NVIDIA (NVDA) 28 Aug 242025 Q2 Earnings call transcript

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

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

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

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

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

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

financial the Let for an community. me event upcoming highlight

XX in chat. where be Goldman fireside Technology Francisco, We on Sachs San the attending September will and will a participate Communacopia Jensen keynote in Conference

With Colette. me the of Our XX, scheduled quarter Wednesday, of let earnings third fiscal turn our XXXX. is results to for call over discuss XXXX November call the to that,

of Data well and X.Xx. XXX% GPU platforms. Networking computing Data revenue than year-on-year XX% quarter. grew by data the while $XX architecture adopt enterprise to was for more and year-on-year, than roughly companies. Customers Center Thanks, up revenue accelerate stemmed billion gearing Center and strong driving our Stewart. revenue, billion Hopper last a their more Blackwell. Center. continue Compute represented and Internet up $XX.X than from our purchases year. providers XX% workloads consumer Key Starting Data to of from up and Center XX% as and systems; AI with recommender growth and and outlook more record, driven of include synthetic generative post Xx training demand workloads; video, up networking service pre database our Vector was processing text sequentially Revenue and our well. sequentially Data billion. $XX the inferencing; was and of image, AI with up and generation; revenue XX% record model processing QX grew XXX% above NVIDIA Hopper, data Cloud CUDA AI-powered another SQL

compute Next to will more generation train is with require XXx models to trend continue. data. The expected XX more significantly to

strength enterprise over HXXX. the and and bandwidth of demand. NVIDIA that we Data from incredible robotics. adoption compared CSPs, XX% capacity, shipping inference X efficiency estimate Center trailing the more of broad throughput than ecosystem enterprises quarters, benefit began enterprise HXXX platform. building Internet education, consumer revenue. Over cloud. of Developers availability in given our thousands the HXXX Demand model care, XX% inference drove NVIDIA's services, companies, of rich offering are The and applications and and desire and memory from in makers, growing their start-ups of Internet every Hopper is Internet CSPs, company. CSPs more ramping and the consumer appreciate tens and the and NVIDIA to AI large NVIDIA's for QX, health and architecture consumers, frontier for NVIDIA the our of builds consumer generative coming upon to NVIDIA NVIDIA high platform companies advertising,

revenue. sequentially contributor to Data our grew a and QX Our China Center in Center in Data significant revenue

prior seen total of percentage levels of export the to remains revenue, below a As imposition Data controls. Center it

We leadership all round inference medals Hopper latest manufacturers, array highlighted architecture-powered both of China to Computex, top AI NVIDIA, inference and NVIDIA's NVIDIA MLPerf the for tasks. be factories continue an and At the benchmarks forward. The building gold to to NVIDIA with on win going computer with systems market competitive of Blackwell unveiled very centers. networking expect Blackwell platform data combining and

expect and NVLXX dollars revenue. the brings mass building shipments is and next Hopper is and XXXX. ramp quickly Blackwell Link MGX reference production XXX time. models NVIDIA half well networking together chips, systems the modular DPU, QX, With and to production are OEMs and are multiple NVIDIA LLM's with is to and Hopper a power for CUDA demand expected billion XX% above fiscal increase than workloads continue to to countries. in Blackwell single the the revenue systems supply sampling. software availability and 'XX. In trillion-parameter and faster The to the partners into Blackwell enables Blackwell ability GPU, several of and for and change a platforms XXx to Hopper sequentially. system NVIDIA effectively. The run up second platform across supply, year. Networking inference and designed scheduled into architecture, increased have begin ODM NVLink all improved. fifth-generation real and our next cost CPU, in fourth this generation XX GPU deliver yields. more the Blackwell of quarter expect Blackwell-based the NVLink in strong to Switch unlocking GPU cases, in the we to NVIDIA We continue GPUs widely as the we Demand industries, improve AI year Blackwell executed GBXXX to to act fiscal is

well sequentially revenue, demand xAI hundreds products GPU traditional line OEM adopted AI launch our from market is performance Spectrum-X includes offerings. plan thousands for with of connect compute on end-to-end Spectrum-X of supercharges customers of the the We partners which Ethernet. is adopting GPUs and today ODM Ethernet cloud scaling support year. world. the by doubled in providers our X.Xx begin enterprises, DPUs compute and tens track Ethernet has cluster clusters Spectrum-X the from support GPU for Spectrum-X multibillion-dollar of to Spectrum-X AI broad platform including within product a new Ethernet processing of to and to CSPs, near largest a for Ethernet millions in and being to year delivers to future. every Our

X.X supercomputer drove NVIDIA. Enterprises Our sovereign national Technology year. The sequential Industrial AI double-digit believe sovereign also AI will AI has its reach opportunities quarter. Bridging in and growth this Japan's continue We Infrastructure started. Cloud to AI revenue wave infrastructure billions low for revenue with Advanced and enterprise society the at countries imperatives industries. of expand as AI building Science and is expertise their National Institute recognize

its including on-prem for across imaging, centers. to working vehicle NVIDIA lower will working is serves as autonomous Now a build fastest-growing a dual billion and quarter queries productivity. Amdocs to Automotive copilots. and NVIDIA enterprise industries customers most chatbots, copilots in on multibillion-dollar on grow reducing and product We significantly for XX%. Automotive cycle with A AI generative revolutionizes lastly, experience offering, cloud to data NVIDIA customer Health every drug build costs. is robots, AI customer their company's applications NVIDIA will automaker fueling new to driver is in XXX AI technology end-to-end to multibillion being And business and using NVIDIA the with revenue reduce copilots. SAP their care AI as AI the range are next-generation is the companies models agents over is history. XX,XXX and and more is Fortune the our build developing using service a for AI also AI key AV enhance of monetizable Omniverse is times using factories processing, growth, agent, using NVIDIA initiatives costs agent X are [ record geographies. dollars the growth new transforming electronic employee as day for in patient generative by their their business over health drive and and build generative AI-powered across to was for Snowflake compute. generative ServiceNow care, medical surgical require Cohesity smart way consumption its discovery. is applications NVIDIA of XX%. ] Assist development using by using

financial across telecommunications XX% AI, Uber. adopting services, generative is develop are moving applications. first health across the AI for and NVIDIA embedding transportation, AT&T retail, model and Xx automate the first deployment. quarter, X.X generative call Meta's model Lowes, customized simplify the energy, and the clients seeking both NIMs enterprise NIM a the AI a to and X.X NVIDIA for business. Accenture software generative models. of new With announced new collection adopt service creating leverage of deploy assist to flywheel. applications. Companies applications This and source foundry AI Agent layer knowledge AI of NIM every custom own applications AI deploying full to their include data-driven time, a into AI customizable institutional that for applications accelerate Llama XXX to accelerate encode service including into generative cost use and building Companies ecosystem. and of AI to transcription build after and their latency can enterprises announced to refine the We flywheel for with models frontier-level its NIMs Llama marks for catalog of can savings AI partners reduction an Blueprint, are Blueprint, enterprises their Agent to capabilities a classification. Over NIMs reference AI. world's Aramco, open supercharge During time we enterprise watershed care, AI a over realized NIMs, the to for moment an suite

Agent retrieval augmented NIM enterprise discovery, first computer-aided customer for drug The include generation. Blueprints workloads service, and

software providers, NIM builders bringing to Our platform, the Enterprise Agent system Agent NVIDIA AI are and enterprises. NVIDIA Blueprints through which solution available integrators, are has NIM NIM Blueprints technology and momentum. NVIDIA great system

billion and We rate notably with year, Enterprise approach software, SaaS AI $X to a support exiting contributing our NVIDIA run annual to revenue this expect growth.

to gaming Gaming revenue remains now devices, tops channel are sequentially Every and and notebook, console, to XX% growing PC RTX saw PCs $X.XX of RTX can applications is Recently XXX model, Moving and AI. NVIDIA to is strong PC. game the desktop laptops Awakening an catalog demand RTX XXX is games on designed generative a from an AI total in with most The our and continues X% AI-powered revolutionize set size and Jones up small and Circle, service. XXX base to GeForce With leading language content AI grow. over deliver and of with with PC ACE, We year-on-year. AI NVIDIA DLSS the technologies RTX consumer million any Megabreak growth ecosystem and installed and added revenue, to AI manufacturers. The PCs. increased and inventory over billion X,XXX RTX healthy. gaming sequential and device and PC. to experiences include of Great NVIDIA Dragon Vanguard. titles, expand continues available including of Age: optimized use cloud NOW The titles generative Indiana of The AI library suite for gaming X,XXX ACE, RTX XB is inference. Nemotron first AI is

drive Omniverse $XXX is that power verticals microservices to AI to and digital across to new manufacturer, to produce NIM digital is and Demand the to of the open the visualization. highly NVIDIA copilots workflows Automotive ability of systems. And growth NVIDIA signed our for up industry Companies Moving into industrial of incorporate content Omniverse quarter. enable million announced largest workloads. global driving Foxconn, graphic physical electronics use and world's implementing manufacturing new as model build pro generative operations. this virtual plants X% workloads, Blackwell to creation USD in NIMs developers key We using contracts to and for their and are by Company. and The to WPP is accurate worlds. digitalize customers build USD its including such fine-tuning several The XX% NVIDIA racing industries AI-enabled efficiency to large twins Coca-Cola were generative Omniverse-related Revenue Omniverse year-on-year. USD connectors accelerating driven and including among Cloud AI sequentially enterprises, being agents was twins NVIDIA Mercedes-Benz factories. multiyear pipeline cases,

autonomous category, entries X% Boston the year-on-year. the ramp customer robots. up at robotics Intrinsyc, and in and Challenge AI Year-on-year automotive and than the to new up Siemens, driving cockpit At growth solutions. more humanoids BYD Conference, increased demand NVIDIA was for robotics. million, -- for and platforms the was by Pattern Revenue Figure, robot and and sequentially driven Dynamics, upscale Recognition Vision Computer XX% are self-driving the $XXX Moving using end-to-end NVIDIA Robotics in won Teradyne Autonomous XXX Brand Electronics, arms, worldwide. Isaac platform the outperforming mobile consumer

Blackwell XX.X%, down were non-GAAP and products low-yielding operating margins provisions and rest inventory non-GAAP XX%, gross toward was we margins form up compensation-related Sequentially, in the mix gross in due GAAP to GAAP expenses share to dividend new Data of QX, the higher sequentially Cash for costs. P&L. cash moving shareholder of XX.X% Now of were billion. material. shareholder. increase were and Center within higher operations per utilized dividends, primarily the returns cash repurchases In and flow from $XX.X reflecting of reflecting $X.X billion a the

Board billion the billion remaining end add $X.X share a $XX Our authorization Directors to approved authorization our at of recently of to of QX. repurchase

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

quarter growth our Blackwell products. of and Hopper third outlook sampling continued revenue of incorporates Our architecture our

GAAP XX.X% respectively, production and to and in or XX XX%, points. ramp minus Blackwell gross be expect QX. plus We expected non-GAAP are margins basis

trend continue shift mix products, the to continues into we XXXX. this to fiscal of expect Center Data As our new quarter fourth to

available expected other are financial securities. GAAP items. nonaffiliated including full GAAP and respectively. to developing be billion mid-XX% website. and are work equity X%, excluding tax to operating XX%, expenses $XXX million, are and next and operating held discrete our Full products. we XX% in approximately mid- CFO to Further and year, from year be gains publicly the in expected and on commentary to any margins IR minus GAAP expect non-GAAP included in losses our range. are are income expenses rates to about upper be to other the of or range For non-GAAP details grow generation the plus information we and $X.X gross as expenses non-GAAP expected the be on $X.X and billion, investments expected

We you are questions. now questions? going Operator, please for to open the for call would us help poll

the question Arya of Vivek your And Instructions] line first from [Operator America Bank comes of with Securities.

in suggested packaging are be that what Blackwell I'm the comments QX mentioned because Just GPU back-end there's of or revenue dollars of will solved you you I to other what that could think mask. several ship anything else? by incremental Is to it change in in prepared any Blackwell changes change kind and the in in of in that Blackwell design. are issues a it. there you any Jensen, your the these reacting means overall then? the changes curious, billion us size impact the all And is profile despite how customers help related, timing,

functional complete. The changes is Yes. change Thanks, mask Vivek. to the necessary. There were no

types sampling And something that built is a those. and it The sampling production we so is, to enabling are of Blackwell-based we of ecosystem and variety functional start configurations systems we're start Blackwell, samples at as Blackwell, of to There Grace as were system functionality we're speak. Computex, are of in like expect Blackwell and XXX QX. shown that our different

line Goldman the your with And Toshiya question next Hari from comes Sachs. of

Jensen, question. I longer-term had a relatively

guys going customers' NVIDIA, debate of As and that CapEx gauge maybe forward. you customers your a a customers What's sustainability that and in means as your how follow-up like dashboard may for to know, watching? CapEx? And Colette. return heated there's market customer at Internally return pretty quick investment the on and what on then the what you try impacts are for you on

think AI about I up improved What's should full the and a your we for year how of couple 'XX? maybe think driving number sovereign the went fiscal outlook billion.

Toshiya. Thanks,

mean out, but production all, starting starting mean QX, up. mean of I shipping production I don't -- mean in to I don't said when shipping First I I ship ship, but

you've On at time. platform me a back. transitions let's say through the step same that the we're going X question, longer-term take And heard simultaneous

reason to purpose slowing from be And first has to to accelerated of general accelerated and to for a time And grow the continues crawl. CPU slowed to the computing is has it known scaling -- significantly. that yet for transitioning been demand because some computing. X computing The is computing from amount quite

it estimate year. single even maybe doubling could be every You to

driving it And the up new would inflation so and be computing if don't be every consumption cost data around have the approach, world. company, a of for energy centers we the up would driving

seeing fact, In you're that.

it's at to and bunch application an course, blog see but computing. directly new that computing this you computing about the enables of whole for much of do accelerated lower answer their XX% for is, that's the of simulations that processing, someone what from is cost or libraries a accelerated just We database of accelerated platform out also really the there's save larger first so know not a fact, talked energy in reason that consumed. to lower applications. the transition, up of And computing, we And you that sped going computing. It a that translates purpose is week, to And XXx. up cost. And offer. And example, computing came core unusual speeds that scientific course, scale, to general

expect quite would You significantly. decline to computing the cost

the by let to model because representation on cost learn so second enabled to to reasoning, and gigantic is which the pretrain incredibly drove down codify and language accelerated how knowledge models, and large possible generative train just computing it or about The training of corpus world's training figure multitrillion-parameter have how it -- the we out the into was now AI its learning knowledge models and revolution. models language neural scale understand caused human to how on and go that it so networks deep

It's the for the back now then a doing it just it universal is about function. is is human-engineered model, a deeply Now our a software. just tell because new it is, of why of Instead fundamental function we the went to not observations? answers. so generative We and approximator AI are tell it's figure feature, expected capability. learns AI, we we not step computer have data. of it's AI, bit the -- the the taking a way a what universal the it that learns function. we what into the a out what's And previous algorithm tell It are What algorithms,

could of that the have that's almost predictable, of. have function anything has that And previous examples you so learn anything. you anything And structure, that you anything

that fundamental now It's in CPU to generative a human-engineered develop done here the It's are produce is generative and form there applications are And type and happening things algorithms of layer remarkable. And computing you so GPU, are algorithms, from to computer of every now new we from AI. with fundamentally could AI. of several machine-learned how science. is affecting

So is And in seeing scale. benefits they're still to models the frontier go thing model, substantial it. the happening we're And -- to the are size scaling. of more the all a the data set you size than double of also have growing of train quite first double that's still seeing the whenever you

that in And so the flops create amount model of quadratically. order to up necessary goes

And so next-generation XXx to XXx, take more not that models unexpected see it's last the than could generation. XXx, compute

significantly we the we drive so generational So to energy to to performance drive continue have drive consumed quite down the cost can up and the necessary it. down do

the so makers there year. frontier surprisingly, on first trained And modalities. last than model X there models And are more are frontier is larger more

That's you more. more of more generative the on And have one so AI. on on dynamics going in

although the the today, see But now. AI. We the world about those and here of largest quite things are we of moving from image the in you've systems are kind NVIDIA iceberg. the designers for to of heard are extensively ChatGPT It's which We, the moving have see generators. what now iceberg, iceberg tip and computing which CPUs. digital generative coding. like recommender that. -- are We me The generative systems use below talk CPUs of second tip largest of the What's at this from AI past, in systems, is the lot course, it's a below are coding

quite generation, search very ads and these systems, and very ad to recommender generative hyper-targeting, ad AI. have So generation large targeting at user-generated now content, applications evolved all are custom scale large-scale

start-ups Of our cloud of generative cloud for billions course, of is opportunities dollars the AI partners. tens number renting of of generating

they realizing they could to have and mentioned AI national earlier, so that us now leading their countries digital are Colette incredibly AI, their companies to Enterprise to of companies company. And out AI world's and that resource AI transformation general you their natural to might talking to robotics. just watching generation enterprises. companies sovereign we Omniverse, about, able their infrastructure build synthetic then the now that And we're are use data have our seen more human productivity so Enterprise starting, physical world's about The are own excited AI, robotics as and robotics. from to is intelligence. have last is from and big NVIDIA that so the start platform to, AI, year are joining able IT just them take announcement that the with many now of work the thinking The like as learn video demonstration every to start from drive systems we reinforcement learning are and building general data now own

many that you AI generative can directions so are different there see And so that just going. is

And the so seeing generative accelerating. momentum AI we're actually of

is answer of our in world regarding surfaced have in own and a their something their language, generative incorporate would desires own opportunity, own to incorporate growing country. certainly growth, the sovereign And AI with to revenue, that be their their question the culture, and able data incorporate own AI your around terms countries generative Toshiya, that AI it and unique that terms goals of to in of

can models more countries. for these around So more specific what and be those excitement and they

are yes, us. some So we growth in front of opportunity seeing

question from of And Joe line Morgan your comes Stanley. the next with Moore

being press it also seems But incredible. the I demand talked October. Hopper guiding anticipation without a you strong. strong in Blackwell quarter you're release, very like mean, in about is Blackwell for Jensen, really

see you like. coexisting sense transition that Blackwell do of the you talk intermixing long sort Just to of demand Blackwell? think strong And of So can most the activities, new you see for Do both? clusters? clusters? people about looks transition what you Do how some

of the Hopper Thanks, demand Joe. a couple it's strong. is There's And for for The true, reasons demand for Blackwell really incredible. Yes. that. is

world. And cool the amount at providers, being of for NVIDIA's background. the it's toolkits capacity none. ones either only data the that are have the the on And the in and data for These basically most really GPU available, planet processing, is deployed data popular reason from -- reason in the data, GPUs accelerators the because if only look they're computing mundane. out one, SQL you they really it Panda's their first own in world Panda's the about we It's accelerate processing cloud words. world's new not example. Data ever like out internally accelerating is data to process it running talk And the workloads, service boosting platforms very and almost aside is the CPUs, The that. which just science every data. processes of because of a way as generate that before, Polar's. generate hardly for get are I've mentioned picture or company processing, single NVIDIA's doesn't because accelerated performance it's steam, data, the But

migration And one of primary along so that's that accelerated case number use X to -- another before computing. generative the applications AI is after is the long #X the

vast renting model capital The capacity create They're second could companies. is, generative a company an of AI start-up course, so of they're their the to the products. help renting use that spends And to makers, infrastructure to invested into AI them majority they rentals. it

it to you can't just And so put it companies simply now. raise to -- they now. afford money, just want need use these They you

You you have do. have that processing to

do got you You year, next to it it today. can't do

reason. And that's so X there's a -- fair

is now race plateau. right to the of demand the Hopper for reason because next second The

introduce revolutionary the The to level of to get AI. next first a be gets -- plateau person to

better The person there same. second the incrementally gets or who about is

and and make AI be and SoCs that the be we we And doing so to I the want race establish the the factories mean, there leadership. NVIDIA is to best. drive consistently the we pace. And world to ourselves we We we we ability that make, hard. constantly world's that's set consistently the that to plateau that, the the is why reason systems networking so to first want the make, we and and next systematically GPUs how show one the the you create.

benefits also to come that we we that society, to all our can see dreams all course, of want and to future and in that Of the see true. want bring we true the we the imagine come capabilities

the makers same. And these model are so

Of probably to up start And want billions so weeks or first. the still course, world's of end and will shipping a They of year, the dollars Blackwell out away. want in to this month the although they be world's capacity standing of at best. the be is the

then generative so AI now lot And between and dynamic. of a is market

And so It's Hopper. everybody more reasons need they just really need even is computing either any want hurry. They in it. general don't and infrastructure build a computing. They purpose that accelerated to operational

art. now, decision HXXX is choice right for course, Of Hopper now or you for if the state infrastructure a clear. is of building Hopper, have between CPU right business infrastructure relatively business that

think and just I the people And so infrastructure the modern a Hopper's state to of established clamoring are transition art. $X trillion of installed infrastructure to

from Cowen. the TD Matt Ramsay line comes next your with question of And

and that earlier on see that and having you all to the towards kind frontier back how priorities And of make question ground people really, focused seeing company spending today regardless folks of ROI of of on sort sense. in having push are this for a their the their the are But on as AGI convergence for that that wanted what this you're of and, the for to this the money if investors will sense level that they're to and you customers just in a like Jensen, said, I'm don't spend with so going really is looking CapEx all are debate industry get up percentage I'm that are ROI. I question many capability, the the doors hopefully, are and capability that to get are priorities to the the and new dollars and because to frames their just of what about this of versus opens circle versus that time CapEx. know an it of plateau to makes investment. discussions trying new distinction sense. distinction putting some about technology about

on investment computing Thanks, people in infrastructure infrastructure, the you right are away. The returns today. NVIDIA best Matt. are ROI It's who can make infrastructure getting investing it

And go so to most way is X way the just it back to to think the through first think through -- probably it, easiest principles.

up, You of than more not? And is, And probably to have purpose CPU-based $X do that you or that every you trillion worth infrastructure want Juniper $X for for of rent less build question the infrastructure. it billion stand worth computing billion. $X of general you

are the this the they That's to infrastructure, get return getting and because already And why it's because on is What's processing out infrastructure, investment. point processing money. soon, it more? one, And on of they and saving ground. build a saves commoditized, so data the they there's And part is clamoring money, money of start when people trillion saving giant reason infrastructure the who $X Hopper-based just probably start tremendous already. Blackwell-based data so

so And so recommender and And you start saving so money, forth, money. on so systems okay? save

up The going second to to get you thing stand generative so AI. is are because create companies everything being rented are founded many

services next And to for next-generation third frontier also AI or away from of on user-generated services, content, your system a for system? So yourself is for own good. stores, really services, return own you your a your is reason And either next-generation ROI. recommender create for own the search fast platforms, Do the so own a the a and investment media your next-generation system rented own or is generative ad your social your your right own capacity Internet want benefit that gets or then business.

The the core, infrastructure ways through you is lot world it. moving there's it's shifting But you to the so software computing. computing software. could The accelerated because And world a of think is AI it put general generative ground purpose the of at in human-engineered today. to could best is computing of

to your you it NVIDIA. your modernize infrastructure If computing build centers, do with the to to were build way it. and cloud accelerated data best That's

your line the of UBS. And question Timothy next comes Arcuri from with

quite did year. the you I your and at both OpEx bullish. purchase term. a know shape in the revenue that's commitments look also the Colette, I the longer growth, near increase and increase if And of had question supply your on obligations, for I

obviously can next of that to great But supply you I speak some to not sort that be recognize or when other year some wondering to if school is a there's chain hand, I'm on any of models ramp? And say, 'XX, which for many start is you can be you beyond to thought the and then guess just that and do about seem I air power some customers get of other something shape is racks these the ready point, that. liquid look smaller? Blackwell going really how the you do year, that Jensen, On consider into to factors of cooled. like, look at gating going cooling, worry

I'm going the to work backwards. I really Tim. question, appreciate

in So very can are clear: remember, of is of builds workloads be the centers computing down data the And that worth we purpose And computing from data to so accelerate future, no have centers, for need to that computing, will GPUs $X center we the worth data of trillion world reason about do world centers. cost trillion to when computing. don't In the all general we computing a computing. every sustainable, more because years drive just is experience accelerated continue we few in the to CPUs. accelerated moving continue so that data the $X be inflation. will single In past, GPUs.

we that transformative to going is Second, generative AI we acknowledge the to called a could computing. new GPUs model be of all for future need quite computing

infrastructure accelerated. With than to be world's will $X clearly I that so the think and working is be the trillion the it And $X think respect backwards, the different way vastly to trillion, will next about we last Blackwell. Blackwell you was form our the I factor we the Volta. that think that will, uses either of if Blackwell classic, And it in Volta. multiple pioneered a offer with of HGX comes configurations of shape ramp,

And cooled. so The Grace we've been factor shipping is It HGX for cooled. form air is Blackwell some liquid the time.

center, data any go X the and center, choose, However, reason could past. in data throughput significant. that compared data that center centers -- can, is of cooled cooled install from because you the whatever we anywhere to in you the number center a liquid to AI quite data deploy to And size liquid for the want data to power-limited is Xx

allows XXX which to allows better, of us GPUs. liquid to essentially Blackwell XX cheaper. have which And cooling, so has benefit and our NVLink, we the expand you is TCO liquid call packages, capability cooling to cooling this it Liquid is Grace

connected is And to GPUs going is XXX click showing low imagine of large that. NVLink very that, so for latency, And you game benefits changer that. obviously NVLink. language in the large next increasingly we're And the throughput be the high very inference, domain a model and

both. people I very And are deploying comfortable so think

we're some are both. almost so deploying with every And CSP of working

that it fine. I'm just ramp confident so And we'll pretty up

yes, is Your year looking be is great second the next forward, third of a question to going year. out that

mind business to backwards is Blackwell We grow the game as changer the Blackwell Data next that to carry same the be is to Center principles, software our going and shifting complete software. computing computing, generative to just following going remember to platform mentioned I or for from purpose is industry. that's a into intelligence-learned working transition computing And important expect -- year. keep significantly accelerated human-engineered is quite really, general going really to on X time. focused And earlier, AI first to your year. through transitions at is your head artificial And on, going which is

Okay.

Rasgon the with question of from Bernstein Research. your And line next Stacy comes

for I have X short questions Colette.

Is QX, billion mid-XXs want where question to If draw adding in you dollars? And say, expected first have Blackwell the margins. year, top second that on of billion of several I [ said additive? if for on that Blackwell let's well gross have strengthen I up. of gross The strengthens I hopefully, I the several to revenue half. to mean I'd of the QX, drivers the range. should And something and that for you're that the for how and that And guess reserves about the think be ] as in XX XX year, mean, we XX Blackwell to rate gross inventory the expecting? yields ramps? margin exit next everything like somewhere kind margins for QX as QX Hopper into in second demand evolution I Does dollars come Hopper is You the year that,

Yes.

So Stacy, about you had and take Hopper your first that let's question Blackwell.

continue to our we Hopper second So into grow half. the believe will

believe many ramp QX. will including we to and QX new existing products our products the start have products Hopper, those in that continuing for We Hopper quarters, moving for our to next new

opportunity Hopper for for So let's that. there HX say, a versus growth is

and of Blackwell we top in have QX. the Additionally, Blackwell starting the that, on ramping

that those helps you I So X on hope pieces.

our Your the margin second provided piece is in provided all terms a at of gross with for our XX. We gross We'll different QX. on our work transitions margin We about non-GAAP gross margin. XX do but provided we're in We do the in we can approximately XX. through, we still also on track QX. the believe or going we're full that that the that for year mid-XXs

our in slight to difference transitions introductions. have product different on cost possibly we're our So going again with that the see QX, structures new we and some

same number are are. in I guidance, have there. lower However, don't not exactly I'm than we where We you the you're but do that believe

And your line comes of Reitzes Ben next question from Melius. with the

I down that There geographies. sequentially. wanted came and the lot up were States a to Asian was the several sequentially United geographies ask the about XX-Q out, was while

well. China did Just wondering obviously, what And very the dynamics are there.

growth the puts if all for means are wanted your in to the company the and then from question given overall that rates accelerate takes? Stacy's I revenue in fourth those dynamics. sequential just favorable You quarter, mentioned the remarks. it What revenue And clarify

to, in up bit XX-Q, that the and/or of terms required sell we piece. sometimes Very terms in with a of of who right a about in our we disclosure even have disclosure as have terms specifically moving customer. eventually and end the necessarily The talk to terms we there a challenging the These pieces of system geographies. OEMs we across to me ODMs most to be our of choice of and the Let we who our come in it so disclosure where key what create X our in in to you're who for invoice portfolio. to product part product is and invoice. not just integrators or where will That's seeing terms travel may are

that their seeing of before you're those to there going and in yes, in they shift is center, are things number those it. from a into both mind sometimes happens what that gaming, the of time So numbers those to our data that China. terms Keep into But there complete And have. who notebooks shift incorporating using are Center, just are to also in configuration pieces is invoicing automotive going time. also we Data swift China full

continue in back and don't we from grow the we grow currently and QX for to looking we're We'll we're gross Going in to seeing your and also are Blackwell terms will what revenue. what mix at here. and in in statement Determining Hopper second of seeing. margin terms continue to QX, exact that regarding what each of half. have Hopper

not that, in see do the to expectations. demand we our growth here do are We will visibility right Blackwell We of have will that QX. guide of On be in see terms architecture. now yet top opportunity that, QX. But the we a

And from comes Fitzgerald. Muse with your question next the C.J. of line Cantor

cadence more only complexity You've and a a challenges product in more, with rather limit likely rising remarkable world. package embarked becoming given advanced annual on

this partnerships, chain supply does impact your if curious, to then consequential taking back, take your and a backdrop margin step a potentially you how So integration, alter vertical thinking through greater around profile?

thanks. Yes,

Let's see.

its believe your continue It's growing high the model answer the models, of first our reach level up, to so it. first believe increase to we why realize question drive by of because it to usefulness And industrial reason complexity continuing we'll velocity is we scale. AI simultaneously is next to and want open scale We that extraordinary -- revolution. to your that that a would the the want the the to answer think growing that down. cost we the and to so is I its continue is

scale. continue And so we're really up to to go that ourselves going to hard drive

every you to unless have ability, design not uniquely, all the the AI factory parts. AI parts. come with integrate, we new have It's possible have factory to the a year We because an fairly all up to

becoming ConnectX going lot we're year, to to than our DPUs of the way next bring had a North CX East multibillion-dollar Ethernet, is which GPUs, is more also NVLink and to -- BlueField West, and of have a a history business data storage us, for in switches, AI CPUs which we so centers, to its for well ship on ever more and InfiniBand and processing but brand-new Ethernet. course, to for company, product for South, to And DPUs, supercomputing we've

what go and factory, X as you sell got of a fact we customers to complete got all the to the access -- market that these you this, you've happens, we it. software. architectural we introduce could we somebody's as it to AI so the an -- AI stack, factory parts, capabilities find know, And got build have to you then new up to and have us ship build Otherwise, mountain to build of it to, has allows

chain it. We to Super that our that And Foxconn, purchased Quanta, service HP, who and in Lenovo, could recently We so forth. able sense integrates fact the Micro. love so be supply to were They so we ZTE. the disintegrated is on Dell, not it's about used and serve

gigantic providers, huge. and centers, all and way supply works, enterprise of into that the is ecosystem of them the ecosystem service take from have, And world integrated and number to is cloud world's partners chain, data ODMs architecture, that in our partners integrated all number vast but a the allows scale because the of that necessary bespoke so our Gigabyte, integrators, we the which we our reach have

infrastructure, part, the ecosystem provided not customers AI the And to but it don't and way doing. and how the we're do want lets know good would like Well, at integrate it. so yes. we design that And that -- we to

the that's anyways, why. reason So

with from And Wells line question your comes Fargo. of the Rakers Aaron final

I wanted to product go back cycle. into Blackwell the

how the asked One Blackwell rack as GB you that you about as how scale you as to of of go-to-market the think you about leveraging the systems think about that tend rack to scale NVLink, dynamic product far as I and get the think out? do mix it questions Blackwell that we to simply, guess how we cycle see cycle. put mix playing NVLXX is start system looks dynamic see

Blackwell rack's a system designed it's The sold the little rack. system, for that We thanks. surprisingly. rack the rack a is Aaron, because in disaggregated and different reason architected sell Yes, it's And whole everybody's components. but as don't

them are some of OCP are them of not. standards, Some

into data be power everybody all CDUs, the for Some of power of of the the And them limits are could configuration bus bars, Choice integration a centers, people's enterprise. and choice different. little different.

components. designed we system architected way is it, work the for the rack. whole the then going the so rack. example, perfectly across we Like to The whole provide the software we and And compute board CPU plant. a and and then system ingenious. modular the architecture. GPU ODMs MGX integrators all integrated into is MGX over MGX. OEMs is completely And It's an have we And

delivered, And so it's assembled data you would to close to has like integrated just got about be that to. would X,XXX-pound like, and any configuration be rack the you center because to where to It it's be close heavy. fairly

from point of forward, moment we centers. switches, ship so and chain the GPU, NICs, the the that the the everything is of the CSPs from the the the the data location quite to from CPUs, that locations supply And integration done close

and many partners. many you world so how And the how ODM with there are hubs centers we've in logistics scaled data our imagine to out can

show the because And and rack it's doing might that we're rendered we so impression left think that as that way, and I way X integration. it shown have we always because the

design the It's That's There's is value-add. customers we you design-in, a final center will. as integration. quite supply integration. hates do data to their Our the The a want integration. hate but fit simple in complicated. into us if doing that chain shimmy do not They as really

OEM is replace, a network and ODM installation, fit-in, design the that of this the the And over the and repair all incredibly we that so And done partners have does cycle the bring-up, world. well. sprawling entire

of the be don't So why integrator, doing want an we're reason provider. be integration to is want doing technology not to It's a we way it. The we anti-reason racks. the

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

you. Thank

to more incredible. AI. in demand with made are remains stack earlier me generative make a comments modernize Blackwell a for computing data computing make Let couple center accelerated couple and strong The of the entire is that steam anticipation worldwide Hopper -- again. I full and the

company. everything the CUDA-X accelerate CPU things, has computing libraries. Let top top reached must with Accelerated the X slows. the Developers our point. scaling of X starts things tipping me highlight possible. Accelerated computing

science for This CUDA-X NVIDIA. libraries New hot is Accelerated for data libraries, vector right markets Polars, new now. incredibly released CUVI-S Spark, including the data leading and libraries, We new Pandas databases. for and processing many open station, go can XG a that gene accelerated. wireless a we Ariel and CUDA base structure for now and data AlphaFoldX now. suite centers is Parabricks sequencing of protein of world prediction whole into for whole

happens it's two, number an accelerated a journey Blackwell the GPU just trillion We the to data one. Number leap modernize AI Hopper. are be not GPU. Blackwall worth platform. over name platform, is computing infrastructure step-function our infrastructure the to of purpose from is computing. to beginning at of an centers our Also general $X AI of but That's

X are package, and As platform than traffic, to a X our best for DPU a customers, becomes software of AI from factories. performance both optimized, computers. to on InfiniBand Xx the X we and the even storage power mounds deploy TCO. one-of-a-kind the CPU, NVIDIA GPU designed Blackwell's and years reveal traffic, chips, Gray and more ConnectX Blackwell they These and to colos of customers fast lead to and as NVLink power-limited of for and data for as cooling stack the provides massive and of BlueField systems, Blackwell AI North-South to all-to-all get the systems chips dual networking, and East-West Blackwell it more extent throughput structured it make Hopper. and to the partners are and end-to-end support nearly want soon and the clear. deliver can Ethernet size The vision equipment hands AI. Blackwell took Blackwell their realize, Customers and factories DPU switch for Blackwell very cables, infrastructures. full AI center for building capital-intensive GPU traffic communications, build in Quantum Spectrum-X sample

NVLink. bandwidth deal switch domain, changing. is game a very XXX connect The NVLink XXX X NVLink GBXXX XX its terabytes with X packages of third is in into GPUs big rack. with second This Blackwell system aggregate all-to-all GPU is The per in us lets an

boost second of you multitrillion-parameter Just the to models kind Hopper. than that XXx in XXX training higher to sense on about need tokens. because terabytes of trillions put of makes perspective, per that's

to data And from be amount moved GPU. so around needs GPU that to of natural

as AI inference, companies models ecosystem started For and Omniverse NVIDIA Enterprise NVLink NVIDIA is joining Agent fast to mentioned, platform AI is per general base IQ. large for AI, value The XD AI more bespoke video year, high-throughput for model the deploying has rate. racing for modalities tip are on infrastructure. And chemistry, opening run NVIDIA's sovereign Quantum Internet installed enterprise understand for used AI for AI exit for NVIDIA vital targeting can then dedicated increase it's deploy is partners a wave from of applications. Enterprises Foundry. the frontier scale token before. Generative countries as to footprint grows the scale-up, anywhere. software Colette at coding NIMs, iceberg. IT AIs, build of Nemo, in NVIDIA consuming also billion billions the text, customize AI Blueprints importance And is and at start-ups can but ad the world-leading CSP's are of images, services AI generators be businesses. help biology. AI, AI to large-scale That from robotics. and growing AI AI Chatbots, physics, image are recommenders, on companies and systems. deploying yearly just millions of per consists and search to NVIDIA of the tens recognizing cloud GPU investing dollars makers and the today. of generative are poised AI GPU and capacity, NIM will tens and X software time, model much Enterprise $X We're you momentum low-latency, an of Spectrum-X the model AI networking InfiniBand Generative for GPU networking we're to for to all of AI AI and safety now help factories, AI our language scaling NVLink and companies platforms, supercomputing We run generation. up $X,XXX CUDA-compatible millions. and have Thank year next than their NVIDIA's us AI plateau era exceptional Ethernet. is the are TAM And significant bigger and accelerating. Enterprise And now and to to AI next NVIDIA transform

gentlemen, your this we call participation. for concludes and ladies And and thank you today's

may You now disconnect.