저자: Alex · 외국인
요약 보기
엔비디아는 게이밍 매출이 크게 줄었지만 데이터센터(H100·Grace 등)와 자율주행(Hyperion·DriveSim), 옴니버스 등 AI·산업용 소프트웨어 사업으로 중장기 성장 여지가 크다고 설명. 이더리움 PoS 전환으로 채굴 수요와 EVGA 파트너십 종료가 단기 게이밍 실적 악재이며, 미국의 중국향 AI칩 수출 제한으로 연 약 4억 달러 데이터센터 매출 타격이 우려됨. 다만 데이터센터 매출은 연 50% 이상 성장 중이라 장기 우려는 크지 않다고 평가. 제목과 달리 명시적 투자 추천은 없음.
다듬은 글
whether you put a piece of your paycheck into a retirement account every month or you actively pick your own stocks you should know that nvidia sits near the top of the s p 500 and the nasdaq so knowing what they're up to gives you a big leg up as an investor well nvidia's big gtc event is going on right now making this the perfect time to decide if nvidia's stock is a good investment but there's a lot of misinformation out there when it comes to nvidia and i have a real bone to pick with how the mainstream finance media covers it let me show you what i mean here's jim cramer on the first day of gtc from earlier this year what it all boils down to is that nvidia is one or two steps ahead of the competition in nearly every market where it competes and i say they are light years ahead of all companies when it comes to artificial intelligence melding with machine learning here's the bottom line i know this is a tough market for stock like nvidia and i wouldn't be surprised if it has another leg lower but when you look at what this incredible company is coming up with i think nvidia owns the future that's why i still love the stock long term and i view any weaknesses a chance to buy more gradually on the way down so jim cramer likes nvidia he actually likes them so much that he named his dog after them my initials are amd and even i think that's weird but nvidia's stock is almost 50 lower from when jim cramer said to buy it in that clip so he must love it at this much lower price right let's see what he said at the start of this most recent gtc and we're short nvidia it's a loser you do not need their cards anymore to mine and at the same time they haven't been able to make the transition yet to being artificial intelligence a virtual reality uh we're just not there yet at machine learning see this is why we can't have nice things in this episode we'll take a look at the science behind the stock to decide if nvidia's future is really one worth investing in your time is valuable so let's get right into it nvidia has a one trillion dollar total addressable market which can be broken down into roughly five big buckets 300 billion dollars in automotive another 300 billion in chips and systems 150 billion dollars each in ai software and omniverse software and another 100 billion dollars in gaming let's start with gaming since that's usually what people think of when they hear nvidia nvidia currently has a big problem when it comes to gaming their gaming revenue was down 33 from a year ago and down 44 from just last quarter that's an insane drawdown so it's worth digging into here's what's going on nvidia's gaming revenue primarily comes from their discreet gpus or graphics processing units which are the chips that bring computer visuals to life nvidia is by far the market leader in gpus with around an 80 market share amd owns the other roughly 20 back in 2020 when the pandemic lockdowns first started production lines were stopped delivery services were ground to a halt and every part of the supply chain was disrupted at the same time demand for gaming hardware skyrocketed since everyone was stuck inside as a result nvidia's revenue from gaming ballooned up to over three billion dollars in just a single quarter marking an 85 percent increase from the year before that's a truly insane growth rate for such a big company but that insane growth in gpu sales didn't just come from gamers it also came from ethereum miners the problem here is that all of nvidia's gpu sales are reported as revenue from gaming regardless of what they were actually used for that's why nvidia had to pay 5.5 million dollars earlier this year as part of a settlement with the sec according to this sec filing which started back in 2018 nvidia didn't properly inform investors about how much of their gpu sales actually went to crypto miners and what that could mean for their gpu demand in the future now nvidia has another big problem the ethereum blockchain is about to switch from proof of work to proof of stake all you need to know about that is that this is a fundamental change in how ethereum gets mined which means nvidia's gpus won't be used to mine it anymore this is where the decline in gaming revenue is coming from but it's actually even worse than that because ethereum miners will decommission their mining rigs and sell their nvidia gpus on the secondary market and at a steep discount so they'll end up cannibalizing a good chunk of nvidia sales to gamers over the next few years and just to add even more fuel to the fire evga is ending their partnership with nvidia evga is one of nvidia's earliest and longest standing partners they've been making awesome versions of nvidia's graphics cards since 1999 and many gaming nerds including me only buy gpus made by evga last week they announced that they would stop making nvidia's gpus due to the bleak financial outlook after the ethereum merge as well as alleged mistreatment from nvidia themselves evga's ceo andrew hahn said despite his company holding around a 40 market share of nvidia's cards in north america nvidia would keep evga in the dark on component prices and regularly undercut them with their own first party branded cards yikes the one grain of salt i'll add here is that nvidia discouraged the use of their cards for crypto mining in fact they added software that actively limits the hash rate of rtx cards it was actually evga that supported mining on their cards and released drivers to bypass nvidia's first party software limits i'm not trying to defend nvidia here i'm just pointing out that this breakup between them and evga isn't as black and white as some mainstream news sites are making it out to be so nvidia's gaming revenue is definitely in trouble at least in the short term but nvidia is not a gaming company at least not anymore in fact gaming made up just 30 percent of their 6.7 billion dollars in revenue for the quarter even before the steep decline their gaming revenue would still be almost a billion dollars lower than their data center revenue which was 3.8 billion dollars for the quarter and up 61 from this time last year this is the part of the business that jim cramer both is bullish on or bearish on depending on when you ask him nvidia has a few different data center chips that are worth talking about many of which were highlighted during gtc let's start with their a100 tensor core gpus these processors are specifically designed for data centers focused on ai data analytics and high performance computing this is a massive market that's quickly growing nvidia's new h100 data center chips were announced at their most recent gtc event earlier this year they're a massive step up from the a100s in terms of performance promising up to a 9x speedup in ai training like the kinds used for self-driving cars which i'll talk about more in a minute they also offer a whopping 30x speedup for inference models like the massive natural language and image processing models behind gpt3 and dolly grace is nvidia's first data center cpu and it goes on sale in early 2023 according to nvidia grace can deliver 10 times the performance of today's fastest servers on the most complex ai and high performance computing workloads one clever thing that nvidia does is design their chips to connect together to form more powerful computers at any scale for example eight of these h-100 chips can be connected to form a dgx h100 server system and if you combine nine dgx h100 server systems together in a rack you get a dgx pod which is the reference design that can be scaled up to offer ai services at the enterprise level and beyond but it doesn't stop there you can link 32 of these dgx pods together to create a super pod and nvidia is linking 18 super pods together to build eos which will be the fastest ai supercomputer in the world when it comes online later this year see what i mean about their chips being able to scale but nvidia doesn't just build super computers for themselves their hardware can actually be found in over 70 percent of the 500 most powerful computers around the world super computers like meta platforms research super cluster or rsc which is expected to be the largest customer installation of nvidia's hardware this supercomputer will connect over 6 000 nvidia gpus to do things like translate the voices of everyone in a large group of people each speaking a different language and their accelerator hardware can be found in every major cloud platform like amazon web services and microsoft azure as the world becomes more and more data-driven demand for nvidia's data center hardware will keep climbing across every industry and it's already their biggest business unit today but it's not all sunshine and rainbows for nvidia's data center business just a few weeks ago the us government ordered amd and nvidia to stop selling ai chips and systems to china and since all the systems i just mentioned are built around nvidia's a100 and h100 chips they all fall under this restriction which nvidia said could cost them around 400 million dollars worth of data center revenue per year considering their data center revenue is still growing by over 50 percent per year i'm not too worried about that in the long run and speaking of crazy growth let's talk about nvidia's automotive solutions next which have grown 45 percent from a year ago nvidia showed off a lot of awesome automotive tech during gtc nvidia's drive hyperion system basically is trying to provide automakers with self-driving as a platform think of this as nvidia's version of tesla's cameras and full self-driving chip but with a few key differences nvidia recently announced hyperion 9 which supports 14 cameras 9 radars 3 lidars and 20 ultrasonic sensors with all these sensors the hyperion platform gives automakers many of the tools that they need for different kinds of driving scenarios but not every automaker wants to offer full self-driving some just want premium features like drive assist self-parking and smart summon nvidia's hyperion architecture is also modular meaning automakers can use only what they need which will really help nvidia capture this market nvidia also has drive map which is a crowd-sourced real-time digital twin of every road that they get footage for if nvidia knows where a car is in real life they know where it is inside this digital twin which means they know the environment around the car like where all the road signs are and traffic lights and curbs and so on then the hyperion sensors in the real car can double check what they're seeing against that digital twin on top of that nvidia has a tool called drive sim which takes in recorded drive data and turns everything in the scene into an interactive object you can remove add or change anything in the recording to run a bunch of different simulations in that environment or you can change the environment altogether with different lighting effects and road hazards and train self-driving software on different scenarios without having to collect that real data of course the big thing to remember here is that these solutions compete directly with tesla who's been able to develop self-driving capabilities using only cameras and computer vision if tesla can really reach full autonomy their sensors and systems will be significantly cheaper than radar and lidar enabled systems like nvidia's if you want to see a full comparison and breakdown of nvidia versus tesla when it comes to self-driving i'll link my video on it right here both drive map and drive sim are connected to nvidia's omniverse so let's talk about that next omniverse has a 150 billion dollar total addressable market by itself the goal behind the omniverse is to provide a toolkit to be able to make digital twins of everything in the world the idea for the omniverse actually came from pixar who needed a way to set up complicated 3d scenes so that hundreds of artists could collaborate on them in real time that's a pretty tall order and to solve that problem pixar developed a universal scene description or usd usd is a system for representing all the elements physics and interactions that make up a 3d scene kind of like how html and css are the building blocks of 2d web pages usd makes it so that one team can work on character models while another team works on environments and so on then the universal scene description itself is where you describe how everything works together well pixar open sourced this usd standard and it's been widely adopted by huge companies around the world for example autodesk which makes 3d modeling programs used in engineering architecture and entertainment has usd integrations so does epic games the maker of fortnite and the unreal engine and apple uses usd to represent all of their augmented and virtual reality scenes so this is already a well adopted standard but nvidia's omniverse is the first application purpose built entirely around this usd system what pixar's shared workflow did for large movie scenes omniverse does for every other industry from architecture and engineering to art and game design that means that one part of a factory engineering team can be working on the digital twins of different robots while a separate team works on the conveyor belt system and a third team architects the building all in different industry-specific programs then they plug their models into a shared scene hosted on the omniverse and if somebody makes an update to one of their designs an omniverse connector would push that update to every scene that that design is used in what this unlocks for businesses is the ability for multiple teams to collaborate on shared 3d scenes at any size and scale i really think that usd and nvidia's omniverse will play a big role in the development of the metaverse but even if you're skeptical of the internet moving from 2d to 3d nvidia's omniverse has a lot of other practical uses in enterprise software today for example connected cars can upload traffic data for logistics and delivery companies to plan and adjust their routes or weather services could upload weather data and models for that same kind of planning in fact nvidia made a machine learning model that emulates the movement and flow of global weather patterns to help predict extreme weather events like hurricanes and atmospheric rivers this is a digital twin of the earth's atmosphere called forecastnet and combined with the omniverse forecastnet can predict the path of these atmospheric rivers up to a full week in advance and it only takes a quarter of a second to run on a single nvidia graphics card that's tens of thousands of times faster than today's standard weather forecast models because these predictions can happen so fast nvidia can generate many thousands of simulations explore many different potential outcomes and quantify the risks associated with these extreme weather events while we often talk about predicting weather and climate up to a week out nvidia's technologies could one day help us predict things months years or even decades out and that's just one of many many practical use cases for these ai and omniverse based solutions so nvidia definitely has some near-term challenges in every market they operate in especially gaming revenue from their gpus and limits on which ai and data center chips they can sell and to who but hopefully this quick survey of nvidia's latest most advanced projects showed you that they might not be the loser that jim cramer is making them out to be and we're sure nvidia is a loser could end up challenging nvidia for the ai throne if you want to learn more about them
Comments 0