저자: Alex · 외국인
요약 보기
엔비디아 2분기 실적(매출 135억달러, 전분기 대비 88%↑, EPS 2.70달러 vs 컨센서스 2.9달러 상회, 매출 컨센서스 대비 21% 상회)을 근거로 강세 입장을 밝힘. 데이터센터 중심 AI 인프라(H100, Grace Hopper, DGX/mgx 플랫폼)의 압도적 우위와 70%대 마진을 강조. 주가가 하락할 것을 예상하면서도 그때마다 매수하겠다는 입장. 소프트웨어(Omniverse)·자율주행·로보틱스 확장 가능성 언급.
다듬은 글
Nvidia stock hit $511 per share last week after reporting absolutely insane second quarter earnings and even though it's already down by more than 10% from there envidia stock has more than tripled this year already so in this episode I want to answer one simple question was this the top for Nvidia stock or does it still have much more room to run your time is valuable so let's get right into it look I try to be as neutral as possible but I got to go on a little rant here don't worry I won't hold you hostage I'll leave a time stamp below so you can skip this entire rant if you want I've been investing for about eight years now and for the last 2 years I've been doing it full-time and there's no doubt that I've invested in some major underperformers over the years but believe it or not Nvidia was one of my biggest I started covering Nvidia stock on this channel in about March of 2022 when I said that Nvidia was the most powerful company on Earth and they will hit a trillion dollar valuation and I've doubled down on that statement plenty of times since then and in True Tech stock fashion Nvidia ended up going down by almost 50% after I made that statement the only thing worse than losing money as an investor is losing money in front of hundreds of thousands of people over the next 6 months I received an endless stream of angry comments accusing me of being on nvidia's payroll pumping the stock and losing tons of money for hardworking people I won't lie 2022 was a really rough year for me and I almost quit making videos alog together but I stuck with it because there's a huge pattern in the market that I feel everybody needs to know about whether it's the metaverse or Gene therapies or self-driving cars every major technology shift that people are talking about has these three things in common artificial intelligence lots and lots of data and specialized compute Hardware whether that means giant data centers full of gpus and accelerators specialized inference chips embedded in drones and robots and self-driving cars or the processors your smartphone laptop and eventually virtual reality headsets so I kept researching these things and now they're the only focus of this Channel and no company was doing anywhere close to the same amount of work in all of these foundational areas as Nvidia so I learned everything I could about them and made over 20 videos covering Nvidia from every angle I could think of even at the cost of covering other stocks my channel is known for like AR invest Tesla and paler and I'm glad I did because Nvidia has literally outperformed Ark K Tesla and paler put together over the last year seriously just go add them up here's what I'm trying to say this channel is about deep research in a very narrow slice of the stock market that I think will outperform the rest the trade-off is that I stick to topics within my own circle of competence and I'm slow to publish content because there's a lot of research and factchecking and what iing so if you want a channel that covers everything going on in the markets this is nit if you want a channel that predicts 20 out of the last three crashes I try not to do that anymore either and if you want me to change my convictions just because a stock goes up or down in price you have plenty of other channels to choose from I'm only going to change my mind on a stock if something comes up in the research changes to the business a specific Market an underlying technology shift or another fundamental change in my overall investment thesis on a company all right that's the end of my rant and lucky for me none of those things happened for NVIDIA so let's talk about their blowout quarter and where things could go from here this was one of the best earnings results of any company in recent history not only did Nvidia shatter analyst expectations but they're expecting their next quarter to be even better Nvidia posted record revenues of$ 13.5 billion which is up a massive 88% quarter over quarter and up over a 100% from a year ago analysts expected $ 11.2 billion in Revenue so this was a massive 21% surprise to the upside they also posted earnings per share of $2.70 which was up 429 from a year ago and up almost 150% from the previous quarter just to be clear their stock price went up 300% over the last year but their earnings went up over 400% in that same time period which means they're technically trading at a lower price to earnings right now than they were a year ago don't get me wrong Nvidia is still highly priced but that's a pretty crazy fact right analysts expected an earnings per share of just $2.9 so this was almost a 30% surprise to the upside this is where I'd usually make a cheeky comment about how Wall Street analysts are really bad at their jobs but I honestly can't even blame them on this one nobody could have predicted the sheer pricing power that Nvidia has right now with their h100 data center gpus their gross margins are over 70% which is a number you typically see in software companies seeing 70% margins on Hardware is absolutely nutty but this is what happens when you find yourself virtually uncontested in the fastest growing Market on the planet as a result Nvidia is guiding for $16 billion in revenue for the next quarter which would be a 15% quarter over quarter increase and a 269 increase year-over-year Nvidia is also guiding for 72% gross margins next quarter so don't be surprised if you see some more crazy numbers coming out of Nvidia in 90 days just to be clear nvidia's insane Hardware margins cannot and will not last forever but if another company wants to disrupt Nvidia in the near- term I think they'd have to provide a huge step up in performance while maintaining a competitive price just like Nvidia is doing versus modern servers right now Amazon Microsoft and Google are all designing their own data center chips today but a lot of these chips have been aimed at reducing the Reliance on Intel and amd's CPUs not so much nvidia's gpus at least not yet but there are only a handful of companies with enough scale to disrupt Nvidia at all let me show you why if you're not a regular viewer of my channel just remember this training is where AI models learn to identify patterns from very large amounts of labeled data like how gp4 was trained on mountains of human generated text Data an inference is where a pre-trained AI model makes predictions on new unlabeled data that it hasn't seen before like when you give chat GPT a new prompt and then it provides you with a brand new answer there's a substantial difference in the compute costs between training and inference to the point where training very large AI models needs to happen on large server systems while inference can one day happen at the edge meaning in laptops and robots and so on but right now both mostly take place on servers which is one reason that nvidia's data center products are in such high demand when I started talking about investing in Nvidia Nvidia had just switched from their A1 100s to their h100 server gpus nvidia's current Hopper h100 data center chips are a massive step up from the A1 100s in terms of performance they're about nine times faster for AI training and the whopping 30 times faster for AI inference but the thing that really sets nvidia's data center ecosystem apart is how it all works together at very different scales eight of these Hopper gpus can be connected to form a dgx h100 system and if you combine nine of these dgx h100 server systems together 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 superod which provides about one exaflop of computing performance and that's just the GPU side there's also Grace Grace is nvidia's First Data Center CPU Grace is an arm-based processor that's designed to handle hand tasks that require a lot of memory bandwidth such as large scale scientific Computing AI training and big data analytics arm is also the name of the company that designs and licenses the arm processor architecture and it's going public in just a few weeks I just put out a deep dive on arm's upcoming IPO which I'll link you here as well as in the description below according to Nvidia Grace can deliver 10 times the performance of normal CPUs when it comes to high performance Computing and AI workloads Nvidia can B Grace and Hopper to form a single package called The Grace Hopper super chip or the gh2 200 which is flexible enough to handle a wide range of data center workloads designed for CPUs gpus or both it's kind of hard to overstate just how big a leap in performance nvidia's solutions are giving to Data Centers over their current configurations over the last 6 months nvidia's CEO Jensen Wong revealed product after product and service after service set to dominate the entire AI staff back for example he showed that if data centers could switch from Intel and AMD CPUs to nvidia's gpus they can do 44 times more work while using less than onethird the power the GPU server is no longer the computer the computer is the data center your goal is to build the most coste effective data center not build the most cost effective server back in the old days when the computer was the server that would be a reasonable thing to do but today the computer is the data center so for $10 million you buy 48 GPU servers it only consumes 3.2 gwatt hours and 44 times the performance let me just show it to you one more time this is before and this is after and this is so if a modern data center switches to nvidia's new Hopper based systems they could do around 150 times more work with the same amount of power or power the same sized large language model for example at just 4% of the cost this is why demand for nvidia's data center products is through the roof right now and it's not just nvidia's dgx h100 GPU systems or the Grace Hopper super chip or even their Envy Linkin infin band networking technologies that companies like Amazon Microsoft and Google need to compete with the thing that I think everyone is sleeping on over this past quarter is nvidia's modular mgx server platform this one design allows for many different configurations of nvidia's gpus CPUs and dpus to go after many different data center verticals like cloud services and gaming high performance Computing big data analytics digital twins for manufacturing and robots telecommunications and of course AI training and inference and because Nvidia will support multiple generations of Hardware in these same form factors they'll be loing in their data center customers for years and years because the costs to switch away from this kind of ecosystem will just be too high nvidia's data center Revenue accounted for over 75% of their overall revenue for the quarter which is why I spent so much time on it in this video another 18% of their revenue or about $2.5 billion came from their gaming segment that's up 11% quarter over quarter and about 22% year over-year largely because of the sales of their RTX 4000 series gaming gpus I actually just made a video about how I think Nvidia is seriously pivoting away from highend consumer gpus to focus exclusively on data centers and Enterprise application that video covers everything I'd want to say about this segment so I'll link it to you right here and in the description below as well the last pieces of the puzzle are nvidia's enterprise software segments which together accounted for about 5% of their overall Revenue but here's the thing Nvidia showed us that they think Enterprise AI software is just as big an opportunity as their AI Hardware business is today software like nvidia's Omniverse which is a platform designed to connect many other professional tools across different Industries together so a scene that was storyboarded with adob tools can be populated with 3D models built in solid works or a 3D asset can be designed in CAD and then put into a scene in Unreal Engine so when one of these tools gets a major upgrade almost every tool in Project connected to the Omniverse can benefit from that upgrade in one way or another on top of that Omniverse uses many of the other Nvidia technologies that we've talked about to enable realistic Graphics physically accurate simulations materials behaviors and even lighting and nvidia's Automotive business includes both software and Hardware as well nvidia's self-driving Hardware platform is called Hyperion 9 and it'll support up to 14 cameras nine Radars three liars and 20 ultrasonic sensors which will allow the rest of the Auto industry to crowdsource many of the challenges that Tesla is currently tackling alone but here's the thing Hyperion 9 is is just one of nvidia's Hardware platforms Isaac is another but for robotics instead of self-driving cars so just like we saw with nvidia's data center stack these Edge Hardware Stacks have a lot of overlapping Technologies making them pretty efficient investments from nvidia's point of view like Drive map which is a crowdsourced real-time digital twin of every road that nvidia's systems get data for by 2024 Nvidia expects to map every major highway in North America Western Europe and Asia that's hundreds of thousands of miles of physical roads that will have digital twins in nvidia's Omniverse then the Hyperion sensors in the car can always double check what they're seeing against that digital twin Nvidia also has Drive Sim which can take in recorded Drive data turn everything in the scene into an interactive object and then Harvest those objects for reuse in further scenes down the road drive Sim also lets Nvidia remove add or change anything in the recording and rerun a bunch of simulations in that environment or they can change the environment all together by adding different lighting effects and weather conditions road hazards and so on nvidia's Drive map and drive SIM can also simulate what different sensors would see in a given scenario so for example they can simulate a city street and then show what a liar would see and simulate the depth map that it would generate that means that they can train their self-driving AI on liar data even though they didn't collect that data in a real City street a lot of these kinds of applications are built in nvidia's Omniverse and that kicks off a major feedback back Loop between nvidia's different business segments because as different Industries demand more and more of these kinds of services Nvidia will need to keep designing better and better hardware for it and vice versa as new applications come online that rely on their Hardware Nvidia will build new software and accelerator libraries to support them and that's exactly why I'm still going to buy Nvidia stock as it inevitably Falls in price from here there's no doubt that nvidia's Hardware margins will go down as they start facing more competition in the coming quarters but they've already identified plenty of software markets to expand into and one day I expect them to have a tightly integrated full stack Enterprise AI ecosystem that no other company will be able to compete with and then nvidia's 70% margins will be much harder to disrupt but there's one more huge AI disruption hot takes on hot stocks ol
Comments 0