META - Earnings call transcript Q4 2024

Due to licensing restrictions, you must log in to view earnings call transcripts.

Operator

Good afternoon. My name is Krista, and I will be your conference operator today. At this time, I would like to welcome everyone to the Meta Fourth Quarter and Full Year 2024 Earnings Conference Call. [Operator Instructions]

This call will be recorded. much. very you Thank Director Relations, Meta's may of Dorell, you Investor begin. Kenneth

Kenneth Dorell

to Good Earnings Fourth afternoon, you. Meta Conference and XXXX Thank Year welcome Platform's Full Quarter Call. and

and results me Mark Joining discuss CFO. our to Zuckerberg, today CEO; Susan Li, are

materially statements press Factors to Form results on from Before release to set information this new make these those are SEC. statements report to today today, and events. include as contemplated remind we on future in results to Actual statements. our remarks opportunity forward-looking or our by these take undertake statements. today's as we would assumptions XX-Q that Any that may a quarterly on we in get result these that based the filed earnings obligation I will started, and differ could with of like are of cause you differ forth forward-looking this call no forward-looking update materially

will and call The are earnings to measures our and present today's investor.atmeta.com. Mark. on release. is press During non-GAAP I'd at this accompanying to now investor earnings like an both And included to website financial call, certain measures. press the reconciliation GAAP GAAP available we in of A turn presentation non-GAAP release over

Mark Zuckerberg

most long-term is a by year. day. for people at going using it is All year. on of right. trajectory our a the everyone, least it be now like year. big really But billion our This lot always more X.X year initiatives Thanks, for a one Ken. ended joining usual, know each end XXXX feels We feels is to apps with today. than Thanks, clearer this I to than a strong big more of the going every like be note of

telling to XX that have on. teams on be I because to going So to weeks be trajectory we is that about want intense this our get keep we the

assistant. people, to going to a that personalized Meta this reaches intelligent I and a assistant more other AI AI, be service I develops kind is AI AI it assistant. scale, long-term highly billion usually when a leading AI the of In advantage. Meta reaches expect and X already that than durable people expect that year used than be more And any is once by

for have road We a personalization. with unique really on exciting vision year map focused a this

going what thing. are I choose People some transformative be the is looks like to like I most We works to use how believe the AI. all that their to their be to to think world. uses think year. them. to AI their and that same it the big to don't this fun to the their interests, there's how people made, of that want surprises their are their think that does one one to just that want about that People going this products going personalized same have we've get culture, I personality, going don't we and AI think context, everyone people continue they AI for and that be

is open great larger too. making the will and the progress models be become our and Llama I model and as reasoning X most think done good Mini this when X in very are AI Llama used Llama well training. year is widely and advanced with well. models pretraining, source looking

an Our and will omni to goal our models. competitive is with goal make X model, agentic capabilities. will for be natively And open multimodal. closed Llama source It's it X Llama Llama X to have with lead. was

use of So couple the forward that And months. for cases. lot to of over to it's to next a going more looking going sharing on the novel, be unlock and plan new our it's year I'm of

Whichever AI when has a profound engineering year the of be this be will coding to the problem-solving XXXX around going a most deploying this meaningful important market. one a milestone build and Potentially, is builds to to that potentially, And shape I mid-level first, an expect it agent the research innovations is in good becomes that think, a going history their like company engineer. as to have and, over as also large and AI I advantage well of it very in advance field. time. abilities possible

I the reason to future. course that's year why So that going set for this another the is think

many be recognizing in just time and real people glasses But they're electronics billions it's a a the hit. This AI year consumer trajectory to some glasses. for Our on overall AI glasses hundreds year generation. if are as defining next for of third AI grind. Many great have stylish and understand a million well perfect breakout the that is factor will Ray-Ban the a a determines being path of towards Meta sold we've will if X of XX see glasses, to history we're or products we been millions that when the And glasses eventually, be glasses like as for are be this as the great platform units, category. just form going computing to longer talking about these and AI this million

we capacity we the X-gigawatt week is it a AI almost to especially Manhattan I a were term. of this that billions These in over of expect hundreds it of that data all long invest online bring of infrastructure part cover there. that that are gigawatt will and will placed AI so we're center big the big last if announced significant And investments, potentially year. dollars a bigger building

We're time, use revenue at AI advances that of growth. investing the these by, planning to initiatives all same in increase this aggressively to fund

next accelerate the pace few over our will going That's put execute a plan growth together And new these the initiatives of next that how hopefully we years. also is will financial over towards. the we've headcount what So of on years. our well a lot few determine trajectory this

TikTok. of important about I continue our happen next the to years. going we're Instagram I billion more has There and And related know expect adding what's than think Facebook of discussion that, and continue the than leading reach I with Reels sign-ups of more apps Threads to million learn are people other a regardless platform day. over going We're monthly well. to number year product trajectory now X and as per that has become on going family this actives on to Threads eventually several to million growing. trends its expect to XXX more been X to

lot on Facebook growing of monthly the used I'm now is than the progress million cultural actives monthly I we're X than has year expect more WhatsApp in is XXX this world. the excited to and And the becoming to Facebook. leading back platform U.S. in in of influence. actives, towards its more by rest WhatsApp a to get OG gaining and U.S. making the share billion messaging like some focused continue it

All right.

for that this metaverse So will a a been number and of is working visually also that going investments steadily to Horizon been a The will using we've inspiring year when the long-term be growing. and metaverse. more stunning Quest make really start of land. has year to people the And on the pivotal number this is

more a Horizon's going to trajectory end think this year. the by about of lot know I we're So

unlock. our a can year We progress of This companies, now redefining our be relationship our innovation American administration values that with optimistic for U.S. big abroad. governments. is also and have the that interests this is and going winning, to prioritizes proud leading And I will that a am defend and technology about

So going this is to a year. big be

seen ever projects, to I try to the most I think AI, our have build that And I a our doing of exciting massive media, Between human of that awesome is this shape dynamic lot have glasses, connection. future the business some in bunch work do. the going think social to and that building future industry. things a accelerate we and to of we're infrastructure

As am you, for I who's journey here's Susan. and us. always, everyone with Thank on this grateful

Susan Li

afternoon, and everyone. good Mark, Thanks,

otherwise on comparisons reported year. billion, currency both basis, on QX X% total our total QX compared up constant consolidated billion, a a results. to revenue and are were with begin All $XX.X $XX year-over-year XX% Let's expenses up was unless basis. noted. last

lines, rate favorable impact cover reflects legal XX point accrual and lower the I cost growth our in percentage note from that Before a quarter fourth QX reductions year-over-year would costs. expense specific I restructuring

items, hiring which were costs. mostly Marketing revenue XX%, of In decreased by cost costs. driven R&D increased offset the employees, year-over-year. well certain lower G&A due in related and billion approximately legal by and generative driven by line expenses year-over-year, in infrastructure AI, legal-related driven as with specific infrastructure reduction XX,XXX by primarily up areas regulation $X.XX XX%, Labs to higher employee sales year terms proceedings. accruals as with driven monetization, restructuring were partially lower and higher XX%, mostly growth primarily of increased flat compensation costs, of legal by compliance. Reality to We infrastructure, ended XX% priority the a over

billion, operating quarter was Fourth XX% a representing income operating margin. $XX.X

Our dividends by finance billion. to per $XX.X tax were securities $X.XX billion with and share. payments Capital income $XX.X principal and $X.X infrastructure. or debt. rate the quarter cash $XX.X flow in billion $XX.X in leases, expenditures, the data driven XX%. servers, for billion billion, centers Net was was on including investments $XX.X in cash network Free shareholders, paid billion We ending in and was marketable year

with begin of Apps our to segment now our Moving segment. I'll Family results.

revenue to was Apps continues of online grow, $XX.X basis. daily we and ad a billion Family than Apps billion, people to Family revenue, contributor least ad up a Within QX up December. vertical year-over-year the on was commerce estimate community was X.X constant Apps XX% XX% of of total revenue billion, one the year-over-year. both the Family and growth. Our Family in Apps our largest $XX.X used of across QX And more at of on currency reported basis

XX%. from Pacific. by our at increased and served revenue strongest and our World user surfaces. Family increased of basis, impressions was Impression revenue was improved regions in North respectively. partially from growth number grew driven growth Pacific in was and services revenue impression benefited X%, growth, ad particularly total messaging the XX% increased growth Business XX% by lower-monetizing On Rest average a Apps by was WhatsApp the other Asia demand, million, This of and Asia Europe performance. geography by In platform. growth primarily of XX% XX%, advertiser offset up QX, from $XXX across price part driven ad XX%. America driven followed ad mainly ad business Pricing by at per

to billion, Family overall QX, of continue of operation Apps. the of $XX representing the Family toward In $XX.X costs of to expenses. operating infrastructure was were Family our XX% Family partially employee expenses representing We our compensation, lower of Apps margin. XX% Apps by direct growth of Apps expenses. billion, our which a investments legal-related and majority development up primarily were were expenses in and operating of X% due offset income

sales billion, loss Within QX Labs costs by $X Reality year-over-year. Labs X% Reality partially were lower and our X% higher year-over-year, revenue segment, operating $X.X by Labs driven costs. up was driven primarily infrastructure compensation, offset and Reality up by employee expenses hardware restructuring billion. $X billion, was

now community monetizing and our the to performance, factors X are time. ability outlook. our experiences over There Turning deliver effectiveness primary drive at engagement revenue our our business engaging that to that for

WhatsApp at On see the double-digit and seeing to on Instagram, rates systems. on where spent time in across Facebook, and In grew growth to video up year-over-year in our opportunities also actives strength QX, year-over-year, to ranking we're States. through continued Facebook, global percentages the drive first, ongoing XXXX the United particular We optimizations grow video U.S. year-over-year. video both daily and time in globally continue double-digit Instagram was

longer-term continue also for discovered. setting are to help several up We're our are our focuses. product to creators platforms prioritize posts on one focused and get success. Instagram, smaller making we central original of bets On that recommendations Creators

have a with We place content. their creators experiment also want to to ensure

helps introduced feature best content who deciding performs with QX that before also their entirely with a them to reel them we test what and it share So and allows don't audiences. allows share followers to first new new to see introduce people to in follow a creators This them.

tools area their the In it we'll creative in. to great provides app is investing make Reels new a creators a to that we're weeks, make another of on for launch full stand-alone called phone. easier tools Edits Creative coming suite

bring we'll We're for X.X connect more the a U.S., coming re-shared seeing note have very together over early Reels launched making we globally is look been liked. people new content expand over that in and recently the of that Reels Instagram. results, it In positive entertainment a easier are to of day, content. your destination left or social consists to in on a we've months. and this friends focus billion aspects times introducing and already the features Another

tremendous On up establishing accounts top Threads of number they our from and progress about. Threads, recent to place come is a recommendation see posts, ensure people year content follow. to keep making made from prioritize as we more focus XXXX, to in this We're the our of and what updates people creators, the care with more surface content systems they

improving topics can they're personalized also so continue on people will feeds custom We interested build in. feeds

preferences. they better and Finally, Meta AI introducing queries remembering relevant more from more Instagram AI with usage to Meta Facebook certain their to updates than interest what for people's actives. develop details enable deliver monthly now with engage personalized million considering to that intuition scale continues will responses XXX on and and We're and by prior

increasing revenue the Now to our driver efficiency. second performance, monetization of

part organic engagement. ads level this first optimizing The is within work of of the

We each surfaces ad lower-monetizing at time deliver surfaces continue ads optimizing to people. place most supply our like and the will while of be relevant on grow video, supply to to on they

we it's This most better growth. locations the in to ads personalize feed for the efficient example, revenue. to the introduce depth continuing supply optimal supply For optimal of ad show user including up, when someone's are and is enabling when

Threads, see test we surfaces which opportunities un-monetized like this also are term, to impression beginning Longer ads on growth on quarter. we

introduction overall be don't Threads revenue it of and the in being gradual expect or on We ads XXXX. meaningful will driver growth a of impression anticipate

thousand in narrow this improving ongoing to The enhancements marketing of of of half ranking In an systems XXXX, part an is learning performance. innovative new increasing are our system ads driver monetization the second introduced work. efficiency ads increase we second important The the the more machine ads of complexity called a Andromeda's partnership far we which few of generative process that models This process pool create system tools advertisers XX,XXXx the run to driven we someone. consider us The on larger is to and ads showing to quality to volumes we've models someone. X% is Andromeda. efficient in to prediction ads ranking the we also see efficiently use personalize for more of ability where as sophisticated part of millions retrieval, of enabling in in show an model enabled more which us AI ads people of a objectives of with down for we tested. NVIDIA future our use positions the has This the increase increase well tens ads. complexity better test

way billion growing delivering for campaigns is of XX% Adoption Advantage+ shopping QX. and increased Another with to ad through in run revenues advertisers a of annual scale automation their rate Advantage+. with surpassing $XX continues we're year-over-year value campaigns

and streamlined app advertisers longer we're sales solutions, Given choose the manual Advantage+ seeing shopping our so we're between a strong or no campaign. interest in need running new other a testing and Advantage+ creation flow to performance or end-to-end campaign

ability their to having later campaigns rolling when expand need the advertisers aspects coming plan all turned for beginning. the from lead will Advantage+ allow to customize on sales, they app In while offers months advantage more year. this take further to. optimizing of Advantage+ more the objectives out of will to campaigns We before in or the in have to it performance advertisers new This still fully setup, the

generation tool are with out we're already X,XXX,XXX now animation, Advantage+ AI months of of rolled our ago. October, in been is adoption it tools, advertisers There of early thousands of Creative that More million monthly. hundreds than using first generative using from another has significant video one we up image ad momentum. six our X seeing area where least creative advertisers at

allocation. approach to would Next, capital I discuss like our to

Our business primary investing into priorities. capital focus remains our the being infrastructure top back with and talent

working the of and ads features many capacity both these On our needs for core fuel we compute infrastructure We're usage and and pursuing to the engagement workloads. footprint efficiency products of opportunities the to scaling expect first, increasing AI of we're by across drive central platform, capabilities initiatives. and our will advance services of our organic the growing as the we meet generative Llama, be increased

efficiencies way the equipment. lives associated of we're pursuing by and is extending networking useful Another servers our

is forward replacing be in AI guidance. non-AI servers time our which for expense, and going we estimate that a of to expectation is Our our X.X deliver them, already longer period both and annual able use approximately be will we'll before This will depreciation which savings in years. resulting included CapEx

our compute deploying silicon deploying our the workloads and inference XXXX, by chip efficiencies started In content. workloads. for MTIA cost and recommendation pursuing ads we of organic ranking to achieve cost MTIA in we're optimizing we can lower our areas a custom by Finally, to unique where

We MTIA training workloads cases for these for expect next to our efforts throughout further ramp ranking silicon XXXX, adoption and before recommendations use custom to of year. extending

infrastructure hiring The be talent standpoint, support headcount to to was our added data strategic cost our compliance. Reality In roles technical the Labs, remaining regulation our In to initiatives operations. within growth AI we XXXX, function. by on headcount R&D year-over-year a was From growth the within growth center our as XX% and will expect priority well as infrastructure, be nearly as priorities. our in headcount generative we monetization, of primarily support primarily quarter, focus revenue fourth across continues driven adding to of continue technical

our We to We're executing functions to by pace. To anticipate be headcount base a in business limited. our productive. more achieve rapid need will areas, will engineering these continue tools remain this help our ambitions growth supporting relatively in we at building

tools and processes helping developer our assistant, code efficiency improvements of our X tools our we've write past and years, over quickly. part engineers As the coding is focus new more introduced AI-powered internal in like our which significant made

will not base. Llama's capabilities code the we engineers, advancements expanding in quality continuous of generating to even code our capabilities forward, code but provide our greater updates changes assist to in we and writing improve reviewing are and expect and only our tool our focused on engineers leverage the to to that coding automate its begin also Looking

our expect dividends. return support shareholders we repurchases to enable strong financial Finally, while through position to continuing will capital to and share us investments these

outlook. Moving to financial our

constant $XX.X expect XX% be the to We currency XX% X% to to $XX.X is currency basis, year-over-year total foreign XX% assumes in the on revenue as billion first first guidance total of or also to of growth of range billion. a This current reflects quarter in to our approximately on the rates. leap quarter This reflects exchange effect growth year-over-year based headwind revenue an X% XXXX. lapping growth day

While we delivering are providing in will continue making we're give this throughout growth investments XXXX. we year strong the outlook, our business XXXX an not revenue opportunity core revenue a full to expect year us

Turning expense to the now outlook.

full expenses to XXXX year range $XXX expect $XXX to be billion. We total the in billion of

expense of We in the costs, largest higher driver expect growth expenses depreciation. and to operating XXXX infrastructure single driven be by

We be factor expect compensation we I areas that priority in employee referenced largest the add earlier. to technical as the talent second

to outlook. Turning now CapEx the

expect be will expenses in billion $XX the to capital of range We XXXX full our year billion. $XX

continue expect business. to majority and towards driven core our of We efforts both in XXXX will to our be CapEx our The core investment our increased generative in support CapEx will growth be business. directed XXXX by AI

range XX%. to tax. we rate tax to expect to any the Absent full XXXX in of tax changes be our On to XX% year our landscape,

and headwinds In the EU impact the regulatory addition, business results. monitor regulatory including in and to continue U.S. financial we and could landscape, significantly legal that our active our an

across priority community. and company, business a products for good this strong our our new closing, In delivering performance our areas innovative was investments with year for

growth we let's set compelling this drive company up that transformative in the us will With technologies continued which the and our open questions. shape develop of to a invest strong and call of industry. have help Krista, future expect the We of opportunities year, that, for

Operator

Stanley. question comes of Brian line Nowak Our with Instructions] first [Operator from the Morgan

Brian Nowak

potential of your then you and utility and Llama all question of announcements use but that of a wonder just you could for the X innovation to sort Meta think high-level value custom changing new lot Mark, course if and AI on about of and of the throughout cases on few sort drive all the this excitement year -- could I XXXX. And appreciate advertisers a vision about examples you. a share silicon, second of the know offerings your throughout come. you sound for I users very one we the for there's maybe your either as year,

third-party Just custom results? chips higher think to able any factors between models how engagement on difference to as learnings silicon? percentage quickly gating And you'd be a we the about your ranking your move silicon the and your custom in main should how of your to and

Mark Zuckerberg

AI a I On a tried comments out very can my lay as access I to assistant that website. intelligent and opening the this on you in apps. our mean bit. one, personalized first across highly we're a There's focused Meta

our apps, You can access too. it of outside

the I integrate last of more lot keep discover also to improved to useful help think a the our this into it's and quality more improving over going that it. ways people to it's on that year. services finding it is just -- We're

that using think why then undoubtedly so kind easy millions to it of what of hundreds it's many are because it. today, discover and people we're obviously, keep doing of I that's using

that fun just to refrain the Llama sure I different of I'm on year. release going thing. Meta what AI. releases. technical exciting a do to do to But I things I'm for one set going a before. to adding will whole But because that talked expect launching I about the It's X X haven't be about. lot know. with that that I gave talked more not planning we're a from to we're this very what keep surprises we of it's It's bit this people we're don't Llama stuff and detail some want on I year. enjoy

a there Llama with models X, different different at dates. like were of few kind Just

X we'll too. think Llama with see that I

our an is we're going very our I I mean, And to soon. going be external it. and advance what then on, advance think the AI product for about working it's that But piece, AI profound goal engineer to don't internally. that's thing. I research I be anytime know think And just own to excited development I'm really a

that's mean will But as will But something that show better over I products and through opportunity market that be year, and -- AI I like really now development. this for think mean, all -- year we're think I is -- a deployed, to road. making I this that you're up widely engineer down don't this works, is So our there think an going I time. changing of the see extremely then potentially

I to think dramatic in be the year where lays become kind of and for this much the going possible groundwork don't Yes, starts is 'XX a I to it. change beyond. that really know. more that's and

Susan Li

custom Brian, your about I'm silicon. question to happy take second

all, long-standing workloads, off-the-shelf compute our versus and different to industry. per specifically isn't So we in the efficiency versus partnerships. needs we're the for also specific committed workloads necessarily and memory greater because And of silicon optimal require different of we're performance might first from a and of purchase silicon we able cost optimize optimize compute, that really our providers continuing third-party to silicon to certainly developing types we very the to of so mix workloads. expect where network, bandwidth invested unique stack our But are can full achieve that because leading those custom to own are we in power and

workloads. inference in-house inference. in ranking for recommendations now, core half XXXX and our program the the ranking We MTIA adopting on and MTIA is started Right of focused first recommendation supporting core

when use those to over for capacity lives. as the workloads adoption end ramping reach incremental of servers of and for continue both useful it they their GPU-based We'll we some replace course XXXX the

some MTIA training we're some expand gen use to year, core our and AI workloads cases. over our time, of AI to support hoping Next of

Operator

Your next Goldman comes Sheridan of Eric from Sachs. with question the line

Eric Sheridan

AI approach go to comments your can open to I your you capital source could Maybe help the competitive back on understand with versus evolve others return and how for continue with open to how to medium the the to bend views dynamic And respect improve term? cost curve Can approach open us on your source. possibly over your around long the in industry? source

Mark Zuckerberg

Yes. we open did on the weren't where first what I is building for best us source, to I Open we think mean the Compute, system. analogy with

a So the got that around then it, to time have building to wasn't big be it really advantage it by we proprietary.

we doing, By it to a standardized then were lot everyone. on the building meant which a contributed that efficient it, So chains back standardizing bunch shared of that of industry on adopted And it. made innovations more supply what it, we for it. prices

cases, their more think is we as likely, other we some others, platforms, that developer Llama the can, silicon providers see example, what used, I and it's basically of it down APIs will using for more and more here improvements and that becomes costs optimize in too. work for use and drive drive that

to So strategy I think that about continue I mean, be the will this. continue to be optimistic effective. And yes, I

standard it's also of think talking light new it's source of to the think be globally. the it's in just -- DeepSeek an from competitor, open kind -- puts recent also news, I think think it going I there's things just of we're working. of one the I I is some about that China,

think American I important it's of kind own national that standard. our And it's an advantage, for

this AI take the us focused some the news So thing using. are anything, only to we want that strengthened on. we the of is system be that people that seriously, think world build to right And our conviction recent the that and has around I if for

Operator

Your with the Mark next of question Shmulik Bernstein. from line comes

Mark Shmulik

of more of out the Or is factor kind remarks? it highlighted the use as assistant smart better do and to AI the track so this answer you've here. appreciate as looking of form Meta your you the way an to get most seen in we opening I progress you in-app today? you the forth, out, a to may But it glasses, Orion experience people in Mark, of complementary kind get view that

we've few And quarters, the pricing driver seen kind then, as the revenue growth dominant last ad of Susan, of growth.

forward? Given for the the how growth think deeper, about engagement we contribution efforts more better the going advertiser around ROI, you've just ad formula and commercial to driving do highlighted revenue

Mark Zuckerberg

your mean form assistant glasses context that can that's it a you while mean, let an understand talk I device an talk on I for the to see you to are -- glasses. what it AI context hear, mean what think yes, see I've can the be your going glasses I a to in which ideal AI life Yes. said you about it's get I and and would because that want you for I hear to that everything able about on on. factor gives

mean, very phones platform I away. computing computers I went it's -- a primary became look, important platform, in So the be like but the not that going think to future. glasses are When computing

some I time. now, to think that from a for finding hard we'll to basically AI aren't that a glasses. who well be phones going there be It's to in are of imagine But world me as of of glasses glasses for the decade who people or have a today, kind lot as a lot the all useful people, thing. don't glasses more wear have

So I'm about optimistic incredibly this.

of I AI think of for much that drive I -- previously is a glasses got AR year clear was we as of holographic going big shared form I factor at thought that these actually full the become I last last to showing displays But weren't started in kind value like as surprises the going it's now major year, holographic think we least one that And Orion. the pretty to prototype until the the is.

to cause excited. a be that's So

a Metas be. a trajectory this were to We lot going we're going I is Ray-Ban what learn And don't the But hit. look, think this year. know for to still the long-term

So really year I that important this a that. for think is

Susan Li

second question And growth. on I can the take pricing

pricing, to continued growth all, show at the where sort term, and mix look we growth have of surfaces supply and the drive both of would we by both When dynamics such ads of be of revenue can our we different what demand So supply I up. across think say so types factors different is growth, over shift pricing the influenced because impression reported as opportunity first by of auction long dimensions. the

like course, lower-monetization efficiency macro then, of speaking For example, video services are broader and relatively factors.

lot kind we the a what and per to performance And of their growth is is cost blending what are road that -- as work price what pricing is doing generally we that have a is -- those actions. really to it's ad an the what happening, to of about are map, the going blending of average output the deliver ongoing of that ad, including of things, think average ongoing of bids But bidding for, able we're overall. their sort expect effect here for think reported, will monetization our I important of are benefiting through about when across we we think price we improvements that really metric. It's be part are advertisers things lot

thing. given different that complex optimize that one into distill different have so it's So very to can values, objectives that a for that there many that are advertisers metric tries

that And served, will impressions. Overall, we conversions seeing trends when cost we to of whatever for lift continuing we action continue driving do, the we're time better And for. CPMs resulting we'll conversions that impression are healthy per in action per optimizing advertisers have because to is advertisers. over believe delivering higher-value they for more the for effect are at get

Operator

question next the of from comes line of America. Your Bank with Post Justin

Justin Post

for it as adoption? changes advertiser and as one U.S. positioning Maybe You one for do for in usage far fact the Susan. better and But the Great. U.S. companies mentioned abroad. you how think of rid got maybe political in checking. Mark, you Mark and U.S. about

it think all? that is So you do content -- at users? could impact appeal that more to Will Could change? advertising

pretty people know I monetizing be AI, the about the CPC about Meta also it How use you on think about do that? Could how thinking Susan, case. Or ads? about thinking but that? are then you excited case revenue And are

Mark Zuckerberg

think that see about policies. expression us. we Notes the need to question And I what is that fact service someone I'm And think, content effective not system better Community system, for system mean, had I time don't checking is want the think that be found our gives to people in trying something I effective that's but you People than more more does than to before. build the admit just we we're look, going for I afraid and the that misinformation, believed while. people. we build over was when a is context. free I quite The to an think best to

to best systems. our and implement the the of I go it's job do sort think best just work and

a Notes I than our if because going I our doing we about had who adding it. that the are lot for of there's care what effective is to before. product a people are this not Community that have like And read don't things on as has system, That's to been we I were of more X So think think actually misinformation. get announcement better somehow that while, what context think just platform actually that right. is

Susan Li

policy to I would also that, advertiser just noticeable content impact seen from to our add spend. we changes on say, any haven't

So advertiser are tools value maximize for their of we're continuing ad AI-powered the strong particularly spend. again, demand, businesses helping see to that

to that to So brand meet of in needs advertisers. we suite safety our our and commitment the is of will tools we invest expect unchanged,

focus There AI, question frankly Meta that's second your consumer right clear of for kind focused pretty premium experience, not time, I monetization offering, a are to Meta building paid we be directed energies development where think, AI but and terms here of over in where all On really and a in our now. about really terms including today. of are initial opportunities of including is monetizing Meta the recommendations of great AI our that's will,

Operator

question comes of line Anmuth from JPMorgan. Douglas Your next with the

Douglas Anmuth

for source over others this How on just or one Mark, cheaper. for as the train And models Meta for required what faster and could a Llama investment trajectory of it up following mean leverage view? does open to potentially your and multiyear Susan. Mark, DeepSeek period? other impact in One

$XX billion think training And as between networking the last And by you to servers on this as data talked billion the does composition and then, much and year, CapEx year inference, we just change following how week? about equipment? post Jan's this should Susan, part followed about just we up from like mix that think when about $XX largest centers the

Mark Zuckerberg

on the start can I DeepSeek question.

I think a that's works, not. this that I that's to And a to every this of an that And technology and have strong kind whether digesting. a think will there's Chinese don't are part from, has of in a competitor around industry advance that. advances trajectory for company the to like new a number that early number it's I of that it's probably has some too know that -- nature the novel things I hope opinion new there advances of launch that going of on our systems. that and they implement the they of that expect really how that things the field sort things still have we're have we or is infrastructure CapEx -- the goes. learns means what rest of how and did

that mix compute There at towards we're here intelligence happening get a around already time infrastructure sort you use of more inference, infrastructure these you are a compute models all go compute trends as to using debate There's or shift models, of just versus how of of going where how bunch get reasoning by are to will we that of into the towards pretraining is whether once. reasoning more much the more of that. the our -- putting it

pretty labs we're a think and all to I and -- that of towards like that pretraining. starting seemed think about ourselves using, necessarily likely the before other the the going that largest go that of something even -- already already aren't more the to were compute lot was That this, pieces

to means that at apply generate new of emerged you level a intelligence the going provide advantage to now a a don't quality of need a as a to higher that's has higher to higher business time less more and But that model of necessarily generally on support this, compute a to one business support is in basis. than mean order to of that that service who which strong I inference we're others the quality model have that's properties service, it company the compute think an that be doesn't sustainable because ability able

thing of serving Meta people agnostic building and going also that is model, of of that implementing we're all a sort are to is and for other serving things things a is to okay, but of AI just The and we're ad people, billions to using many these different lot some how that, into be are model level, because people pretrain all products serve AI, it's the us start like people. from, we we're when you like Like feeds it. to at just how expensive which

effect is. sure this of And so of I'm of all net not what the kind

We'll just the everyone from basically do what as making to of kind incorporate The the from month. move releases to over learn just a field as contributions this something continue not who There's -- not does ones interesting, field to into to ourselves. continues novel quickly. we last the that well lot

a over major CapEx early otherwise I is point, and that kind that serve And we heavily the advantage the but want bet to this think way scale continue to time. at going is to. going to be build too and it's at we'll call to It's very being service to out I quality infrastructure that just to be that. able And investing ability for point, think would that I learn in of of that both a possible the advantage the infra strategic some

Susan Li

to color little plans, add XXXX to second happy your more about a CapEx question. I'm our

is biggest to expect So of the be XXXX will across remains growth components That CapEx described. we our grow driver. certainly budget. X of all going the CapEx portion that those overall you Servers largest

we existing AI and support We a expect as invest refresh AI. the growth of in including to efforts and to in are business, base core But both capacity to our also we continue engagement in AI meaningfully in capacity higher invest servers. gen growth our expecting core we as support non-AI

density side, We're AI we're phase. to XXXX build-outs core in be On are non-AI of that our higher power for data the expecting driven capacity to centers that data the and by center clusters cases. center higher use anticipating construction use our large core spend data and primarily training entering

traffic. side, We're training to networks we to handle the On training as cross-region we accommodate our AI-related with networking to build large along higher-capacity in fiber in and expect investing spend 'XX gen core AI grow also future clusters. growth the traffic, in non-AI networking

to workloads the AI is or the be AI perfectly, non-AI building -- and can business so some for servers for to of of caveat infrastructure procure and the increase gen our to non-AI and use that's so core use majority that, expecting terms not we we're gen can each measure on AI, spend that gen easy directed with core AI forth. GPU-based centers cases, data total breakdown AI within cases the and in core for support as with CapEx we're then AI 'XX versus re-purposed And our in core of

overall, said. would But Mark I reiterate what

building models leading to foundation and committed applications. are We

we we're our expect to support that of to cycle and exactly the in We investments inference objectives, going where yet. we and are know training don't big that make

Operator

the line of Your Josey Ron next comes question from with Citigroup.

Ronald Josey

positive. use want to Mark, marketplace want to the and I groups on bit Facebook, Local could expand. more the comment back to and get back Video a that how is understand cases been on OG I your have getting all clearly benefit. to a little

AI, seeing experience doing XXX-plus Meta on people just Meta the given adoption does Facebook? AI? back we're insights And OG any the MAUs, What So are here? on with then to user evolve the the how

Mark Zuckerberg

Okay.

part Facebook and last interesting make maybe an of a think I opportunities important their lot haven't that that development fun will are sort it it that have over product a of a think that's there today. day, as every And lives. it's -- way And So directions focused of take much the for lot use than to on Facebook, culturally some interesting more of people that and our is in several I goal we influential years.

other and of product on. one area near And think just maximizing than mean to know of doing, this we what to more areas specific ahead it's we're time of I near some an term, investment that something it areas of the some mean So in results on I have is kind focus don't might to trade-offs kind focus year. some the term business I on in anything this -- going spend going of it. a be for this on that I that I'm my much make

know, half-year many But follow but specifics some to call stuff it, I'm doing on up think into And I rolling overall, be not start this or of I about was some it don't And used get now, excited originally I so in we'll to get the we'll exciting we here. fun. really out. how I'm of will Facebook year back over think this going day, will -- next, kind stuff the as get back

Susan Li

people I'm are Meta and happy to it. share what a little about AI with doing bit more

We learning a people phase engage with we where are in the are that really from way Meta lot AI. a

most WhatsApp one-on-one support are threads, perspective, along app queries, engagement from for continues to with Apps. our the AI Family messaging. So an group though use using People information cases. WhatsApp frequently there the emotional Most it in across see some strongest seeking of Meta usage and we is in educational of usage see

Meta AI them. we're people our integration about content which lets Meta ask driver deep engagement, from second-largest strong seeing on And engagement the that is to dives that questions of is Facebook, recommended feed AI the

casual query social signs and it across, see for to editing, They that recommendations. gathering, communication. new our information it AI all use I Lots talked for cases. conversation. and would about research, we writing people is of apps people So helping continue use and interaction types, leverage use for Meta humor say, We

to And look on AI Meta forward as make doing our really we more we to it focused feel XXXX road map, in are more personalized.

of makes including certain remember on, dimension the able be working most that of experience. really chats, increasing to to I some for really we're people ability great recommendations exciting We're details so and enhance people So say its to would and then able today. the share details features and AI improving personalize memory -- valuable Meta the for its of sort and content deliver one-on-one in responses, Instagram what example, Facebook those use

Operator

from next Wells Gawrelski Your the question line Ken with Fargo. of comes

Kenneth Gawrelski

me, Two please. for

reductions or are you recent I see a -- yet the it's know talk you the on hard things laid little parts maybe needs or about you to the we you about we this tell hiring future externally on think as year. think that think You for this could hiring, I'm planning Could hard how First, you sense then the of resources us as any could talk just a 'XX bit term CapEx -- that? should where we we a you this about should a year. curious, thinking about either think you Could bit early out OpEx and give bit seeing you're ecosystem? And you it's think about little as the about where just in the -- of guide gave or performance-related second 'XX, -- how But little you -- But talk side about for near and budget internally could to the needs the earlier and capital the bit intensity a for little just -- for your you seeing just constraints, announced one, about developments, 'XX, be more going. beyond? we constraints about and knew the talk forecast. talked any CapEx one you're

Susan Li

of Sure. those. I'm happy to take both

updates So your our will our staying is against question influence are constraints do to obviously, XXXX. where in supply in any first have CapEx the ability now. on That plans, see But we spend share factors CapEx supply on of our that on execute just we availability don't on of right availability. one really we certainly to top

prior needs. and I it we're with When long-run you those and engage think We to like. look comments, to intensity, both really Mark -- determine capital to what GPUs capital I own our intensity think how is is are not alluded to vendors planning about going our invest in as too of we'll to to in meet continue early XXXX, significantly and really ramp in our silicon up deployment to asked

the can from what ultimately, AI our then other gen performance they our hardware own adoption of so innovations, third next-generation products, how are be, underlying advancement gains investments pace models, party, monetization many different case come factors, efficiency AI both or and unlock. what in gains efficient what There is the and use our and of

long-run have So yet. capital I anything to would intensity share I sort about in here, of we we the we say, again, have think don't are we're -- early of -- and kind journey don't

thinking hiring was question about needs. second about Your

in So technical generally That expense are by areas monetization, the keep next it functions, growth driven Reality segue and largest growth compensation where I headcount compliance. means in those primarily employee comp regulation And XXXX. after that and infrastructure. that Labs is growth more it's generative base is driver those more AI, a constrained. expecting also are is of to mentioned, here, relative Employee to cost And business headcount a infrastructure good organizations. higher and we broadly

of to we're that that And in to to is want efficiently. a feel areas I make in a win, the would at say staff -- best that us same we critical we we're are are -- way focused terms positions we we really making those period time, that. and sure priority sure running on company in investing do But we the it like

Kenneth Dorell

Krista, we have time one question. last for

Operator

comes of And from the that line question with Sandler Ross Barclays.

Ross Sandler

Mark, for Yes. One on agents.

Operator OpenAI's week. saw all we last So demo

products, into moves And commercial how reasoning as I are Llama from some agentic experiences these AI monetization moves might these X how agentic industry to Mark, intent the guess, and So new AI? Meta more potential you about for AI? drive then for chat behavior Meta help thinking of

Mark Zuckerberg

Yes.

lot you're say a One to couple of is perform multistep being things about this. of able thinking on guess agents that So reasoning, when I is this I'd and a tasks. about

you systems the now, something it responds, and chat like. way of say kind a that right is then it's and almost these lot work So of

able an But to task. sort able be going you're compute of be a or as use I arbitrary be going to going to want much think use a as task, it and do to it's it of an it's that the direction off is to intent able give that and you to on to go amount

for Some a tasks thing. buy Go people. be of might pretty the simple specific

hard like good really Some of this possible. might an optimize as them really and write like make code go be app it or as

think, thing, becoming to just start of year type that And and prevalent the more next or going I more over X. is

smarter. in generation very just it products. like it new these like feel a sort form won't think will It In of of ways, chat. of sort like of So But because it's as sort getting it's some the be much current feel, will factor I others, are another products exciting.

think So and -- excited I in lot it's there's general, to build about. just a be

We billion that these just try focusing I you product that guess, reminder my or that if reach really And of caution my kind it's scale at at them will, more. scale periodic our to development monetization. on on people to products. is they're once build note X usually of process, or start we point we

before. experiments up example. But meaningfully monetization We sometimes, reach typically to So for see experiment ramp or big we'll things were contributing with with them don't the we scale. business quite Threads running we really as until these a some now,

the think sort have that a that major So improvement that. of used. the methods is be going of kind sufficient year, kind to flywheel introduction the going it's advertising and AI that durable at be be will going a and business and and from to starting meaningful that, taking are to this improvements product to will applying that the scale. the to to year. be business advantage. is the of this scale my year, mean expectation of people year But recommendations things This the get a using At of like to be feeds is I and that have thing it hope doesn't to getting kind Last year and contributor them was a this sufficient and the AI we to it this and of

and people communicate for an here. I our part. thinking and AI opportunity for as to important about the business these think agents actual that interacting remains the for you're AI people and prospects and outside that's for internalize So Studio with AIs of most And to us Meta thing business 'XX,

We I'm But take We scale we it. but it We do. what some it like very many large. to nonetheless, we've be product. run this business build time. out times. make process We optimistic, good. going it's That's a to the a build around

Kenneth Dorell

We forward for with Great. everyone, again and speaking look today. time, you appreciate you, Thank joining we your us soon. to

Operator

your participation, you may Thank concludes call. today's you This conference and for now disconnect.

Meta Platforms (META) 29 Jan 252024 Q4 Earnings call transcript