NVIDIA (NVDA) 20 Nov 242025 Q3 Earnings call transcript

Good afternoon. My name is Jay, and I'll be your conference operator today. At this time, I would like to welcome everyone to NVIDIA's third quarter earnings call. [Operator Instructions] Thank you. Stewart Stecker, you may begin your conference.

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

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

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

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

reconciliation measures You With which of financial find GAAP me non-GAAP turn financial let that, a can posted on to to over in Colette. measures commentary, CFO website. our these is our the call

record another quarter. QX was Thank you, Stewart.

Center. of and up All and NVIDIA adoption strong AI. Starting deliver Revenue billion. was growth, and posted fueled year-on-year growth. of XX% to Data XX% and incredible computing sequential by sequentially continue above billion of platforms with We our $XX.X well accelerated market $XX.X outlook the year-over-year up

NVIDIA and billions, delivers half approximately and and doubled large ramped faster fueling CoreWeave, AI Ampere the HXXX-powered Google as Azure year-on-year AWS, largest models cloud Cloud inference infrastructure infrastructures cloud year-on-year are NVIDIA world. business to sales up of sequentially, Revenue Alongside companies available regions double-digit rapidly service OCI GPU Center. more NVIDIA jumped rising recommender to installations fastest and and engines, India, serve Another revenue HXXX agentic and significantly and Hopper and and improved for regional XXX% prod inference training soon. record year-on-year. CSPs from and instances revenue sales in content providers exceptional, increasing for Xx and billion, coming deep our up in than in is XX% NVIDIA Pacific cloud NVIDIA up inference demand inference creation with of and next-generation history. platform Internet their networking their Cloud to the support workloads. scaled instances revenue CSPs, company's now were $XX.X than The performance revenue was and growth Asia XX% to up thousands to to significant Hopper AI, increased growth learning achieved with multimodal, Microsoft AI NVIDIA sovereign infrastructure Center sequential the training, NVIDIA generative America, are ramp our customers. North NVIDIA workloads. HXXX scaling of Data year-on-year. our Cloud inference AI grow and TCO. high-speed HXXX with Data DPUs NVIDIA from of demand tens as Hopper is more deployed Xx build-outs. Consumer

software Xx. token throughput inference time performance optimize in and by NVIDIA and TCO algorithms large Hopper to encourage boosted by to an Xx incredible X improvements. Rapid year developers advancements base deliver for ecosystem Our first and NVIDIA and installed software in rich cut continued

the Blackwell air data market GPUs, scale liquid to third cooled, increasingly successfully market. Blackwell up a NVLink to to AI now NVIDIA quarter, infrastructure, installed returns engineering with boost DGX NVIDIA samples a hands xXX every to economic they the Our center the GPU a samples upcoming full We in is full is training major release all in stack, racing to NIM customer the address AI data from is cooled growing in system customers and be inferencing the of first a InfiniBand XX,XXX are after for to entire are ARM, first shipped Blackwell OpenAI. Ethernet of to to switches, centers. executed working customizable inference of by change. to of and and and performance production our including an of full diverse one performance Hopper additional bring Continuous base. Blackwell to partners, and configurations from optimizations their needed will NVIDIA X.Xx. drive is hallmark

of a Blackwell Midjourney, NVIDIA model. created. Mistral, Hundreds the Podium, are platform leaders wave of with are headliners. results development announced Harvey, racing Anthropic, customers. great are Bridge building is Meta, that Blackwell NVIDIA of meet center Microsoft models. Yesterday, to leap cloud Microsoft, to first instances offer, scaling, the demanding Cohesity, us. agents AI scale these clusters Cloudera, or start-ups Blackwell GPT-X GBXXX accelerate The services diverse the systems the with services is and We time to of compute. Xx and Industry are and companies Firefly, applications micro NIMs run running The is response of years. Purser, build to NVIDIA, incredible services. Enterprise and industrial first Enterprise, AI-native deployed and delivering NeMo the on per staggering, Salesforce, is billions are SAP, start-ups seeing Enterprise faster relentless shift, Adobe delivering the in that new placing Blackwell with AI thousands of into over AI AI technology pursuit Switch reasoning Blackwell AI data With oX HXXXs to at Blackwell its AI an for build-out. to next cloud world's help applications train and while we be and a to the our XXx compared AI new already in AI the enterprises help Perplexity, global recent to deploy are and coming Runway, launched are to computing group to cost. NVIDIA in excellent potential they leaders ServiceNow level new GPUs demonstrate customers AI-native private be AI enables scale the a are for reduction AI. benchmarks performance of CSP gearing drive the up some Accenture Copilots new XXX Deloitte InfiniBand. enterprises. demand next-generation to Cadence, to world's scale. the by the Blackwell can Zettascale to with throughput and powered this agentic and Quantum supply XX agents. Working are like with in Google, XX,XXX also on which of X.Xx GPUs of NVLink business benchmark throttle. and performance operating are wave in our integrating inference to NVIDIA MLPerf OpenAI a facilitate XXX,XXX demand OpenAI's like racing configurations required architecture are success platform and new Hopper. week, taking success. Through most a Oracle up of and Customers debut inference training announced Accenture results, Tricks, cost Blackwell-based on The are NVIDIA AI of professionals NVIDIA great the NVIDIA the includes full Consulting is deploy trained AI. GPU using down to Last made round most will preview, Light every over of sweeping are NetApp, inference

including internally, speed AI NIM, campaigns Additionally, companies is of employees, AI its the case manual for in Accenture cuts monetization. in is using NVIDIA XX%. with uptake over NVIDIA-powered applications evident steps XXX,XXX to Nearly X X,XXX agentic that XX% and NVIDIA Enterprise leveraging marketing are

to robotics year NVIDIA are enterprise and at We expect NeMo over at NVIDIA We operate pipeline that This physical developers Enterprise annualizing we billion. Industrial year support accelerating. AI agents. software, exit by for models build. the over physical AI in robotics. expect Omniverse is Xx world, billion, and AI our train, build, and annualizing triggered last service $X understand industrial from to like AI, and revenue to foundation revenue for breakthroughs Overall, built AI to is full our increase $X.X and this and NVIDIA continues year

Some world AI the NVIDIA the export-compliant hour the drive levels our adopting expects in achieve to Foxconn, reduce to its a shipments and operating Omniverse workflows, largest Hopper new NVIDIA in data alone, sequentially facility reduction bring-up over their XX% using China to geographic automate industrial industrial factories the revenue In speed Foxconn is are Omniverse usage. From and grew manufacturers Blackwell new electronics efficiency. in twin levels businesses, manufacturer, to of industries. built on due to Mexico of a and largest products its efficiency. to accelerate of of -- of perspective, their world's center annual of digital

As total remained a export onset center of to controls. it the of percentage below well levels revenue, data prior

market the in remain competitive going very China to expect forward. We

customers. export continue serving while our will comply We to with controls

AI are tens as consultants NVIDIA upskilling by new year-end, India's to initiatives building our on agents with adopting product and TFC, for by boosted developers industrial services they factories Wipro accelerated of most platform. In clients a SoftBank momentum to AI. communications revolution million continue By building Our InfiniBand. and run GPU country have Enterprise nearly NVIDIA Blackwell the XXx. AI is nearly are in data for gather AI NVIDIA deployments DGX the GPUs. of AI supercomputer embrace countries NVIDIA and and AI Infosys, powerful and thousands build Quantum powered leading computing nation's include will help CSPs Japan, NVIDIA X.X

also with Aerial RAN distributed NVIDIA that a the network process NVIDIA AI telecommunications transform CUDA. on SoftBank XG can partnering platform both into and is AI with on ARN network to AI

to massive build power Ethernet. leverages platform to deployments. Traditional cluster data Utilizing Xx application maintained degradation Spectrum-X, achieve compute. Ethernet Ethernet InfiniBand clusters. and customers supercomputer our NVIDIA XXX,XXX-Hopper not throughput uniquely NVIDIA with AI XX% and DPUs. companies Japan, multiple help pipeline sequentially launching are of consumer NEC, And for centers year-on-year. for to and to adopting continues are NVIDIA revenue of increased their Enterprise large enable HXXX designed planning down, in the revenue networking with for traditional versus CSPs are Colossus X Consulting technology U.S. technology AI NVIDIA QX. AI demand anticipate revenue include across to growth consulting Leaders industries. Networking growth and XX% and including the using and Though Strategy adopting strong and and and major was sequential bring Areas SmartNICs switches, and InfiniBand networking T-Mobile. growing, NVIDIA EY Fujitsu, sequential including we Japan's BlueField in InfiniBand Ethernet increased experienced XX% over was latency will same Spectrum-X the AI. and Internet exclusive supercomputing year-on-year. NTT Spectrum-X new to previously CSPs We scale is companies, of xAI's revenue

Now coding. This time. for to AI editing, with a moving Microsoft's Copilot+ we to desktop executing QX driving supercharge revenue increased to Gaming celebrated These technologies PC image devices ray consequential end revenue graphics sales gaming, XX% great of GPUs console, XX% applications. Channel RTX Gaming The and QX. the video and the first was back-to-school are choose RTX strong to revolution. most demand anniversary power from and in RTX technologies and XXX, with GPU. and with Gaming power quarter, photo, growing GeForce as and PCs. harness began NVIDIA's up behind up by the was and remains machines world's new and quarter transforming holiday have TOPS all AI gaming, the sequentially $X.X AI GPUs AI healthy and sequentially igniting We we force to anticipated and creative, some GeForce our season. year-on-year. the generation to gearing RTX notebook, for AI XXth MSI and XXX year-on-year. fueled the billion tracing consumers of of continue inventory shipping the GeForce the of been past ASUS capabilities of AI

engineering-related XX% sequentially continue professional Moving workloads. the to NVIDIA and RTX graphics, to preferred million $XXX ProViz. choice workstations be of power was and year-on-year. up to X% Revenue design,

powerful prototyping demand vehicle driver, generative and use productivity-related for AI entertainment. emerging content creation as is including and media Additionally, model AI in simulation, [indiscernible] a autonomous cases

and out and end demand sequentially by Moving electric SUV Automotive. Strong XX% on growth rolling was NAVs. Orin brands XX% driven Orin Revenue DriveOS. record up NVIDIA for year-on-year. Global fully to and was a market million, Cars is NVIDIA robust built its $XXX self-driving of up

GAAP engineering shareholders form higher-cost incorporates rest QX, continued turn of gross is for Center. operating and up the outlook HXXX expenses was plus gross cash we shift to systems Blackwell architecture $XX.X to to moving margin fourth billion, product non-GAAP of and and infrastructure share initial in margin Okay, expenses driven to and Data Sequentially, within by products. P&L. higher returned the billion to Total me compute, due XX%, operating repurchases GAAP of and was revenue sequentially, Well, down costs development systems a were be complex which the primarily the to demand and of new XX.X% $XX.X more quarter. X% mix the introductions. In X%, Hopper or for ramp let expected the our dividends. the minus for non-GAAP

While supply, estimate respectively, quarter or gross exceed to as expected be to customizable of fourth was due infrastructure AI of NVIDIA-built XX sell-through dollars non-GAAP demand track increase. On our QX, billion and options previous Gaming, expect with revenue we minus XX% into although and we revenue X visibility decline to a to XX.X%, greatly supply plus strong exceed chips, networking is are air cooled on sequentially Blackwell several and Blackwell margins for our and continues are different data liquid supply constraints. multiple types centers. to points. basis GAAP in

the system current strong availability, of and providing our Our demand, on is configurations optimal ramping to focus customer. mix to increasing

we margins expect billion, Blackwell ramped, expenses When XXs. $X.X billion GAAP expected moderate to expect gross low the $X.X respectively. in are we to operating and approximately mid-XXs. to non-GAAP As and Blackwell be ramps, be to the margins fully

center-scale data company. a are infrastructure We AI

expenses and new are and Our GAAP be GAAP other upcoming are X%, details rates our nonaffiliated on plus website. In commentary investments IR an tax $XXX financial non-GAAP gains non-GAAP of excluding highlight or let community. and approximately centers from excluding million, closing, for investments. XX.X%, to building are to development income losses data support of in and expected hardware the and financial the expected software other be Further and to for include information minus income CFO stacks me discrete any included our and items. events available introductions.

the Conference Jensen in UBS will where next session We CES Please will a X attending on the X. December and Scottsdale. in be January a host January for And X. day will on Vegas, Q&A at financial we AI analysts Global join us on deliver Technology Las keynote

fourth Our quarter earnings XXXX to call discuss the XXXX. scheduled results of fiscal XX, February for for is

please? for for now will Operator, call the We open questions. can poll you questions,

of of C.J. first Cantor comes from Your question the Instructions] [Operator line Muse Fitzgerald.

very as for debate part here on front. question large the a Blackwell. have obviously, from How stalled. just to would here, work are these guess but on they I then for you benefit early through you to And language helping context the is your models around clusters your yet of discussing whether love we're Obviously, scaling customers hear thoughts we're have issues? that this

greater driving demand for Blackwell? So even is this

Foundation and scaling model pretraining intact it's is continuing.

ways however, it scale. now to a this is the know, we're empirical fundamental other What law, it's you scale. But physical X not post-training scaling. that learning, is As evidence that an law. is we've is discovered enough, to continues not that One

it's And now we feedback, generated produces. of of our like Of all it thinks, of your reflect in and one have intuitively, assists scaling. And and bit to thinking of test techniques and head called forth. doing exciting it's answer and the like first scaling, of generation forms on chain Strawberry, developments little us one multi-path post-training data thought planning question. human higher-quality inference And and and considers it what most and time a synthetic answer learning AI reinforcement it in so does course, we so which all longer scaling. necessary better is is biggest The before the data approaches -- the feedback kinds oX, ChatGPT of that OpenAI's but oX, learning reinforcement was time events the post-training

result X And so demand we is ways now scaling that, have all infrastructure seeing great. of ways X a really scaling. of and we're the for as And our of

about models generation XXX,XXX now of see Blackwells. were You foundation next starts the at at generation the last tail end that at XXX,XXX Hoppers. The of

that inference of importantly, respect very a And kind you is to scaling, then scaling. the gives now time industry post-training scaling, and sense of moving pretraining where with so

is really for we're demand the scale so great And to our starting simultaneously, remember, up for of reasons. But inference seeing all those company. really

the inference. inference world and extraordinary Amperes everything that Hoppers. infrastructure large. for was large And platform for base largest models, on is our incredibly move it training foundation behind installed Amperes the on today so a installed in We because inference are and we base leads And as trained Blackwells to Hoppers of it

And demand up. to so adoption is really the number course, of inference native starting of And AI latest see AI the We're companies time scaling seeing to We go we're agentic seeing continue of up. we're grow. see rage. enterprise inference go

seeing places. so lot of lot a different from And a we're of demand coming

next from Your question comes Hari of Toshiya Goldman Sachs. of line the

change the mass earlier some this reports executed the some were over you year. heating Jensen, There about weekend issues.

we've On this, super Ruben coming HBN? as that Are that's to be year speak to GTC sort map to on to constraints year then a of helpful next speak road execute on supply that some your to multitude Is of investors B. quick and would it you is getting if sort that? helpful. And constraints, ability well. causing Ultra of at supply worsening? in And Any the is investors it componentry execute the would or constraints? out this this part can On you a had supply they to it better? with specifically ask Are the your questioning be super back sort you presented are time, Can the color about ability of of so that, 'XX. transition

production is Blackwell in to let's So Back question. full see. first the steam.

Blackwells previously estimated. earlier, we this as deliver In quarter than Colette had more will fact, we mentioned

our supply working AI to do of we're generation as exciting of case reasoning that's that understands and AI, doing expected continue Blackwell revolution partners And going that of physical we're beginning increase is that in new the as to all models able to really exceeds team the our work the AI do we're incredible Blackwell, of an we know. one is foundation the physical areas next this increase the generative with long chain is of a to beginnings at the now course, of and to hard so job structure supply year. It through And are supply. able demand And world. to the And thinking.

And so very is strong. demand Blackwell

Our lot that obviously doing we're world. is a of across going there's execution the engineering well. And

now see and You up being are by systems CoreWeave. that stood Dell

saw you systems think I stood from Oracle up.

You have and from Grace to Blackwell they're about Microsoft, their preview systems systems.

You have systems are that Google. at

And go chips, architectures to see ship you in supercomputer are NVLink as a racing respect for cooled we're do of or the we us XX. so process, reason this quarter custom and X to are order we full cooled, That all or all XX, done the or and disaggregate Blackwell around that to The custom something of NVLink infrastructure, lot With and the built integrate what point. But being X chips we greater it in this systems that because AI NVLink are still still XX we engineering is different stood systems. but generations now. estimates. for that X, that CSPs deliver several that and The shape. the engineering you Blackwell NVLink air systems liquid data with all as we mentioned of supply we in as rather be NVLink full them build there is, chain, is the And previous world. know, planning to the happens X of earlier, into great stack these we've first. it complicated. And this Blackwell all our very is of than from at up, We're centers integration the supply at there's it's good although

systems short have or data into the xXX all those nothing Grace. miracle. centers world's We And integration of is of of a the

systems in a ramp you scale, the was supply shipped is last this we X. will much look the how terms billions, of how total measured Blackwell Blackwell go which which take in at necessary component ship this so chain ramp And at back to incredible. and have quarter, And is quarter, much to

Spill And is connector factories and involved almost chain. they've KYEC, and and partners, so involved be SK in Hynix seems And our incredible, partners every of Blackwell, Amphenol, company, Foxconn everybody TSMC built we've supply Quanta quite the in Micro, really Dell win. and I'm great And just company Amcor, in gosh, to and Super we the and and I the and really course, from, of and and Quanta. incredible ramping there's missed Vertiv are got appreciate. the the which And Lenovo sure companies company. world that I've HP, of that number Micron,

some XXX factor data data could size. past, when center of very be it's the so accessible. But that's in partners. it that But a data now so important realize course, cost And our are and expecting course, we data always it's performance, are. more continue a could to by highest the the It the highest to power this in in of factors the We're our the then watt doing do of for limited. respect data of doesn't AI annual road point. And fixed the training. there's so now to the anyways, our so, per center fixed be, We're of you're megawatts of into matter we're is and shape And when is really the at how on some our to cost that much of centers to important centers megawatts performance of question of performance execution directly and reducing The several -- and when And data your -- I note We're at power-limited other an inferencing. tens lastly, to Blackwell great reducing hundred on the reducing is size map. platform. to think it and that we're of center, about increase at megawatts, most increased annual road the translates our we're planning map. gigawatt really revenues map we're centers, with large the able best on cost road ramp execute we're time, the also

And so our X revenues. perf But because greatest for to good we generate the our hand, possible watt our road compared so the customers reduces on on anything per there, annual map the hand, cost. other is out

so annual important rhythm that. And of and have that as is on far I we really us, do everything as And to continuing to know. track is intentions every

next Timothy from Your Arcuri UBS. of comes the line of question

billions if is recent can in talk think like this in April I'm wondering also, just the more I going you to that said about quarter. the that. over know to Blackwell better you than did several trajectory you're sounds Jensen, of than of dollars crosses But about I you said being think you do Hopper January. Blackwell talk going ramp months Blackwell how I It year. in had

So I guess I had X questions.

down in is then margin First think Hopper it kind April? you will it, Blackwell Colette, And gross cross Blackwell way that to still of the over of XXs all, talked about bringing as ramps. that to the right about low

worst So us. XXs sort April. if I'm is soon can is crossover, of of wondering of as in if gross the you're margin? going the on the be I the for low just that guess pressure that April as shape to kind you So

start? you why Colette, don't

Blackwell the going chips focus in that the question, bringing gross your experience on beginning first our and are let they Tim, very making that are regarding ramping our the gross you, we margins. sure discussed we the up. best start And many to we with margins we our as to market, thank are for configurations, different Sure, many we customers that as have stand different me

be but XXs into first in do gross start believe we in will that our low program. We those margins the will growing

you're the that, quite So start get after that increasing the margins, as hope part quarters to our following and of quickly you mid-XXs we correct, round. look at to gross as we will

and ship continue year, ship next surely through than next first that we'll after the of And will next quarter more more Blackwells meanwhile, Blackwells Hopper several the first this, we than year. the quarters our demand will quarter. quarter

it And that of puts so kind perspective. in

at significant. really beginnings fundamental is shifts computing quite really in We that the X are of

creates learning coding no widespread that coding machine to moving at to networks first this fundamental are that that And GPUs. is that is The companies learning not from point. who from do neural CPUs to runs learning. shift machine machine on on There runs going are

enables And also AI. generative so machine what learning is

happening And the data systems the thing trillion being that's learning. so of X machine is $X centers is modernized and the worth first on hand, for around world computing now

something. capability AI. On guess, AI secondarily, I new the generating be of when going systems factories. we're type creating top be these other hand, to a saying these say AI, going that -- are that data on are to we're is of essentially They're called And we centers generative really

electricity is these going Just just if like of be of now many large, large, factory. electricity, just customers is as we generators are running AI AI services like And the are number generate we're to going consumers an XX/X. to XX/X, generating today, AI. And be the number running of

online, the to what new it it close see this type data It's we're so a come going as because to unlike And past. of center AI factory an and really I system call that's is. of

X really these so And just fundamental are trends beginning.

to this several happen, this growth, for modernization of to a expect creation we this new so years. and And go the industry on

comes line of Arya the Bank of America of Your question next Securities. Vivek from

it's should you clarify a of of 'XX? mid-XXs the to wanted think assumption this recover that prepared be hardware they when of calendar the hardware Just historically, is because continue you required just Blackwell? So of get main back just do calendar premature gross how of of to clarify, we a think to way And too seen think half included Can to long-term the to Just my fair that. see we do kind have is along kind period could Colette, satisfy margin in NVIDIA grow some first to we how we deployment? you're digestion seen think 'XX? Or this start wave? inevitably it phase? cycles, to a then many deployment secular what into just to way. have right, historically, at you question, Jensen, that you kind quarters of have digestion along discuss shipments to the When do of

you the thank Vivek, Okay. question. for

reach Let regarding gross definitely question ramp us to yes, how it see goal margins. assumption but me the I of But for that And mix or clarify reasonable do, in the think we to yes, half a mid-XXs of we'll it just is is Could have your possible. goes. next second year?

centers, we if The I be and is that, when just Those data vast through for centers. It's until will the time is Vivek, thing to there them we data you way CPUs. on -- we that anymore. world's applications digestion of just modernize sensible the to look hand a it at think no ran by $X believe majority wrote with not the do a built trillion

have they center -- it data ready for you data every of plenty they're machine they if CapEx, to if have learning tomorrow, company's and ought because a build generative old future If build AI to centers. of

know, the XX% over say. let's data into of the so assume IT you and happen And could be going the world's continues that number as what's grow X we X about let's XX%, grow course centers years, a next of year, years, of course to to IT, over as the modernized

call And let's data say computing by world's XXXX, is, so centers a couple dollars. of for the -- but trillion it,

to grow We have into that.

one. center have modernize We learning. from coding the data That's machine to to number

the new is segment had. of never producing a world's world's generative now part type we're of it that never AI. known, market And the new capability second The a

completely new. when wasn't it It's didn't at something It really It's, that's the anything. brand of If as replace in you look anything. a came, brand completely iPhone was replacing ways lot new. OpenAI,

going that. we're more and services, of And their they're like essentially and create intelligence. more to see generate, out so And to companies going

would intelligence of Some Runway. digital like be it

basic would Some like it intelligence be of OpenAI.

there it riders, just intelligence marketing companies, like would -- AI-native Some every And cloud-first hundreds. of these There and Harvey. recall. are like you be companies. was shift, so legal companies, in mobile-first they were Digital are intelligence what as companies. There there almost platform on Internet forth. were And number -- of were these companies, called, the so

they're AI Now natives.

platform a opportunity and a do shift, people are new. because companies brand-new see these And created to there's so there's something completely that being

to of artificial going two, is these And factories a continue production out that going we're the number sense one; for be are to modernize number computing, for AI so to new IT, then to my intelligence. industry and build modernize that create

line of of question comes Stacy from next Research. Bernstein Rasgon the Your

had into so like Is sequentially like quarter? the XXs I QX? of versus it mean, revenues Blackwell revenues, if say Hopper quarter or a guiding just Is you. QX Blackwell question, said the clarification mind? Blackwell for as off China XXs have was any And pretty kind a can be now on data question down Hopper of count you be because you would next and that. But is do low into it why? low clarification, the The behavior color must still Colette, the also supply helpful. of gross a constraints? next dollars. strong, Hopper margins, ramp than when give been several else I But strong. XX.X% in like and does is, something bit rolling for you're you really be it billion and more my total So up us so up sounds should And China in total center sort has is you

have well. So to course, first, XX%, be margin at below could say, define XX%, We're we and there, the let's year through Stacy, to be in first we get might going about going we of of go that We that make range. want low. So question Low, our as product mids the ramping starting and the to higher are improvement, regarding our We're comes that in We improvement improvement rest we'll gross be than see on and of to is maybe how the just yields, as the your the continuing XX.X%. terms up by of to in terms that do through. the it point. that sure mid-XXs

that not statement doing? was an seen. amazing and Hopper We orders but and quickness The ramping only regarding we've our second fastest of for growing substantial standing in HXXX have what terms a of that is are out. terms the growth and question our in those It that is it's seen the product Hopper

our that in China. But include is their QX looking of out configurations mind sure. may that Blackwell. terms do selling board and continue in will keep terms for this to at configurations, our in what across in quarter, folks time, We different of are of the That the be all build we Hopper also, same to in

have a little we've see. got But grow but yes, It's to is to So QX. bit between QX? and possible of happening for both in we'll it possible QX just Hopper

of the comes of Your Morgan Stanley. next Moore line Joseph question from

talk if what wonder I bit you little market. in inference you're the a about seeing Great. could

of these projects. Strawberry that scaling the and use also of chips you've the for you You've as But could that talked influence clusters some about possibility some some of talked the Hopper inference. ramifications about of longer Hopper age,

to guess of And in outgrow inference the kind training there. generally, do next So just XX your you thoughts time months frame? I expect

does inside really Our supply ton group company of hopes and exceeded legal AI department And when for coding engineering, single and when and their of chain the doing companies the that's of and department world is course. and dreams every is a department has course, inference forecasting and inference. that someday, marketing their is

tokens whole Outlook hope And native of PDF, of and your that PowerPointing is or bunch tokens. time startups there AI inference are sitting Excel, we a bunch of with every you of XX/X. And open generating will from there using doing constantly And that AI a be computer a generated a every it PDF, you're read experience generating so to generating you're native tokens. thousands when And every whole that AI. aspect startups, company

every of the of my through I is archived to paper to put I One application out. came it it living applications listen it's just because NotebookLM, it used fun. And it. Google as just scanning as daylights PDF, out favorite into well that this every

goal so called of And that so whole new you AI. models a it. a if to now physical is that's genre era AI, think these of new train will, I the AI people And there's use whole

future. the physical what's AI the Physical if structure understands Just valuable and predict, process, incredibly short what and what you AI happen That and how will. and capability and the those does could of out the sensible what's roll understands not it is robotics. but the understand industrial for meaning it a large only models thinking won't. And language human not language understand and world. And understands can it

that's just so so feedback, from learn you're instead why human we robotics built Omniverse feedback. companies we companies now To AI learn fired and reinforcement, probably in physical created to it's the created we be AIs a capabilities, that physics native physics learning can and companies and many hearing enable and is data that generation enable up these about. feedback was And can these AI and of so physical that And really reason Omniverse. synthetic Omniverse Omniverse have it's so AI.

The goal that to And so starting and is to see happening. generate -- the we're to tokens. inference, that is goal growth

about So I'm super that. excited

thing. the Now more is the me is X why be just reason say is high super inference super X And the let to need hard because you Inference accuracy hard. hand. on

to you need as low. And to understanding the throughput these latency be high to do. low context could the that but understand. they're want are latency asked to need hard build. to as possible, be being You the what well be inference applications so as lengths computers to high-throughput context cost also have incredibly as want They within be to of able is because they the that low long And

And larger. the length context so and is growing larger

multimodality. models getting larger. On are the the They're hand, other

everything of And and were happen, top their is ecosystem if we ours. innovation if inference innovate can CUDA is architecture NVIDIA's more is they work. what and so innovate probably architecture, fantastic. number -- the of And our quickly that something that Just Everybody because on knows it's not is innovating makes rate of this dimensions top NVIDIA's code and they know to that likely should great incredible. they

single same time, that every around so that center a single base is to having the in every at all an really the edge that innovate quite create way and could whatever installed the NVIDIA you to broadly ability on capability direction systems, deployed computer land robotic large all data And into in the so be phenomenal. world out

the Aaron of of from Wells Rakers Your next question comes line Fargo.

Data and Center ask strong to as wanted the focus you then the When about seeing kind you Blackwell networking were results the your quarter, very I that Colette, think about you, that down was were last cycle mentioned XX% demand. sequentially. look business. of I at But obviously, business comments on we the this

for mentioned design that You large-scale CFP also wins cloud clusters. you had these multiple

business and had of pace that billions progressing seen going in you've dollars about. the So previously I'm confidence your curious of maybe multiple unpack constraints could and where if you the some what's Networking on you that to in talked just Spectrum-X

the The we growth been Mellanox tremendous. of of our focus critical networking. a about networking center. the is has work The in terms in And with start acquisition part do our beginning first of together that that. Let's really of building the the year-over-year is data since such

center continuing Our do networking and our are systems that data quite ability sell to well. with to doing in our grow of is many we

terms of existing only quarter will this large using we're growing. systems incorporated these down networking be more and the going our that slight systems going dip but be So providing up them a be in not to is also networking for and just in lot back are right more We're we of that a is and getting Blackwell ready to. to

of Malik comes Your the Citi. line next of question Atif from

mentioned I can Colette. explain towards update demand quick situation billions. And low ones you're gaming? Can the on supply-constrained have an for because X that? you center? last Is provide data then in that that shifting double-digit sovereign on Colette, is the earnings supply your call, you in you

they building world. in and AI, countries the call that a starting lot first something are today and of of surfaced important doing. onset that them the growth, on with see AI individual really of the around is So And we and a talked the we work an part of such models in terms generative them, lot of sovereign the about

AI is AI in terms also see of in in working of up many and/or own that So build well. as only parts stood see to this going you'll in language, forward terms are areas are those growth but think in our This intact within terms and this being sovereign those not and be this as seeing working terms our focusing are growing terms foundational that clouds may our their their culture, those absolutely in these models own for pipeline countries. And factories you sovereign that enterprises regional of with opportunities in in of the Asia you're growth AI. I is is in continue the still of of Europe, the where Pac to

Gaming. flip me your to Let that regarding you asked question second

ramp Gaming gaming all a we fast. given now saw right make what different quite So supply, our we're of our was to in through our busy supply, selling can this that we from sure moving trying And case, products.

we year. into is be that Not supply to the get fast how turn the challenge corner to this we'll We're going more new worry, we as supply the ready calendar market tight for getting quarter. be on for think the that Now we could have track with into this just back quarter. I

line comes Ben Your the Melius Research. next of from of question Reitzes

and with to regard Jensen sequential Colette I wanted ask to growth.

would you and I the with regard have then some administration any wondering but of So may you comments sense you prescriptive just situation, in was X%. be on? if go quarter, China not Blackwell could too get there. And any to your in or Any China tariffs what had this with supply that strong very there So to growth more first half, the in or from be sequential thoughts overall sense that's the catch-up? the gotten change anything Jensen, or on going about how wondering to conversations about your take probably here just place business? any may you Do guiding you're U.S. as there. we reaccelerate It's imply and early there

X guide at a We quarter time.

the to we're what and need Blackwell. in ship of that in building right on working terms now are we We quarter

terms ramp that on we'll in supplier once see the every help that. that And with that. have after next quarter, next to working to seamlessly you get the understand us of planet going We to quarter we we'll

decides, administration And then our just of do. highest we can that's mandate. new always course, that, -- best as we do And the after administration. the Whatever the we'll, support and the

in comes regulation any customers have to best And will the with the of support to all so marketplace. of simultaneously comply X we do our we to abilities these along that and simultaneously. things our and We'll compete fully and

line New Your Ferragu Research. final the of from Street question Pierre of comes

there growth pretraining? into clients overall inference of Jensen, one becomes compute And reinforcement? as the you How and if in the of AI where comments, high-level splitting was out your itself. buckets? reinforcement the large pretraining, sense have much language you much how inference your it's have is then how more you well inference ecosystem, for today? models most and that of are as And goes models actual or more of for mentioned and much today, the out an that have these into important How well? training you the wondering each how sense the a of I learning have typical Do the and like and much sense in for you have a the of maybe like important any one you

do And as Well, that you to because, as a as pretraining so inference low know, online. for you coming the could post-training, vastly would post-training, in just cost today, it's and new model foundation are in you technologies could whatever the be try do possible pretraining everyone.

However, many there that so a could only you are things priority. do

and on-the-spot have always to do thinking and thinking you'll context in so And a reflection.

based actually models foundation incredible. is it's we on to trained petabytes model, we scaling that the on, that area multimodality And so are all foundation and amount going are fact these the be I have think foundation of X the very where now models video that in And are. sensible

And so for and revolution going we're why that and hard by as scaling the the is as as the the to increase drive to we're we need time be and we the the is time drive continue more increasing pretraining, compute. And have to going. performance keep I out revenues continue X cost well we're as which post-training and scaling reason future, so foreseeable going expectation more my down AI to going factors think to can get inference to of that can to

remarks. closing like the back for call to to Huang I'd you. Jensen over Thank turn

X The is driving being NVIDIA Thank you. of global fundamental tremendous that trends computing. our by fueled in business are growth adoption

to is base a undergoing platform of neural Software trillion is for a GPUs. networks stack which learning computing code on being center from processing The the learning, to X.X, from shift CPUs machine machine traditional to First, data on AI. rebuilt produce coding reinvention, infrastructure installed $X executing applies

a grow more Demand can capability ago. The a full are introduction than of the steam. full which Second, computing multi-trillion-dollar and startups and AI There a ChatGPT pretraining industrial now native exponentially. successful is but emerged. There age test new makers model that a number AI of reasons. year ever production, is are law inference not for in the several with is intelligence, anticipation new AI foundation a Blackwell, create were just oX, post-training factories for industry there to incredible now services the more has new revolution and called industry. manufacturing AI continues new Hopper is Generative than are for AI time digital OpenAI scaling of in rising. And scaling scale a of with oX, software

better. Investments awakened national diverse. fundamental of NVIDIA's their upon expertise, the developing entire it's stack for hyperscale AI from clouds, and full full ability trends and cloud and every robotics we are seeing cloud, Over is Thanks across of faster regional of company, Omniverse AI models importance and robotics age time, is recognize Enterprises infrastructure on AI employees industrial private demand generated foundation robotics. to multi-trillion-dollar ahead adopting industry, to the lets deliver to of catch country. edge train driving the AI and and consume of sovereign next surging All large AI joining AI, is in infrastructure. The coming. revolutionize in have are and synthetically to a computing. new opportunities and us as data. scale, deal age workflows. agentic breakthroughs the petabytes in due physical us world researchers on-prem great infrastructure industrial time. training world robotics assist up and video today, to are The transforming enterprise AI Countries AI serve will jobs and to us and performing these coworkers every their of AI

call. conference today's concludes This

You may now disconnect.