저자: TickerSymbol:YOU진행자 · 외국인
요약 보기
엔비디아는 데이터센터 매출 급성장(분기 매출 260억 달러, 전년比 +262%, 데이터센터 비중 86%)과 높은 마진(78.4%)을 기록했다. 주가는 1년간 약 180% 올랐지만 EPS 증가율이 더 커 PER은 낮아졌다고 주장한다. 차세대 칩(Blackwell 이후)과 이더넷 네트워킹으로 TAM이 확대되며, 공급 제약이 지속되고 경쟁 위협은 제한적이라고 본다. 10대1 액면분할(6월 7일)과 다우지수 편입 가능성을 호재로 꼽고, 진행자는 엔비디아를 최대 비중으로 계속 매수할 계획이다.
다듬은 글
Nvidia is the most powerful AI company on the planet because they make the most powerful AI chips on the planet almost every AI service runs on nvidia's infrastructure today but after announcing a 10 for one stock split on one of the best earnings calls in recent Market history Nvidia stock blew past $11,000 per share and whenever I see that kind of grow this finally the top for Nvidia stock or are their waste your tim walk you through nvidia's insane earnings results focusing on data centers I'll highlight their current product road map including the next chip after Blackwell I'll talk about nvidia's biggest risks right now like their competition and Supply constraints and of course what I'm doing with my Nvidia stock as a result especially with a 10 for one stock split on the horizon all right let's Dive Right into to nvidia's earnings Nvidia posted record revenues of $26 billion for the quarter which is 18% higher than their most recent record breaking quarter and up by an insane 262 year-over-year Nvidia also posted earnings per share of $6 12 which is 10% higher than analysts expected and 7 and a half times higher than a year ago things get even crazier when we look at nvidia's data center revenues which now account for 86% of their total revenues today up from about 60% of their total revenues just 1 year ago nvidia's data center revenues came in at just over 225 billion dollar which is an increase of 22% quarter over quarter and an insane 427 year over-year that means Nvidia is now on Pace to make a hundred billion in AI accelerator revenues this year alone while AMD is guing for around4 or5 billion and Intel is well I'm just kidding I'll cover nvidia's competition later in the video but there is something important that you should know about these crazy year-over-year numbers I know it's hard to imagine but Nvidia was actually struggling pretty badly a year ago if we look at their revenues by business unit we can see that their gaming business which was their biggest business unit at the time was in freef Fall I don't just cover the good parts of my investments so I made videos talking about the issues with nvidia's consumer gpus the overall weakness in the PC market and the resulting collapse of their gaming segment as they went all in on Data Centers in quarters 2 and three of last year a collapse that nvidia's gaming segment still hasn't fully recovered from while their data center revenues have more than 5x my point is these low numbers from a year ago really help make nvidia's growth look unbelievably good today but since their insane growth started four quarters ago now we should expect nvidia's year-over-year growth to start slowing down as early as their next earnings call where the comps will be against much better quarters so don't say I didn't warn you 3 months from now Nvidia also reported gross margins of 78.4% for reference a software as a service company typically has gross margins between 70 and 85% so Nvidia is making software likee margins on their Hardware right now but this is what happens when one company dominates the single fastest growing Market on the planet nvidia's gross margins are up 2 and 1/2 points quarter over quarter and a whopping 14 points year-over-year see what I mean about these year-over-year numbers this crazy growth is why Nvidia stock is actually cheaper today than it was a year ago wait the stock is cheaper now even though it's near all-time highs let me show you what I'm talking about Nvidia stock price is up by around 180% over the past year but their earnings per share Rose by 650 which means nvidia's price to earnings ratio is around three times lower today than it was a year ago in fact fact Nvidia stock is trading right around its 5-year average price to earnings ratio even though it's growing much faster today and has much higher margins now you know why Nvidia stock is actually cheaper today even though it's at all-time highs you probably also know that vpn's protect you and your online data by encrypting it and making it look like you're somewhere else but did you know that vpn's can also save you a ton of money that's where surf shark comes in the sponsor of this video surf shark is a super fast and easy easy to use VPN that you can install on unlimited devices with one account which means you can protect yourself and your entire family and since vpn's change your digital location you can also use surf shark to access websites content and services that are blocked in your region even better some online stores raise their prices based on your location so you can use surf shark to always get the best price on Big Ticket items like plane tickets hotel rooms and rental cars so if you care about your family privacy and you like saving money you can use my code ticker to get up to four extra months for free and surf shark offers a 30-day money back guarantee so there's no risk in trying it with my link in the description below today and a big thank you to Sur shark and to you for supporting the Channel all right so far we've looked at nvidia's top and bottom lines their year-over-year growth and why the stock is cheaper today despite trading at alltime highs let's dive into their product road map next since that's the real science behind behind this stock last year Nvidia accelerated their Hardware release cycle from every 2 years like they had between the a100 and h100 data center gpus down to one year between the h100 the h200 and now Blackwell which ships at the end of this year if you want to know where we are on this road map right now Jensen just hand delivered the world's first dgx h200 system to open AI this past April as you can see Nvidia has another yet to be named chip planned for release in 202 5 along with big upgrades to their Quantum infiniband and Spectrum X ethernet networking Solutions I'm showing you this today because during this past earnings call Jensen confirmed that Nvidia is now fully on that one-year Rhythm he announced that there will indeed be another chip after Blackwell and that Nvidia will release a whole set of ethernet-based networking Solutions what that means for investors is that nvidia's total addressable Market is growing massively since ethernet-based data centers can now have their cake and eat it too they can deploy nvidia's Hardware without having to switch to infiniband here's what nvidia's Chief Financial Officer Colette Crest said during the earnings call in the first quarter we started shipping our new Spectrum X ethernet networking solution optimized for AI from the ground up it includes our Spectrum 4 switch Bluefield 3 dpu and new software Technologies to overcome the challenges of AI on ethernet to deliver 1.6x higher networking performance for AI processing compared with traditional ethernet Spectrum X is ramping in volume with multiple customers including a massive 100,000 GPU cluster Spectrum X opens a brand new market to Nvidia networking and enables ethernet only data centers to accommodate larg scale AI we expect Spectrum X to jump to a multi-billion dollar product line within a year but the real money maker isn't this next GPU or even these ethernet offerings it's how they all fit together which is this GX200 nvl at the very top of this road map and to understand why we need to talk about Blackwell since Nvidia will start shipping it at the end of this year that I focus on investing in technology platforms that other businesses pay to build on top of Blackwell isn't just a GPU it's an accelerated Computing platform and chips based on the Blackwell architecture are going to go in self-driving cars machines and factories humanoid robots and of course inside data centers and supercomputers and just like Hopper Blackwell has two variants the B100 and the b200 the B100 is a drop in replacement for the h100 meaning data centers and supercomputers can swap out one h100 tray and put a B100 tray in its place so Blackwell won't have a hard time ramping up in sales since the infrastructure to support it is already deployed anywhere there's an h100 today the b200 is the state-of-the-art Blackwell chip just like the h200 is for the hopper generation Blackwell has much higher performance per watt compared to Hopper and since power is often a bottleneck for data center operations they're going to want to phase in more Blackwell chips over time in fact Blackwell is four times better at AI training and a whopping 30 times better at AI inference which is what gives Nvidia so much pricing power over their competitors and why their margins are still currently so high the next level on this road map is the gb2 200 super chip ship which is what Jensen is holding in his left hand during the Nvidia GTC keynote while he casually holds the10 billion prototype board in his right this is a fully functioning board and I just be careful here this right here is I don't know 10 billion do yeah I totally would have dropped that anyway the gb200 super chip connects 2 b200 gpus and one Grace CPU and this connection is important they're currently connected via NV link which has up to 900 GB per second of bandwidth that's fast enough to move 150 feature length 4K movies between chips every single second but soon these AI chips could be connected by Spectrum 4 ethernet switches instead like nvidia's CFO mentioned on the call whether these chips are connected by nvlink or Spectrum 4 switches 2 gb2 200 Grace Blackwell Super Chips go into one Blackwell compute node which is one tray in an AI data center rack coat Crest also mentioned these Bluefield dpus or data processing units on the earnings call dpus handle most of the networking workloads so that the systems CPUs and gpus don't have to 18 of these compute trays go into one gb200 nvl 72 system it's called that because the gb200 super chip has two blackwall gpus there are two Super Chips per compute node and there are 18 nodes per system all all connected via NV link 2 * 2 * 18 equals 72 blackwall gpus connected by NV link hence gb200 nvl 72 now stick with me because we're almost home sandwiched between the 18 compute nodes are nine Envy link switch trays with two Envy link chips each connecting four Blackwell gpus so four ports per nvlink chip time two chips per tray time 9 trays is also 72 ports supporting all to- all GPU Communications at a mindboggling 130 terabytes per second of bandwidth that's the back that's the that's the back the dgx MV link spine 130 terabyt per second goes through the back of that chassis that is more than the aggregate bandwidth of the internet we could basically send everything to everybody within a second more bandwidth than the entire internet now you understand why Nvidia doesn't have any competition and why being able to support ethernet-based data centers is such a big deal the 18 compute trays and the nine nvlink switch trays come together to form the gb200 nvl 72 system but that's just one Rack in an AI Factory this one rack has 1.4 xof flops of AI training performance and just to be clear there are only a few exaflop supercomputers on Earth today and Nvidia just built one in a single 120 KW rack and at the top of each single rack supercomputers exists a single networking tray to connect multiple racks together today that switch tray uses infiniband but soon it will be offered in Ethernet as well all right now that you understand the different parts of the Blackwell platform the B100 and b200 gpus the grace CPU the gb200 super chip that connects them the EnV link or Spectrum 4 chip to chip connections and either infiniband or Spectrum X to connect the racks you also understand that Jensen announced way more than one new chip on this past earnings call this new GPU will also have a new Grace x super chip new chip-to-chip switches and new networking solutions to connect the racks together and just like the B100 trays are dropin Replacements to the h100 trays it seems reasonable to expect that the next one will have a dropin replacement version as well so as more data centers by Hopper or Blackwell systems they're already on the upgrade path to this next GPU platform over time I know we spent a lot of time covering nvidia's data center systems but remember they account for 86% of nvidia's total revenues today so you just saw most of the science behind the stock at least by Revenue I haven't even touched the software side of things which I'll cover in a separate video now let's talk about the risks the 10 for one stock split and what the heck I'm doing with all this information since Nvidia is my biggest position by a pretty big margin at this point the risks to holding Nvidia stock which surprise surprise really means the risks to their Data Center business let's start with their competition Nvidia has two kinds of competition chip makers like AMD and Intel and hyperscalers that make their own chips Amazon Microsoft Google but here's the thing when most people buy a car or a phone they only own one at a time so every sale for one brand is a missed customer for the others Tesla versus GM Apple versus Samsung data centers don't work like that when Microsoft buys a GPU from AMD or uses their homegrown accelerators that's not one less sale for NVIDIA because data centers need to handle a wide variety of workloads as a result data centers are actually portf folios of different hardware and software Solutions designed to meet their customers needs as demand for a specific kind of workload grows cloud service providers will buy more of the right hardware and software to support it but in the end data center operators are always trying to deploy the right mix of systems to optimize their costs most workloads still don't involve AI today which means Nvidia systems would be total Overkill so data centers will buy chips from AMD and Intel for certain tasks they'll build and use their own chips for others and use Nvidia systems for heavyduty training and inference applications and make no mistake inference is becoming a heavyduty AI application as the industry moves from predicting the next word in a sentence to generating frames of videos or structures for novel proteins so nvidia's quote unquote competition might be able to handle inference when it comes to large language models but their performance remains to be seen for text to video models like open AI Sora or protein in models like Alpha 3 by Google Deep Mind one of the other big talking points that AMD and Intel had was their support for ethernet while Nvidia uses infiniband but as you just saw Nvidia is rolling out ethernet based Solutions as well which cuts into AMD and Intel's piece of the pie not the other way around that means worries about decreasing demand are equally misplaced the market for AI infrastructure is expanding as the world moves from retrieval-based workloads like delivering you to a pre-made piece of of content to generative workloads like creating an answer or a website or a video or a protein or a robot's next move based on your exact query and your exact preferences so inference workloads will keep needing more and more powerful chips as they keep getting more complex Nvidia made it clear on the earnings call that demand for their Hopper gpus went up after Blackwell was announced not down and I've shared chart after chart in video after video showing how crazy fast industry analyst and experts are expecting the AI Market to grow for the rest of this decade and Beyond and that just leaves us with nvidia's Supply constraints as long as Nvidia makes such huge leaps in compute power they're always going to be Supply constrained that's the risk with making something that every business wants while wait times for the h100 eventually went down from over a year to just a few months Nvidia is currently Supply limited on the h200 and even though Blackwell is now in full production demand could exceed Supply well into 2025 and by the time those wait times come down I expect their new chip to be Supply constrained as well the truth is I'd rather see Nvidia run out of Supply than run out of demand and I know they're working closely with tsmc and the rest of their suppliers to meet as much of that demand as possible when it comes to their supply constraints at this point I think nvidia's leadership has earned my trust and my money speaking of which here's what I'm doing now that there's a 10 for one stock split on the horizon in theory stock splits should shouldn't matter Nvidia will have 10 times the shares at 1/10th the price and that's that but in reality studies show that companies that announced stock splits tend to rise by an average of 25% over the next year versus about 12% for the rest of the S&P 500 one of the big reasons for that outperformance is that many brokerages outside the US still don't offer fractional shares so nvidia's share price going from $1,000 to 100 lowers the barrier of Entry tenfold for those investors another thing that nvidia's stock split might enable is their inclusion in the Dow Jones Industrial Average which is a price weighted index not market cap but stock price the Dow Jones tracks 30 large cap companies that account for a significant portion of US economic activity it currently includes companies like Microsoft Salesforce Apple Amazon and Intel in my opinion Nvidia is a great candidate for this index which means Nvidia does stand to materially gain from this stock split which takes place on Friday June 7th between that confirmation of their next chip after Blackwell and video rolling out ethernet-based networking Solutions and their price to earnings ratio falling over time I'm going to keep buying Nvidia stock to get rich without getting lucky so congratulations to my fellow Nvidia shareholders and thanks for watching
Comments 0