NVIDIA (NVDA) 21 Feb 242024 Q4 Earnings call transcript

Good afternoon. My name is Rob, and I will be your conference operator today. At this time, I would like to welcome everyone to the NVIDIA's fourth quarter earnings call. [Operator Instructions] Thank you. Simona Jankowski, you may begin your conference.

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

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

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

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

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

record billion. XXX% another Revenue $XX.X up of was QX quarter. well outlook and year-on-year up Simona. above of sequentially Thanks, and billion was our XX% $XX

fiscal was billion, infrastructure from year. to a is XXX% Center Center. of from has purpose Center year. Starting The Data tipping tripling XXXX, base Data revenue Data year the computing from accelerated and The was $XX.X the the for world of fiscal era. $XX.X trillion-dollar transitioning prior than computing. new rapidly prior reached up general installed For billion point more revenue with XXXX

the performance, purpose-built in valuable InfiniBand XX% workload improvement generative to tripled Center Moore's computing companies AI. In year-on-year, revenue drive may Centers, Law factories, every and skywalk, in Data end-to-end demand $XX.X to companies driven TCO up Compute billion Data record, of modern efficiency. revenue last fourth to have than era refer and a same networking generation the quarter, intelligence was revenue At networking. we produce platform, raw while sequentially data Hopper refine to computing NVIDIA and the and Xx to year. accelerate of possible time, by of with along the As next build slows AI what as energy future continues up grew more XXX% GPU from started the

delighted that are supply for is We products architecture strong. Hopper Demand of remains very improving. Hopper

generating Center revenue, grew cases are on-premise. more past internal their the at Many Data services and for supercharge than GitHub months for of and Data suite cloud including quarter the providers Microsoft use revenue Business Center AI large its next AI cloud workloads Microsoft reached virtually has deep Companies deploying adopters quarter, noted models year, to AI search strategy. In and products contributing developers, and adoption improved hyperscales, conversation are revenue. hybrid cited and including inference one customers Copilot in of of our more estimate We external across the across a systems. inference. Building to every improving data public and fourth constrained investment first a and of accelerated broad cloud performance stack AI approximately Center through environments, and consumer use the video in supporting more Copilot largest cloud NVIDIA and the and contributing computing AI services and on-premise learning-based and news revenue infrastructure range organizations cases, customers. training for are click-through early and represent extends or leading companies seamlessly releases. The that a productivity Enterprises was GPU training quarter been performance scale. solutions of deploy AI our in NVIDIA's to industries strong have by acceleration AI multi-cloud latest both our AI These their across processing XX,XXX or engagement, and investments represented to use half of XX% social its using expect from recently industry. customer specialized GitHub major ad AI by as private operating significant Center faster its as allowing training than CUDA media, companies XX% to with cloud year-on-year. Data of X generative cloud on inference, providers, advertiser generation models growth of return exceeds regions. high of e-commerce, growth across a a predictions driven customer platform far previous many language to and accurate for return The workloads. X Data We broad versatility set enables be demand and and XXX XXX our entertainment rates. recommendation supply to Microsoft both supply. Fourth accelerating enterprise industries, in Internet categories. was than And Meta large

their AI ad latest inference partnered net of family customers customers to in descriptions and automation notable quarter creation, AI value creators, we've already of release. for are contract content largest new the training help In content seeing productivity All through applying customers are assistance. product commercial new software generative investing generative ever product Internet any support and in addition, contribution consumer generative generative and both advertisers and companies AI shopping to online realize AI success. ServiceNow's annual Enterprise gains. drove for with companies are products their AI tools

breakthrough platforms leading models many of Adobe, with of as working and Google, are foundation Mistral, platforms and software Anthropic, Getty building generative amazing with enterprises well, Snowflake. The are serve Microsoft, is are to We and AI. continued Exciting SAP, and AIXX, Cohere, Adept, large leading including Databricks, like enterprise in other Images, Inflection, Runway OpenAI AI field Perplexity creators. companies language xAI Character.AI, thriving, and

and to driving foundation creating the Recursion, for specific providers. cloud the serve to LLMs And generative are many New regions. demand hyperscale customs are pharmaceuticals biology. world's like These companies models industries NVIDIA for languages, are and cultures GPU-specialized different or startups address infrastructure biomedicines creating entirely others of AI through

in Just accelerate to model this data NVIDIA morning, Google we cloud, performance and Gemma PC. new the on we their collaborated with its language announced that optimize GPUs to center, state-of-the-art inference

cockpit NVIDIA AI to on estimate working Data vertical the the frameworks in $X as Automotive ingestion, autonomous plus adopt fraud to using Tier labeling, with across Center the last X through XX including as the robotics, automated NVIDIA exceeded trucking, discovery, drug contribution financial and through working curation, such of train DRIVE for AI industry of trends multiple company manufacturers and help infrastructure accelerated year NVIDIA. cloud vehicle enterprises applications. In data is autonomous most software new across revenue AI that is AI detection Almost billion the development Automotive, such year. global offers past by application robotaxi or and full-stack AI training, of solutions notable domains the companies significant or systems verticals effect, driving, care, for and and One services. learning AI infrastructure of NVIDIA's over models low-latency our adoption the health vertical computing platform. We driving, every are nearly on-prem OEMs, for leveraging and include AI validation simulation. vehicles, LLMs other driving Automotive suppliers energy machine

care, cameras, equipped transformers move to to drug more video AV demand with imaging, biology digital AI processing Center reinvent Data are significantly. In generative medical and As health are we discovery, grow cars devices. surgery, and wearable NVIDIA's automotive algorithms expect and helping to

models be past are the BioNeMo for creating platform health fraud the and using set model generative Recursion is discovery for can NVIDIA BioNeMo to biomolecular domain cases have We features a proprietary use customer over management develop, growing AI drug risk for ecosystem. In available computer-aided detection. discovery. the service growing discovery a BioNeMo, of financial drug customize AI trading processes. care and making announced end-to-end expertise applied and AI to customers service services, from We AI the NVIDIA models through Clara for care of decade, built drug AI and collection the deploy to that pre-trained a deep in health foundation

American Express Growth October. geography. detection by fraud accuracy strong Center all revenue revenue was the control using Center imposed Data For our by in except improved regulations example, to for X% following declined regions U.S. NVIDIA our export across Data AI. Shifting government significantly where China,

Countries market. the percentage we alternatives shipping first started stay don't to the China, has outside have quarter. In from ship require license world driver. for additional in a China, we local their AI not and their China products we U.S. Data our demand are a and the restricted similar in of the computing infrastructure ecosystems. From revenue in majority facto as for building product Together, the on they large and models emerged along around domestic support of language a of Hopper of language was de have driven support and received QX, range an represented the U.S. the government research licenses accelerated Although expect have perspective, own infrastructure. to in vast in sovereign to our that architecture data mid-single-digit China investing a Center with standard in of the and networking. it become AI revenue by regions to AI InfiniBand enterprise

initial with in the billion HXXX We quarter. exceeded inference are HXXX. nearly $XX ramp a track revenue HXXX is Networking second as shipments the strong of performance run on rate. to Demand doubled annualized

solutions centers. networking define end-to-end modern data Our AI

the the highest-performance grew more is Our InfiniBand solutions Xx Quantum for year-on-year. Quantum AI-dedicated standard NVIDIA InfiniBand infrastructures. than

Supermicro processing with technologies end-to-end their for networking Spectrum networking and stack HPE, networking Dell, now entering in over Center. OEMs, worldwide. Ethernet. traditional new DPU launch new Technologies designed compared solution including introduces enterprises higher BlueField X.Xx sales an of are channels AI us our Spectrum-X We partnering Ethernet AI Lenovo for Data and the to AI. Spectrum-X to incorporated our with that are with for for software AI-optimized space switch, global with the performance our Ethernet Leading purpose-built the are offering expand deliver

track this to quarter. We on ship are Spectrum-X

Cloud with also of run annualized which QX. R&D partners as DGX and of NVIDIA's as great NVIDIA NVIDIA to list Microsoft rate model AI services its for joining Azure, in CSP developers. include DGX custom to We an and development our and Cloud announced CUDA the well ecosystem used progress expand will Cloud. $X We revenue AWS, software made offerings, It partners. that reached NVIDIA own Google brings is Oracle Amazon's billion Cloud,

performance during Windows, AI Gaming. AI up AI OEM. to $XX.XX image ACE tools and PCs tracing won XX%. generative has awards. faster the consumer and of and games, GeForce that libraries, developers to generative and better GeForce PCs. And GPUs. RTX XX-inch announced the RTX installed faster for CES, to DLSS we AI. gaming Gaming AI AI With better NVIDIA next-generation RTX avatars. also SDKs, large are Starting which performance major be demand Xx Fiscal coming XXX productivity. The Avatar NVIDIA and light digital end-to-end performing library everyday bring billion, integrate XXX flat offer incorporate making of thin wave new up and perfect our powering our AI CES, our PCs generative AI $XXX, up open-source outlook NVIDIA AI-generated, CES, moving to At platform XXX start. of for XXx, further AI-enabled Tensor workstations. an inference XXXX into for and holidays. of we a deliver workloads. by to with we capabilities. run accelerates Okay, was generative next With TOPS, on our allow gaming to for and on up revenue can DPUs applications Xx year-on-year, GPUs billion our Sales Series with sequentially RTX accelerates every AI of building up into family can This AI XXX XX% deploying is of X applications generative RTX several AI generative gaming, great software services announced AI RTX quality. performance a and wave of than over PC. AI Series high-performance was and year XX the up AI X laptops. the latest These $X.XX language these performance, models, can deliver PC we Cloud up models Super CES stack RTX the on incredible their was developers state-of-the-art Best to at GPUs platforms. At creating they off NVIDIA solid pixels revenue them fueling TOPS applications ray million for wide factors, for generative At base AI and Cores the rich XX best our TensorRT out AI resulting the to for PCs technologies, including microservices, laptops increase range a of With Engine over AI capabilities announced form the and are to to way. in and from NVIDIA we PC LLM includes

also include support and workloads Automotive, developers, as build, as RTX workstations Ada retrieval their it robotics. In to driven workflows NVIDIA Visualization. their Automotive we architecture to of building up was to simulate, XXX% preparation, experience XX% car bring to Pro technology industrial $XXX CES, driving industry generation. adopter buying in Omniverse-powered of was million digitize Omniverse realism new X%. ramp. key of configurators. At experience. cars. and rich early demand GPUs a and personalization, levels Brickland, generative and partners Revenue AI-related adopting and automakers are to augmented to design, operate of to year announced These $X.XX factories has and mix ZeroLight like Enterprises continuing and been are and by are interactivity the Moving year-on-year. up Fiscal Lotus Leading Sequential that billion fine-tuning sequentially, seeks quarter was the the revenue creative LLM from car an refreshing including up manufacturing, growth of such data verticals WPP

by successor, AV year was NVIDIA NVIDIA monitoring, is of revenue Motor, year. software-defined first for billion capabilities functionality the offers adoption computer and $X.XX continued next AI billion Wall down year-on-year. parking, time Thor, automakers. on cockpit car Revenue Orin be up customer vision this and available choice $XXX fleet. EV DRIVE AI DRIVE of platform the crossing was for XX%, a DRIVE EV and Great of AI and transformers There ZEEKR, will and driver all quarter. the several Moving into vehicles NVIDIA new for were designed NVIDIA. X% $X the Its the platform, announced announcements performance and more of up to Fiscal range built wide sequentially Geely X% integrates single Automotive. mark on compute passenger including and intelligent of driving premium automotive Xiaomi Li Auto, subsidiary a autonomous AI million,

strong XX% margins of and gross rest margins to the and Center to growth GAAP Moving the Data P&L. non-GAAP gross expanded sequentially mix. on XX.X% to

up favorable dividends. Our investments higher in employee In and up in expenses gross returned cash the margins X% costs. QX benefited infrastructure were compute repurchases of were QX, to $X.X we from growth. component Sequentially, form shareholders and primarily reflecting GAAP operating share operating and expenses X%, non-GAAP billion and

returns, fiscal share 'XX, including utilized in cash During towards $X.X billion year shareholder $X.X of repurchases. we billion

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

and expected XX to in points. non-GAAP minus XX.X% by Gaming. basis GAAP gross component and We QX, benefiting XX%, or growth favorable are and Data to margins margins from decline gross be expect ProViz, sequential offset plus costs. partially respectively, in Similar Center QX seasonal are

non-GAAP margins the gains Beyond expect CFO range. included nonaffiliated mid-XXs in closing, QX, tax GAAP are our XX%, invest non-GAAP of we in continue the non-GAAP billion, are $X.X percent to items. to $X.X X%, $XXX range are respectively. Fiscal mid-XX% and and information expected other on GAAP year, and are be of of GAAP investments. for and year to other or website. In be and me GAAP for commentary IR non-GAAP to to available million, expenses approximately from discrete expected Further operating expected us. operating the be to return grow to the the in billion approximately minus rates income expenses excluding losses upcoming details plus events let as ahead opportunities remainder income excluding some and gross we expected expenses an XXXX community. financial highlight large the are any and financial and the

on poll on Conference the Care annual will the Health for for will would in years. Media and Morgan and Stanley please And in conference TD in us questions? course, questions. off you XXth Boston Telecom in be time Annual for we please day, will GTC in XX March March our Conference the attend to of keynote host Cowen's analysts for the We Monday, March San a Jose, with join person X are and X. first Technology session San March kick held next time, open California, call and X; financial for this XX. At starting GTC Jensen's Francisco Q&A now Operator, we the

first Toshiya Goldman question Your of line from from Instructions] comes [Operator Sachs. the Hari

evolved really the expectations extremely over XX Clearly, you're My you question can for Congratulations Data some hoping the results. how And the the days. like 'XX curious Data calendar answer have your was 'XX in touch doing within Center well the on the software. strong past Center, I'm Jensen you of as is on and things for buckets newer business. on question, business. I

Sovereign several AI, we in medium- you market. the in guys recently, think that vocal that? an should there about about years? how to how NVIDIA been And if any market long about that I term. think so, playing Is And credence think participating you've to to there over ASIC article pretty the potentially next about was

Thanks, Toshiya.

was... There see. Let's more One time? three questions. were First question

OpenAI.

they've your I expectations evolved. Center, for how guess Data

industry-wide. fundamentally, let Okay, calendar calendar know are at industry-wide growth, We're beyond, why. at transitions, of you guide for to Well, them and 'XX and me conditions tell we both the X continued you quarter the are X beginning time. But yes. of a excellent and 'XX

the computing. as for steam. is including centers, out from you own is starting run extending throughput of And by reason you data to general-purpose transition to computing, one X like extending computing, and GPUs used General-purpose There's can't you accelerated years. and fundamentally enhance dramatically know, the with no could update when just from more many its tell to CSPs our a X to first The general depreciation you to.

been NVIDIA to efficiency. you you with so This what have energy computing, has pioneering dramatically And your can is time. accelerate everything. improve And some for accelerated

XX:X, data That And sure enabled it your plenty the AI AI, industry-wide so of called in going In is speed that during we incredible I'm of You huge by second AI. created. call. a is new improve software, we're It transition dramatically software course, to speed. processing can generative about cost generative of talk application. being enabling types It a way is doing a computing. remember, new But is generative way the a new new numbers. of

AI do computing. on generative traditional can't You general-purpose

the turns Midjourney days that. that or supercomputers ChatGPT or A These type data. Center first You AI. AI augmented of new of something the incredibly And basically, whole built, at, sense accelerate generation, it take Center worthwhile very and me is and factories. But Data a it is as into step is now industry. is in a these AI data raw describe a third for sovereign new about tokens It is, not that heard material, the have look We that transforms and the last a and it. And data it to are enabling are and with computing which search company. employees have what back generation these it time, about just industry people this is connects takes to storing it the them amazing tokens. new by whole serving you've to Data on factory, about experience question NVIDIA AI the valuable it your an and AI

it. augmented that, goes your that All are along of by with the hyper-personalization recommender now systems

chemicals the All generating generating of and biology incredible on. digital start-ups proteins list and goes these in

diversity. a these And so are very of Center, specialized and generation in And all We're it Center. type Data seeing of AI this AI call supercomputers Data generated tokens we factories.

amount One, time with of new so the that Every videos interact we our videos that build editing, that. the way diversity markets you into because in or larger, are inference about estimate single being in. we're the do -- of off you continuing out. is into doing that these and foundation the of of the The generated NVIDIA's these time But the business is inference all CSPs Firefly, -- of Runway, it One the also seeing amazing time is Sora inference of other tremendously, models we're The diversifying Midjourney, has is XX%. you you new reasons Almost just that inferencing. The are us inferencing. amount see part you're grown large they're now. know still every are ChatGPT, larger we're getting is that increasing. training we to use inferencing. industries. amount charts Every manifests The continuing

They and in category could from GPU-specialized their CapEx But see CSPs. infrastructure, AI You new a specialize their there's GPU-specialized discussions. NVIDIA whole CSPs. called

seeing AI. ServiceNow You're example. software really really, a deploying platforms great just is enterprise

You the see there's others, and SAP Adobe, others.

services You all augmenting have can to that even see are of Internet past more with now created. consumer their they so the AI, be generative of hyper-personalized services content

You AI. industrial see generative us about talking

belongs they businesses: multibillion-dollar of total, course, The now. And the industries and the different, are own the auto, the for businesses has represent in with multibillion-dollar knowledge, health, region our reason vertical industries financial each do their to history, to them. the data their language, that that provision their own AI material raw AI. are digital fact to sovereign our like to the use of society. culture services, sovereign Now train themselves. own it data, Each their The useful to would of to world, to data. one most They intelligence the is it them. It create data regions belongs with harness around and the

to protect to they the they it want into AI so transform value-added services themselves, themselves. data, provision And transformation and those want

in being is other infrastructure regions. in Japan, Canada, So France, built many we're seeing so in sovereign AI

every are West around expectation surely AI what experienced here in the factories is replicated will in And be And the every every world. United States, industry, being company, in so is generation that my the to going region. these be

doing becoming last AI formed, a a whole new way being really whole space, of year, driving new whole a think and I computing, industry the this is last application so growth. seen we've that's our generative new And --

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

inference. I number XX% revenues wanted you're you maybe how was of -- sense assume It's that? to I us give you inference. year in are bigger than before, solid measuring seeing from the same of number growth a follow where measurement? that are is up Is expected. that LLMs coming Can I much on the around cases, it's, used from And and that some how How training inference? the some GPUs for

The backwards. me is that probably estimated the the music, it, year you -- go let as you estimate I'll videos, understated are and the and to the when X things recommender a that -- has but of why. products but the news, a and know, -- trillions the ago, many how trillions people inches being you there, trillions, I Internet recommended out know your ago, Whenever -- year you run phone don't is we the Internet, systems because, tell are squared.

needs needs the AI path acceleration GPU needs It of systems has something now systems It acceleration acceleration a generative recommender the nearest for fit it to for you. be learning now augmented GPU deep into ability recommender based the so such search. to the generate amazing system information these really put systems. These a GPU the And for all for And embeddings. to information and It CPU acceleration. system, real estate But all called GPU the recent migration them neighbor accelerating to recommender on used for an directly to approaches. needs through small that reranking. GPU of down is

system are world these the I planet. Almost large consumers, So largest ChatGPT, major these, run just and GPUs every use of on single are Getty, XXX% they're mentioned, Whenever systems. hear company is Firefly recommender none from And on. and you of generative know, in of it's the the number every that we software inferenced. Adobe, a to year as the now. goes these And do a generating see Midjourney step in recommender just single Whenever recommender for things ago, all you a has that engine work The you system Getty new. with list as models. when you existed the about being

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

to Why Colette, get constrained that What how constrained? of do -- be do the you generation that? through Hopper I calendar to as be the the wanted expect do easing that all assume the the wanted I Blackwell when generation long to you comments you to products, that to constrained Like driver dig bit? that into I Like Can that those next you supply on touch a is ease? is expected your constrained. of way And up? expect like start does next 'XX? be to little

Yes.

you transceivers of course, The name we and first the is our of wafers, of chain list Everything just is to it, supply us. the our incredible thing ship. networking for is cables, the and the supply improving. regulators from, overall, and components an Overall, doing job packaging, that all memories, power the

things These for GPUs people Hopper reason. we've XX are like NVIDIA know, call that weighs it NVIDIA XX,XXX a you As AI GPU chip, things is good has think but pounds. parts. an complicated the supercomputer People built. It really

the has at mind-boggling. look Center, the ever cabling system back systems, the cabling is the It most you ever system networking is for complex Data dense, If seen. of the the world

Our InfiniBand is business supply grew really year-over-year. supporting us. Xx The fantastic doing chain

overall, the improving. so is And supply

provides, our be year best. going continue supply expect The our do through do the the and to demand than to improving our continue we're stronger to and times are best. we'll cycle We will and

up. it as that a doesn't you you can't large we X know, to It ramps up. from However, products, do overnight. and number, very new have ramped whenever step Everything is

traffic keep way product Spectrum-X. of our a brand-new the for have for routing, the HXXXs, on control, like world there's ramping a no right so isolation Ethernet, incredibly whenever is augmented, ramp. products InfiniBand new And Spectrum-X, top Ethernet Spectrum-X, or Ethernet. Ethernet of It's term generation with into noise we of very layered new short AI-dedicated systems. fundamental so is we system. But doing the We're as congestion Spectrum-X. AI. with we've in are on ramping Ethernet optimize with and scale-out isolation up we could can We're standard demand well we capabilities now, just adaptive we reasonably that not good

be will And AI-dedicated our infrastructure, so and be networking, our will is AI-optimized ramping. Spectrum-X InfiniBand that

the So that's supply we'll products, the just But nature products, and can of catch up than -- we kind demand with to work we overall, supply. very new all our overall, nicely. is greater of as net-net, with fast new increasing is as And demand.

Matt next of Your the question line Ramsay TD Cowen. from from comes

for ask, to your very the to "fairly think which your very, of of industry. product first compete companies across your speak of Colette, getting on? And a wanted turned at that with I Stacy It's improving. I hear good that of interested I be enabling. at of terms your of of about of supply but sort the about significantly demand even you bit of about are, government. a X product the guys being for a across how the be the smaller guess care was in then into and that there's it to you company results. good I thinking speak sort I'd comes monitoring the on interested any a really Congrats allocation guess, buildup more on to you guess, sides a you're of company customer yet the of health what though I'd question is and are things product putting like and than I allocating" hear kind thought arena about also unique really if your to readiness other, be And guess, in deploy question, for you Jensen, to supply for customers, allocation each bit you. you are might to how not X-part industries many technology start-up the And

of what these time both they first providers need allocation in your set are how me up out in but setting Let thanks, for partners they their work and that working many have as start into we they they been are them customers one also us folks their years at the across products serve been Many with multiple what question, building so we have with end look different as going many about we how our well they assisting process. what we are needs have instances internally. that GPU customers The to up with with, cloud customers work as as with our internally. our

working course, getting but want. those are them with So they their outlook them helping clusters and in wave they our they and for not only And that demand getting will architecture their the next interest about the new that Hopper thinking on need. of advance, our understand discussions continue

our the what terms So being built with us is and helped and it's allocation for customers. helped the both their use of that really process end still in has them. with they of always us build purchase, is of relationships a with sophistication will moving the communications what what and But in understanding built we've that

where our with First, road them clear to view of transparency which confidence that CSPs CSPs have and our product And gives place when. and a our and very map transitions. products the of

-- timing to they the know allocate our course, ability, and, can unnecessarily. know and quantities the we of We allocating the to allocation. allocate do they best allocate fairly. We and best of of to avoid fairly, fairly And best so our

As is difficult not you mentioned more Nothing around. earlier, anything when ready? something a sit is Center than Data why have to allocate

is -- our customers. them services asked about and that an and partners, we about And we is that really care question interesting giant CSPs, our agriculture bring companies, there's robotics There allocating end suite robotics to know end bring robotics with you just ODMs, of have excellent markets, are to are the to biology robotics OEMs, robotics, companies, companies, And markets. developers, unnecessarily. model AI bring the avoid so ecosystem developers, robotics robotics very warehouse emerging. autonomous financial companies. allocate customers, importantly, that companies large of where health OEMs, that we NVIDIA a and companies, unique companies, language we we What fairly CSPs kinds really vehicle The all do humanoid the companies, you surgical

them to same -- and can have companies, health by working and the we start-ups the bringing We directly. a large CSP services, a time. on And twofer care, the allocating oftentimes, customer at these of NVIDIA's CSP financial are and platform. such to auto support All

with those waste looking ecosystem, end it's avoiding for the that And for fairly you're it, But We're core at opportunities vibrant. time. and so right to absolutely we opportunities this allocate all want partners users. of and connect the looking to

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

plus Obviously, you're of purchase of of didn't touch. that I lower? lead into and inventory to the suppliers you maybe sort state it the the about the closer I of talk as your you of is supply, revenue. inventory up because add commitments, have your come just backlog read backlog? order that a was your down your your some converting reaching you're you How actually how should wanted we and to your products commits ask lead purchase if down supply, sort saying steady that prepaid times filling for don't to about Colette, are a Or financial book quite but aggregate sort times need you're your that? commitment a Is where much take to bit.

Yes.

of highlight on different areas those three look So our me how let we suppliers. at

You're correct.

Our allocation our trying come we're things hand, work customers. our inventory, to, to given into to inventory immediately that them as we're on, ship on

customer to looked ability I the for. our meet think we've that appreciates schedules our

our The commitments. of is second purchase it piece

different. the also capacity many that have all different components of for it, into we that are manufacturing or Our need capacity component length the need procuring The but need. purchase we we for components are length often of commitments that

be be regarding X may the prepaids. for of of same multiple some next but say can them may years. Some them for quarters the I our

as suppliers capacity prepaids Our the forward. manufacturing several that as are our we need we to that make predesigned look have of we reserve sure

our wouldn't supply. increasing read numbers the about regarding we So anything are into same as approximately

that All lengths lead them things just in sometimes to we to things have us. different need buy for times or of had as built long have a be capacity

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

gross to And also, And don't results. Congratulations if you of products? the about I margins talk new it the comment in think mind HBM unpacking to that do that. on the are Colette, regarding and drivers back go to mid-XXs, that you your wanted the content should comment? the is what due that

some opening of quarters those QX quarters QX. in about thanks for for Both question. our outlook the are chain unique results include in really gross of margin highlighted our networking, and our remarks, both benefit different manufacturing Yes, supply kind are compute as unique. favorable of Those We component and the in X their our and from across our they process. stages cost also several in

back we gross have taking where QX before a and mid-XXs of forward, visibility here. were looking to that the fiscal this rest So for we the peak year, had us margin we've into QX

So really drivers. we're be really driver our and the what our of the we for looking Mix at those going of always will just year is are of the rest be just a largest mix. balance to shipping

Your next Cantor question Fitzgerald. C.J. of the comes from line Muse from

Bigger inference playing clusters you, do do in picture decade similar their tomorrow's today? GPU they're How you the about compute the investments over improvements next, today's clusters? you how for and long-term question about training become out? the usability see in think making customers that million-x your for improvement Jensen. last When think expectations NVIDIA of this the Do

the C.J., really thanks cool question. part. that's the for Yes,

two able architecture to to revealed that it's we've two, Geoff way it's It's literally transformers, One stack only every every performance And so we along, single at the every reason Alex have Ilya deep transformers that improve is through AlexNet, and into has our optimize deploy programmable. the it because and and now transformers and to beginning characteristics brittle. Krizhevsky Hinton when species able amazing to RLs from it, installed gone LSTMs sequence support Sutskever very and is If RNNs AI platform. first look and every base. This of our very, RLs to it's single single that much, beginning, our the that the species and every have it variation, to at you every very single NVIDIA for why the great all is accelerated, vision multi-modality about the not we're that has -- been come part. along, to really version come time stuff. CNNs

formats invent can Cores, technologies at and Tensor improve of Tensor Cores, one generations processing Transformer architectures and meanwhile like the same of base the installed like supporting new new our hand, numerical the time. structures On done Cores, new like different we Engine for the with we've Tensor our

hand. inventions, industry, on of it of as take so installed algorithm runs a inventions result, models the and And the our new new of we inventions, one on software all all of our base the

Transformers, Transformer revolutionary, the brand implement like we something hand, other Engine the future. can Hopper it and new whenever into see On like we something create

it base bring to the we better better. ability And installed have so and making keep simultaneously to this better software and and

So our our customers' with installed base time software. is new over enriched

large made breakthroughs, will new our all for Don't capabilities. hand, revolutionary CUDA, technologies, they made in installed will be be to other language And possible. run of surprised models base. the in were available breakthroughs a be if because On the which future those sudden, generation, some create of amazing in software

carry us giant hand. we other the so on one with hand, And the everybody on we make breakthroughs

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

started that I ask solutions that know China. you you wanted the alternative to some in comments, China shipping prepared I business. the about said into

to you out also that percent to mid-single-digit expect Data that Center a of contribution continue your total business. pointed about You be

that the China to solutions products guess come expand the opportunity not market participate of is, into question we why market? extent I should is that So alternative the that today again? your And expect maybe the and to what breadth in shipping you're other

will, two wants if the possible. And as be China like U.S. remember, capabilities core, those to successful of Within AI you pillars, are two the At restrictions. U.S. computing accelerated within would the us limit in government latest Chinese the NVIDIA's see as and constraints, the to government the those to market.

in so that that in is way We some had to any reconfigured so And paused, came way. hackable restrictions our products the the And new a that pause we understood software out. we when immediately restrictions not took are, what time.

in now China. And to And to of to and best that reset customers do the our product we're the in China, so to we're marketplace and specifications marketplace succeed restriction. within in that offering going our compete we sampling

the in as shipping -- in We the marketplace. marketplace. paused it. we This our stopped last business And so significantly that's we quarter, declined

and We see after we same. it do we'll our can quarter to business compete be and out. hopefully, this turns that, expect the But for about our best, go how

comes Sandler. question Piper Harsh from Kumar line the next Your from of

so little for could bit of software us help words, us a and all, But pleasing business. NVIDIA where and coming Colette the bit the guide. understand that or unpack software quarter if about can congratulations billion. parts to are pieces Jensen it it's growth In a and $X I what a and talk wanted stunning on Colette, that is business, better to your we it's I Jensen, just first a team, you other over understanding help about hoping, hear -- was different of from. little get just

fundamental Let why a successful in me reason explain software. the and NVIDIA step back very take will be

and working fix, accelerated And the complicated our So them, any cloud the teams, and first, of know, with their we teams operate in we the and way have software as assigned large are cloud, we engineering involved in you large In computing. own issues, to are really teams, there work engineering cloud. providers we that's whenever manage patch maintain, directly in really a stack service teams accelerated with to them that computing the we we business. grew allows engineering their have enhance, them and

you accelerated computing. As different very computing than know, is general-purpose

You're C++. not starting from a program like

The to CPUs. PDF, to images structured models, SQL for all versus compile text and which to unstructured, large processing, and -- stacks every it computer data speech software the your systems. all You and domain is learning on data SQL vision recommender to classical language run things from versus of necessary machine all

NVIDIA has All things libraries. why of different That's the reason these stacks. of require software hundreds

software, markets. If new can't have you open don't you

you software, have new can't don't you and enable open applications. If

we long people easy. that's understand computing accelerated difference general-purpose between to Software work really computing that CSPs, a really fundamental with understand. is way accelerated is time that This most computing. And fundamentally and key. And that necessary for the now took people is the software

have We their are teams teams. with large that working large

and across all engineering of throughput computing optimize their companies software enterprise don't to just to generative embrace possible, is clouds longer enterprise of now accelerated large and when embrace computing, no through improved private essential clouds longer every However, it likely because all every computing, it to be now these is sustain that and stack general-purpose accelerated AI world's and enterprise enabling no the enterprise is able teams on-prem. to companies anyhow, run to and to have and software company software maintain

So year. time on an with the the we we our tuning, every AI as Enterprise. it software GPUs. like NVIDIA on-prem And for to the operating of management, way And we Enterprise software optimization, into per private our GPU an a enterprise market run are going per go think $X,XXX system. stacks. all in in installed containerize software do charge And base It's optimization clouds and them every Enterprise, the clouds system guess the operating we the called AI the especially patching, that all their will the for now world, And obviously, of for run my artificial stack to NVIDIA and that company is are is NVIDIA intelligence. enterprise that deploying AI

and is start. really going to likely already very off getting rate this a be mentioned Colette great a we're that time. started. so at run And $X billion significant to it's over just business And We're

Huang, turn over back will I to Jensen now the closing for call CEO, remarks.

generative accelerated and human center energy. increasing data world. the enabled incredible the NVIDIA of The structure the base AI, understand computing demand keep trillion-dollar biology a of will transitioning simultaneous two the data The whole information with to computer at industry paradigm, is accelerated to centers cost The speed-up platform learn, and installed any making with Every and same software so where of can purpose can shifts from throughput general up computing -- language is generate that enabled time. managing new the XD from world be NVIDIA while computing.

investment it of And last intelligence. are of factories is next two of generation built this AI. Generative into trends drive a beginning believe cycle power annual proprietary fundamentally with generation dollars AI infrastructure massive will open supercomputers AI We a trillion AI speed, world a the from world of and centers applications next factories. center industry X up new new not journey years generative doubling AI base and industry of company the every kicked will providers Internet billions. This revolution. future, of represent possible is cloud off of new and board, hyperscale consumer opportunity are on We services on data NVIDIA and every whole to developer to automotive generation to scale, now new stack their to proprietary raw gives business Like at Every the telecom, the process AC application the essentially plants of the intelligence, now companies started become to has and data industrial industry-specific will their today. and an installed revolution, every infrastructure these in AI in companies. AI market an build health a the hundreds AI-dedicated whole refine with help and company. digital We every the industrial the us industrial media infrastructure where industry in the a platform reach financial to data frameworks AI industry help care in ecosystem to huge full entertainment. and partner computing to company, NVIDIA's

much look at We you forward on next month's so share sure next so We progress you to us. in have join updating to GTC San to our with be quarter. Jose,

conference today's concludes This call.

disconnect. may You now