave week's International Grace NVIDIA which of Germany, opportunities adds CPU BlueField-X, center announced applications. in a Lastly, will is sampling centers. Grace efficient based generation At to new Conference previous Grace This Hopper the Supercomputing Xx and University energy-efficient CPU on customers. Bristol super the across more new cloud only the CPU/GPU a and coming Superchips enable supercomputing energy momentum data AI of both Grace growing

and was $X.XX solid, Strong end revenue down sales driven Overall, XX XX% Now, and let's RTX XX for energy sequential consistent featuring up DLSS desktops. seasonality, efficient resilience challenging and RTX GeForce GPUs was industrial to Tracing, and in Series sequentially, billion Ultra-Low against performance gains technologies. XX% GPU of four inventions, was creators. move demand spending growth gaming. demonstrating great a to Series by Gaming the Latency and of The notebooks GeForce battery X year-on-year. both off with are They RTX deliver Max-Q, consumer gamers for and design, backdrop. Reflex laptops life rendering, NVIDIA Path rendering, AI tremendous start, a

joined XXXX capabilities. XXXX, offers And large our XXXX, XXXX AI Xx installed-base or the launched support desktop, and RTX nearly technologies, XX upgrade. the RTX laptops spectacular a for previously ramped offerings, GPUs. Ti unmatched the XXXX creative tools and and we including is software content faster acceleration our In which RTX the The like science RTX and desktop XXXX, giving than data Studio NVIDIA Omniverse, and creators workflows, Series platform

newest Last launched price XXXX we be week, to the just in first today, and our XXXX time core performance XXXX July. The available points. starting architecture available XX at world's will $XXX. gamers the mainstream the starting at bringing provide the Ti, gaming Ti of family, for console RTX XXXX, the while the latest is Xx GPUs These

NVIDIA PCs to their the Build engineering, Microsoft drivers, on this help system end-to-end on and spanning NVIDIA time. Developer Core GPUs the to and Windows have for be Generative creation week, will operating content Tensor showcased GPUs, LLM gaming stations how we AI-powered graphics NVIDIA and a make Windows transformative and framework core. the collaborated RTX at platform supercharged AI NVIDIA from At with will generative NeMo's AI. earlier software Windows be Conference from Microsoft RTX development to work run to

announced Xbox we bring GeForce quarter, Last partnership a PC to games to with NOW. Microsoft

Gears richest set partnership, any with X,XXX gaming available games service. now to content There months. now GeForce on NOW, is this more X available released the from the are coming game on first be The over cloud in

Lovelace driven sequentially, by and was Healthcare, down believe cycle. the us. Visualization. workstation demand with The stronger Pro Ada mobile million our desktop correction verticals across workstations to major was XX% year-on-year. is and Public GPU ramp Moving Revenue a Automotive. form product of growth in and Sector, such kicked-off of strength XX% both architecture factors, Sequential in as behind We inventory $XXX key channel up

the announced AI rollout we major coming new Generative desktop for NVIDIA-powered workstation. for RTX GTC, planned in and workstations, workload with six At laptops further a quarters. GPUs is new

generative into for harnessing with to designers, ideal Microsoft windows elevate collaboration creativity creators productivity. Our AI and transformed the and platform their

Using the Cloud NVIDIA we GTC, managed can Omniverse of be in in available design, and service Omniverse NVIDIA stack year. environment, industrial that cloud Azure fully an applications. Cloud, starting develop, NVIDIA customers full includes Microsoft infrastructure. running of full will NVIDIA deploy, metaverse this applications second suite the OVX half Omniverse manage announced this and At

generation has adopter also will and design connect Jaguar to and Group, Rover. NVIDIA their to industry Omniverse. factories companies content been [indiscernible], such marketing. Microsoft by with Land XD including and The being leading companies applications automotive from used of Omniverse BMW digitalize General workflows Cloud as early engineering a Omniverse, Office smart for Motors, XXX is

X% Automotive. was up million, and to Moving a up Revenue XXX% from $XXX sequentially, ago. year

strong growth DRIVE the energy across of driven was vehicles. of NVIDIA ramp new a Orin year-on-year by the Our number

next $XX billion from pipeline up ago, $XX into As March, six announced us billion, years now our design years. continued a in the we visibility win over automotive growth stands over year coming at the giving

China in schedules to demand adjusting customers NAV's some growth. as are their moderated than growth expected Sequentially, reflect slower production

this the dynamic rest linger the to expect We for of calendar year.

the we NAV's. BYD our of leading with the expanded partnership (ph), During quarter, world's manufacturer

BYD's next-generation Our of start new Orin use DRIVE production and Dynasty, design to high-volume ocean vehicles, the to XXXX. win set its series extend calendar of in will

to were software. by innovating valued and absorbed P&L. prior gross non-GAAP now GAAP peak XX.X%, more and have as and and rest as largely to margins higher Moving the costs, delivering products recovered gross and more incorporating higher Gross products we have the well margins was of them margins XX.X%. level, offset

We gaming also GAAP down X%, investments through at We're and last OpEx the the delivering in operating level expect while and visualization. business were held inventory X%, corrections the roughly expenses Sequentially, leverage. the quarters. have operating same in professional working down to four expenses increase over operating past were now non-GAAP We

our returned have of through we December QX, repurchase $X million to share end shareholders XXXX. billion approximately the in authorization form At We dividends. the cash under the $XX remaining of

expected turn X%. plus me outlook billion, the second revenue fiscal quarter $XX for Total or the to is be Let to 'XX. minus

this steep and higher data second extended demand substantially for expect a data large we generative of to increase year. in to has the the related procured few a language driven growth by demand largely We center models. have quarters This our half sequential AI reflecting and be supply out center, visibility

million, non-GAAP expected are non-affiliated of gains investments. operating GAAP are respectively, and margins be other and $X.XX GAAP and are income plus $XX or and minus billion, respectively. non-GAAP XX% and gross $X.X be billion points. approximately expenses from income XX an be to expected basis XX.X% and to losses and non-GAAP GAAP to excluding approximately expected expenses

items. our and available Capital expected to on other or X%, Further expected GAAP to million XX%, in to are discrete excluding commentary plus be are financial $XXX IR website. tax are rates million. details approximately information $XXX the any non-GAAP minus CFO expenditures included and be

upcoming the XX person Monday, of will let May this local Jensen the U.S. keynote highlight me Taipei which evening time, closing, In in in COMPUTEX events, address will some be Sunday give coming the in

attending be The on June New AI Technology Global will June we of Rosenblatt Technology Francisco the Virtual Virtual San Summit on and addition, June X. In XX. Conference BofA on the in X, Transportation Street of Future And on Age Conference

call fiscal quarter to Wednesday, discuss earnings 'XX the Our of is August XX. for second scheduled results our

for the Well, poll that covers you now our We're would going questions. questions? opening please call for to remarks. open Operator,

(Operator take Thank from you. Goldman Toshiya Sachs. Instructions) question first We'll Hari our with

Your open. line is

Hi. on incredible outlook. Thank the the Good much taking you question so congrats afternoon. strong and for results, and

about you of problem to, beyond it's April a the one the what this from second July this what on visibility and supply in quarter perhaps data was sequential what more construct point, the your I'm come some data there, will of more to at center. drivers speak into the the curious can the you Colette, increase you quarter supply kind revenue growth majority chain mentioned are Thank you. half I the importantly, Just key of July guessing vast at is point? question can talked from support if of this center. sequential year.

Okay.

different lot there. of So, questions a

sure So, comments. can Jensen your see let start have following am some if me I I and will

QX, our sequential QX language our our So start-ups. models and AI our our and both when AI that the and Internet expecting large and are our growth between broad-based about a consumer across search it's we talk CSPs, we're companies, enterprises, demand, driving generative

our networking in our recommenders where growth. as as also same And at is driving interest the This not both Ampere well surprising our as Hopper to see computing growth area deep both often key in in we our also of time. generally This as architectures as of architectures, well our is a architecture. also we architecture both of are is latest business. It both our sell expect

we that So, think baked QX. guidance those are some provided when in we the of have to key things we the about

for in surfaced we also half. working a this amount quarter, We on our of remarks are the for today supply supply we second that opening but also procured both substantial have

chain see, we demand to and a We that we significant serve is some have supply of significant customers. across our that this wide customer see different flow range demand

and things companies. for at the They platforms but up the also are building the consumer of some large setting largest enterprises, AI CSPs

has there that's second for data led I'm So, our we add have visibility to quickly out right extended a see supply to now that working and us few on half. that center to more. little quarters demand Jensen if pause bit substantial and wants probably procuring the a for going

that color. thought great Thank you. was I

Next Muse we'll go ISI. Evercore C.J. with to

Your open. line is

part And cycle relate Number support kind you as managing terms taking another of commitments center, you manage afternoon. that times, and acceleration best Thank in Good we with yes, I with with other to where to demand? their to match your for into of questions are Thanks as much. question. you are that times doubling -- of where essentially years, servers thinking data and lead you kind are you the the Yeah. match two natural guess coming to AI? so to driving that, TSMC want best quarter-on-quarter, about mind. one deal come partners, with one, supply how of in there your longer

question. a on the large of we -- came. start when transition when and service models both I'll to it language the a remember, for Thanks -- moment chatbot. C.J. Hopper in how the helped I full product technology based were ChatGPT crystallize production Yeah, to of everybody backwards. and from And The Ampere

vector came human and search, it and moment, of it feedback, it's an all be together what the knowledge, helped call amazing knowledge to reason the the what that product everybody iPhone systems integration guardrails alignment all a proprietary really where The bases way in for connection have. can why that I can technology realize came and it's together reinforcement and data capabilities learning of wonderful

full chain our is significant chain we so NVIDIA's in as production. flow already and you supply And supply were know. very

and them in boards build XX,XXX of incredible the and and volume. are build these course, components. but GPUs, a we It And stand-up on of have that data in switches, system systems volume, GPUs, has the giant our we all and the supercomputers networking, includes, to NICs, NICs, us together transceivers come the the the fiberoptics, And the for and in Smart center. to other order

the And really significantly so second said. we for our were when production already Colette procurement in to as half the substantially came. We moment had full increase

computing Now, let been heard about accelerated Why it, time towards you've some for me if XX a do it centers but moving the data It's it me years. is we problem, stack taken has entire bigger about us that number and in could large world's of computing. but are talk known the is full challenge, domain talk a picture. full accelerated stack application successfully

data If entire be sufficiently major centers' the applications that almost could accelerated.

it you the didn't reduce the but lot forth, It and and order it substantially the and of have a You it you and takes center so build by for data for all of amount energy our software all the magnitude. do to have of XX -- so have cost amount because of years. do on an we to and to consumed could everything systems money costs

app happened what came computing when for that's is, time. this AI And for some along, in generative been triggered killer platform it a preparation

it's CPUs nearly all ourselves But a it trillion, (ph). over $X course. dumb transitions. in worth installed. by $X NICs growing And trillion on trillion And a so, simultaneous infrastructure basically billion entirely unaccelerated. of last is The completely and the see and call world's today, populated $X now two we $XXX of It's years, center CPUs four data based it's year,

the that AI world's future, the dramatically primary of very moment fact will that it the we speak. seeing with as now now shift that is towards the very workload accelerated it's clear with information, of this right generative budget is data and fairly center clear accelerated that the most centers -- -- now data the generating efficient, now. computing becoming so of and We're going you're that through energy computing In

CapEx data retool world's time, the but at While centers. the we're incredible is world's data orders center seeing to limited, budget same the

build starting beginning it world's to And so, XX-year the I the you're think and data computing. accelerated seeing a transition -- or out of it as centers call basically recycle you're reclaim

You'll have to workload from the with be accelerated NICs, and the predominantly course, the spend to computing, GPUs, a dramatic switches, computing and shift smart AI. pretty of center traditional data in smart is of generative going

we'll move And Securities. to next Vivek our Arya question, BofA with

Your open. line is

wanted environment, other very Thanks center clarify over just continue competition QX data the the what to or more is competitive competitive given custom other question. does solutions? that to see the at question and for this landscape solutions QX in GPU invite of the next to in sales do they levels? I of do change ASICs? then to strong does wanted competition landscape? or that, I invite years? sequentially demand terms in that. visibility sustain more it How kinds it my three QX clarify just Could mean you terms to can grow Does do of Does two Jensen, And

for Thanks question. Vivek. Yeah, the

can half will than HX. little of will second bit Let substantially the add if have supply for be me see the we year We color. more believe larger that the a that I

in increase last expecting we of demand our the half the Yes, this on we to but that. be not in a first seeing So, we to the have not here. the in quarter, second here But also are the second demand are have just to we on something that that QX do careful just for compared We year. the saw half. plan in half forecast, second we substantial of only planning the half guide

every over Regarding direction. from well-funded competition, startups, we really-really and innovative of world. have them all competition Start-ups countless the

existing competitions from have companies. We semiconductor

many CSPs of of you have about know We most these. projects. from internal And competition with

cost all time. We're solution. of the are mindful core the we're TCO and the so, the NVIDIA's at we time, get And all proposition lowest value competition lowest competition solution. But is, the we

of that because libraries for is of engineer it's is, problem, is two you all optimize things of we the of that as talk architecture stack one integrated the and frameworks, for it full computing, the I way often, algorithms, software optimize it's distributed all entire the work computing hardest know. And the which have and the center, and a accelerated into data to not way computer fundamental It the them really the all just is the is reason all the into amount extraordinary. challenge, and quite about And engineering stack into the models. frameworks full a but ship and science all architecture computing the an the

whole and a stacks. optimize it's one, number you thing of and just blowing number the so, challenge across to And full mind stack across have it the

We acceleration libraries. have XXX

As and we the that of amount you accelerate is blowing. pretty libraries know, frameworks mind

like is that that is data center chip, center is computing thinking system not the is part you're And another the generative scale is a fabric, what your to networking it's in engines, particular networking computing operating the the understanding data second your that The the a computer happened center or data system, it's never data before. and architecture it's way trying problem, distributed systems, the the entire of operate. large switches environment, computing scale to of center the your it's computer, that's gear, the and your the AI problem, it's and of this this and computer

the that's understand data have get and center you what best to performance, understand is. And so accelerated computing to order you stack in and full have to scale,

that utilization talks the accelerate thing -- amount our about architecture utilization, which second The high. types the diversity you of that is of applications of and keeps can that

is stacks, -- problems. problems center GPU so many that's one number then one the throughput, scale universal that's makes you data and incredibly largely fast, your and and And architecture only we software-intensive one do up. it's so incredibly hard that is, in that center is accelerate our and thing can thing data to and fact If utilization a software thing do and underutilized high,

is The over our data helped centers we've just and third versatility data into world's integrate clouds. We've of center the problem build and second expertise. our data world all architecture is built we centers the own utilization and five companies the all

about of measured to not up standing if is deployment, the up see, data the half operations -- center good a of a to delivery at not you're it, coming product the let's supercomputer, the about year were not it installed delivery From world see it of the do proficient – to in a take moment a time some operations the the at of and supercomputers unheard online, months. so could to and – in largest ago it of and and not now they're not a Standing it's of year.

delivery and it and the supercomputers operation doing measured that into and turned And data in is we've expertise so. taken is of to incredible, weeks. in we've products, team Our centers the and

possible highest translates So, the the infrastructure, our the of throughput all final technology lowest value analysis, in in this proposition is in cost. the

very, I very very so great. But is -- is the think And large. course our competitive, really-really market of challenge

we to Rakers with Fargo. Aaron Wells Next go

line open. Your is

the question the congrats for quarter. Thank Yeah. on you and taking

of terms of revenue forward? your agreements only service of of just -- that various drivers grow? terms the how I'm suite other going of expansion we're the AI you should monetization to what drivers data cloud kind considering forward, curious, we effect of enterprise where software software, business software of think approach unpacking of curious and do in As at think I'm think the growth continues different that the about the going center we in of Colette, about

Thanks for the question.

platforms. and is software are important products we including Software to have accelerated amount Not do nearest that that essentially, we in only have. of our a really our architecture we substantial all

different now AI to help accelerated with generative and computing. their are models many start in customers work We

that So, those as essentially anything importance or and a helping them NVIDIA here models providing discussed that we've build system operating services, of we AI. from AI the for have Cloud DGX enterprise,

availability as should this things both well. offering, as architecture Jensen, to all needs if ability going both that to as infrastructure, over the we [indiscernible] and turn of well grow to I'll our as the monetize the So, it forward, add. he continue to

growth models, can AI CSPs, as models. the deploying refining the the see of real-time and We Yeah. models, for generative in training the well both as

earlier, AI used As major Colette said capably because of so so now computing driver applications is already. accelerated is many a inference generative in

of software, industrials. all are Enterprise requires stack of segments two capabilities that ability large principles, and two because talked proprietary your about, we've your own There segments whether enterprise to own it it's and the that need and new language and align operating requires the many have stack domains. use-case your for data, your enterprises models, the software and proprietary a a to new to are adapt,

the You have want it have deploy you create your clouds. do Cloud be vendors to multiple and a that chatbot or we deploy might your on and agreements able sandbox, you AI the computing in on you a Cloud, to and depending model. that ability in and performance applications, DGX any Cloud, services to high Then because want to you with have various

accelerated have for NVIDIA a AI stack that that and you custom X,XXX AI do, are to different Enterprise safe, There patching there of And AI are AI Enterprise in supported. create GPU that constant engine NVIDIA the helping we Foundation the accelerated for Enterprise the have Enterprise buildup is NVIDIA workflow. enterprise, packages world the end-to-end you operating and we operating engine, NVIDIA have is models the stack, AI and only Enterprise. entire represents

XX%, that to data big time you data, to deal. training. lot data, model, your It's and have the a computing that and have so, have obviously, AI prepare whole could very That in amount order process package an do of of just align ingestion, up it you train and kind of curate, the to processing, processing is data from XX% of bunch data a XX%, you to to stuff of consume and only one data for there's its some and

And the continuously refining aspect X,XXX then security an for training to is the NVIDIA AI every for that and just is and software. wants model all as like patches as that packages order Linux, model, of And the to to incredibly of Red patches well deploy Hat supports the those has second be in want their and it it of to this Enterprise model is in inferencing. Enterprise be deploying deploy they software complicated has engines, and secure, it on-prem, to deploy supported. third cloud

AI second NVIDIA point. so, Enterprise the is And

third The Omniverse. is

the as feedback. In that an And need you Just technology reinforcement AI reinforcement Omniverse the Omniverse the in ethics ethics, industries. robotics, for starting includes vital an applications to engine realize need to of AI And for feedback. robotics, for align a it's are human learning physics. people for called software to is align case aligning AI learning and you to and defined a applications robotic same industrial

also so, Omniverse service needs platform. And cloud a to be

And Cloud software of you partnerships the in the us runs well Omniverse. make cloud, think in team -- both into GTP cloud on order and in types marketing and AI it AI we clouds, software AI I CSP of as sales of have With world's need that world's team the the the to as putting Foundation, Enterprise software clouds have into DGX with to it's partner world's order leadership and partnerships Cloud the and and vendors. is and the a for into great have us the AI to stacks for great three to world's need and so the new we the all cloud, of partnerships, new Oracle, in -- reach way all integrate the all the for it and them happen integrated Enterprise industries, you extend of into AI in and Enterprise beyond so AI stacks, by that partnerships Azure, Omniverse two DGX the three stack, our we with.

Next UBS. to Arcuri we'll Timothy with go

is line Open. Your

Thanks well. I as and a then a a question clarification lot. had I had

sort on with first crowding a -- you the the Thanks. question solutions the side the your on clarification, So, having on out versus there shipping argument, maybe compute you InfiniBand that the low Was timing? of that side? is, to core that's sort still Ethernet rate debate attach can on playing buyback that can about AI, you InfiniBand low Jensen, share out? then is authorization. the InfiniBand you to Ethernet you but the of of like And despite late for see what talk share and billion latency you're the know you Colette, of maybe are wasn't whether the it similarly how $X just about I repo and need speak compute

you Colette, take the go how about question. first? You should

That is correct.

not We that anything have $X in go consider we for forward Thankyou We but we'll do authorization as available repurchase repurchase did billion in we as last opportunistically recurrent and well. this quarter, repurchases.

costs infrastructure if different. and continuously for a two exceptional have both that record data has applications are is InfiniBand number, it's going a And very InfiniBand $XXX really quarter. different to going the and for InfiniBand It's their a designed that in They right. case center data Target to a an All networks giant Ethernet an few is a are for NVIDIA's have you record InfiniBand a doing place. AI pick Quantum But roadmap. million. If applications it We're has specific people factory, year. you, -- incredible. will. and a had a center. really few running InfiniBand be use

is data using center shared so, cloud jobs, datacenter and your a of great a too little It's and data InfiniBand it is little $XXX -- effectively why of if difference And and bunch if The jobs difference and use center InfiniBand difference basically $XXX of is is people. throughput the a infrastructure overall the millions could it. multi-tenant. in one The free. million, an just and in free. bunch be between its XX% and to application Ethernet people it's by XX% that a for and million InfiniBand XX%, it's you That's in XX% ignore, spent to is is reason you're throughput. however,

is really is line fantastically numbers and it I is Ethernet generative AI referred at AI announce workloads. new a But record which GTC. is Then InfiniBand it's major for of AI answer? Ethernet at factory per clouds. AI middle we're I segment cloud. where focused Cloud quarter-on-quarter to going do not this It's generative At There's a and year-on-year. application segment, doing doing to still COMPUTEX, a we're COMPUTEX, wonderful it the opportunity type last product This a the se. but only multi-tenant But generative a new segment wants run becoming Cloud is the in to

Next Research. to with Bernstein we'll go Stacy Rasgon

Your line is open.

versus Hi, generative a guys. for on my I question Thanks training had AI. for inference question. taking

bigger about scales if parts inference XXx, two it you you're if even one-and-done. there Is bigger? it's sub is guess, those influence us very Xx, how bigger, basically be can give can are within versus on qualitatively, just give Is inference it sort a think you the So, helpful. opportunity. that, training vice-versa, some is a like And more two do you talking much would of than like being that. like large AI, usage us the it it anything generative as if besides is I of to it's with opportunity, like training is or like is workloads like

I'll work Yeah. backwards.

training. with done never You're

with You're data. new collect new every the you data. always -- deploy, time When you're collecting data, you new you train

so, never And done training. you're

large that producing model. done never language database a You're and augments processing the Vector

You're have. data you never -- collected done with vectorizing all unstructured structured the of that

these recommender language system, a but very so, core of if applications obviously are you whether It's three are the running. of important the stuff, bunch a you're are a very building these all three computing. major of future other database, of And large the ones. three vector probably engines, always They a will, model,

and processing companies You're data, factory the dedicated forth. data in intelligence, so more-and-more largely and case, on to going so have they see that factory a particular training representation intelligence for data of and to realize learning that and and an vectorizing it's

the -- of of of in of themselves, connect all of course, of is be or them will could could an APIs But many kinds in APIs inference Companies. there partnering that are connected And use can Internet of can company, come that it, they'll application. be part companies that Adobe into part integrated generative XXXs be applications, workflows their The all create whole of are APIs that from some workflows. some to AI Foundations. a bunch built AI they APIs APIs Service of them ServiceNow of into kinds, them we're And as open then that bunch companies like and there'll with a APIs whole either that

we in very that significant of So, APIs. heard center The about seeing the In you're we've market Moore's AI growth enough time actually and XXX% Law. is a the really didn't inference end in by the be practically simple has the data -- the but segment for think before, and as places keynote year it this segment it now the first Factories, as very of weak ISC exponentially, of know I there a that, to simultaneously CPUs. they end, think in a trillion growing a way world $X it's used exist was the installed future, well I think with that to for AI a is about

the heavily accelerated center that next net would think That the lean CapEx so them the Every is generative GPUs the think think it accelerated into trillion of would everywhere scale-out or used as connect with AI, now computing forward about in four quarters that largely compensating well general-purpose about is of chips and infrastructure. networking accelerated easiest way AI seen easiest then got lot budget a into killer $X and that and that's that that training so data and that XX to switches all budget and it over adjusting all. as most five, growth the the still, will probably inference. generative to and to trillion way -- and very be that's $X the years, for got number the in to path is computing in it's places now easiest you it's We've reasonably data computing that centers accelerated from AI CapEx your and of way a that be about can't app infrastructure generative the capital

go to Stanley. Morgan with Moore Next, Joseph we'll

open. Your line is

I focus on are on significant want Great. customers models, talking the is terms that It's inference. but of about large query the to in to by per big opportunity around a really very clear cost that, Thank also reduce pretty trying you. follow-up language cloud amounts. this

ramifications guys, how get cost customers come at you about the of that talk the you down? can that per to and going the are your for is query specialty GTC You where of products launched in just insurance help some that you

Yeah. all the align so you the ones, on into the surprising, large But the and can and you or it PC -- a And the large seems of it size. and they That's great tiny biggest so can a phone one have they your tiny one things. ones, thing. language start lot more by be teach on large you have same how the zero the amazing model, would generalizeability and put large versatile the the model more in forth Whether distill medium, And so large could sized do to your a question. and building one that very can smaller and they AIs much and your smaller you all version ones shot so, and to use and good and obviously, forth. you but to generate all language use -- small model the large prompts language is and -- do them good so

have large of a train by start And smaller then very ones. And you to bunch whole also building so you ones.

reason Now, why inference. the our have so of sizes we that's many exactly different

HXXX LX, HXXX of model have LXX, announced you sizes there kind also HXXX we I that so NVLink multi-node which and -- saw NBL have -- have then like. with is you any that and You HGX could

to out. are have video in, The in, other important future, out, in, thing ultimately video the is, XD proteins could the text XD out, in, they And are but text these in video out, graphics image applications that's in, out, models, connected applications. image

of The output ignored. pre most post-processing one the and and it rest chip the like the arguments processing of is of length and post-processing, it preprocessing The post-processing. call the is of fall pre decoding, the that. be the all of is security, that because So, a specialized requires lot things overall only XX% the of things model can't And of inference. input itself it's kinds and -- this -- of the data and apart about

multi-cloud, going great services. ServiceNow, have in that done just why we so, Enterprise done the on-prem. done It's have with I we of day, multimodality going great a that's to be clouds. other be on-prem, reason a it's called the multi Adobe, announced going third-party of Cloud reason of have integrated be Helix, partnership they're be the Dell whole a be going AI to to inference, in That's the all the And think AI It's with generative it's capabilities. creating the why because going into and aspect project partnership the to diversity in reason that's why the we to the inference, we bunch

there's you so, to I really need broad, fundamental described, diversity, the all it. like AI just so generative And in and order the whole address of space some what reach to have very of capabilities the is

Harlan to JP go Sur we'll Morgan. Next with

Your line open. is

and an appreciate I the congratulations compute really on Hi. some the today mean, integral strong your of the the products. Good focus afternoon, really your maximize in of of platforms. sort to results performance execution. and I networking it's of focus more part or full

quarter plus part a $X X.Xx I minus, think billion so guys right, data growth acquired networking revenues of business years is you driving from that's center three Mellanox. per when of or ago,

complexity, on? accelerated datasets, attach team networking So compute your platforms, the compute further requirements up keep with more very then, your bandwidth better forward Ethernet going networking And with doing what just given latency, in the the high predictability, run-rate of for and unlock is stepping to InfiniBand, so the to is traffic shipment? compute growth, increase solutions, lower strong pace significant line but your in very

appreciate really I Harlan. Yeah, that.

some Magnum and possible chip early-on are but everybody for the the DOCA about understand, it we our this and GPUs. is they nearly think makes in networking, about point I've it it's the software IO. of completely. accelerated fact, of accelerator very So, computing hard AI, call And pieces stack the misses software connect that the chip, about announced to it it, of mentioned a These we who crown company. library have before thinks Nobody stack, about to acceleration about networking of talks and the of nearly called jewels ever us X,XXXs XXs that remember, two whole because

about networking, which the connect the of company. the things great How performance so the great, of that's do Mellanox that networking know if why the XXs companies was reason center, in why of the data highest and performance high our unambiguous two quite and -- have have operating reason infrastructure, the GPUs, world's of that's XXXXs well, as we is you in insanely the And one so are you system leader together. we're not obsessed we is the

memory lot also more running GPUs, all it the with a a GPU. fiber our are low then, InfiniBand, are expands You an talk and latency which NVLink, optics that NIC, full COMPUTEX from starting to about production I'll These NVLink, that And of network and And we in we includes we're optimized a things I that described, going and that And see that references, inside at the few which super rates. with the end-to-end. really using multiple a is it days. at communicates the BlueField-X then gets connect switches, network fabric computing not package. in incredible take beyond connected line the and SmartNIC

complicated. stack together, COMPUTEX, have -- brand AI through the announcing we'll is switches, want at factory Ethernet this units type fabric then the and we And connect software beyond computing this all if GPUs that smart a that, AI into whole the you fabric, and smart networking, all computing -- so your of the area computing be way extending -- connecting of of to these fact new factory the through the the insanely

-- because as think be. as you computing so, whole the of thing delighted should and this it particularly, platform a understand out, we I'm it don't break but -- it, And we're we

it it us We Cloud center why of way That's but centers whatever kinds, as reach. in like XG all world's the components, NVIDIA's edge it, it doing this kinds run is out the way it's all data up, or so makes of to from it integrated integrate different we they anybody's all so we and style it but on-prem the reason to can into that to for stack. all sell our of do would it architecture different more we be of it they way and way that software can computing data still incredible break that complicated complicated, gives to world possible the the into architecture really to the

last with come Cowen. Ramsay Matt our question will from And TD

Your line is open.

much. Jensen, whole very and team. the to Thank Congratulations, you

I little One wanted dig into of the the a DGX is bit things to Cloud offering.

it over then hyperscale your been rest guys some on your you lease business, working the behind back scenes, and for to own for hardware in partners the You time sell have the few found this where and in business. And months to discussed of Colette of data out the us of high that publicly ago. it the visibility number a look forward center quarters as we next kind about

Cloud own a upside in DGX little is potential of seeing mix what forward in, about internal systems customers you're Maybe to business. sort first-party Thanks. of your you buying you've own versus hyperscale for data what customers third-party, could that their that the it that of selling center with you've offerings workloads of their going potential the you're and support learned own big talk since versus launched bit their about

infrastructure, our NVIDIA CSPs numbers, and and like like us deeply CSPs the performance. Yeah. the possible is reason too with Cloud the gives the the but to XX% and NVIDIA It ideal Thanks about highest-performing is way is something mix DGX -- ideal Matt. the partner the to It's -- one. stack -- being specific create ability DGX we Cloud scenario, XX% the It pure very NVIDIA stack. clouds, the without and best architected number the achieves the is

with two, with never partner the create partnering us we're Omniverse stack. that, had like, world's to it bring example, a cloud to the for world's allows like Azure industries. Number computing the markets CSPs to to the system And

the stuff, with virtual a large virtual kind high-speed the all generative physics world before. the existed all and incredibly low-latency and Now has database that networks, stuff networks really and and of XD never AI stuff industrial

we Microsoft cloud create inside Azure Omniverse with to partnered And so, cloud.

on but one; that having create they through cloud. two, number us amazing number getting new together of that things computing So, our by GCP platform benefit all OCI one we together. two, And go-to-market and the in develop security by Omniverse to benefit instantly applications services services data and number team of to and it access allows us the have, and markets cloud, they Azure and and as amount their we and and can our customers new have

a it's win-win. huge so And

NVIDIA's applications, And can for the it these that cloud customers, they the anywhere. works way early for do

standard to to one take in and cloud themselves, CSPs would NVIDIA it NVIDIA the will and So take themselves software but manage clouds little run if that, long-term, runs NVIDIA the a their AI run by it they longer, is and Foundations. all AI like stack going on we're this clouds. the CSPs And Omniverse in Enterprise, because delighted

Okay.

incredibly. in So, were in markets including reasons out run we're primary on-prem the and our architecture our doing to -- the goal and customers that everywhere, -- provide applications for new to and with is deeply and new to partner flexibilities worked really creating their drive their so, in the it's those it that

we spectacular. CSPs three quite in the great. currently Cloud works force and marketing teams, is with that team DGX and have and really partnership sales their It leadership Our their

back Thank closing Huang it you. Jensen remarks. turn I'll for now over to

and is general-purpose strong is era. is $X has best accelerated racing yet computing based data full The that supercharged. simultaneous computer computing industry center AI. the now global now are generative method AI data trillion the to the generative through computing accelerated in computing, path era. generative deploy stack center pioneered computing CPU last Companies and slowed, approach NVIDIA demand scale infrastructure, forward. scaling going There installed the transitions, AI Accelerated with on for is the a of and two

of Over the data most will next accelerated. decade, be world's the centers

significantly models models, generate language applications. encoded AI in surging forms. fine-tune, Large labs large are its models generative our to Guided to content meet by can and from emerging principles to the can to amazing supply We and facts, models industrial is many AI is learn increasing ground with on way the information guardrail, demand. align generative and guiding language

platform platforms building one with the partners help we enterprise for on-prem or their enterprises CSP our Adobe bespoke world's Helix, Internet leading on safely. we generative can AI a or Whether a expertise are ServiceNow of with and Foundations, within AI As Dell with NVIDIA like harness enterprises. also largest whether securely domain leverage and we to service data cloud and scale and providers,

Spectrum quarters, coming and sustainable data will products at next. in our production. that Grace We also X are is chips you. and show They BlueField-X HXXX, COMPUTEX energy-efficient are we'll the what's our in Grace platform. of help scale next Hopper ramping networking week computing. They and deliver and a computing wave Thank us Join super and including you center all

This call. concludes today's conference

disconnect. may You now