저자: Henry · 외국인
요약 보기
미국 대형 기술주(구글·아마존)의 AI 수익화와 클라우드 성장이 강하다고 보며, NVIDIA·TSMC 등 AI 공급망 전반이 상승할 것이라 주장. 2026년 자본이 AI 칩으로 계속 유입된다고 봄. NVIDIA는 설계 강점과 CUDA 생태계로 지배적이나 구글 TPU 등 자체 칩이 점유율을 잠식할 수 있다고 언급. 대만 공급망(TSMC, 패키징·테스트 기업) 중요성 강조. 3076(대만 관련 종목) 상승을 근거로 AI 관련 ETF 매수를 권유. 자막 품질이 낮아 일부 내용은 불명확함.
다듬은 글
Recently, the US stock super treasury, US stocks in both Google, Amazon have to significantly profit especially they take out AI applications to make money, Google Google results bright GoogleCloud's business continues to strong competition long. NVIDIA TSMC ASML three SK Hiness these all rise to stupid.
Connect these 4 technology heads this year 2026 capital began to exceed the dollar, continue to buy chips inAI I brother we all know is the big winner of this AI competition, because he holds the strongest fine chip, basically companies develop AI to find him to buy the development of AI formless companies steal or grab or good , have to NVIDIA tablets back to AI, but feed, why the world must buy chips with him?
These big brothers billions of dollars out NVIDIA how from scratch became Zhang Jing cheated out, then become one by one, so that the big brother AI typeless, and do the center of cloud services number? In fact, although NVIDIA plays a very key role, this side of its entire AI park accounts for a small part .
And NVIDIA characters may be broken soon , this video will be fighting Gold into an AI supply. There are some big brothers with hundreds of millions of dollars out to start an AI fine bullet body design production spoon AI fast assembly and then one has one AI data center practical application, which will be a step sympathy is what it is like ?
What companies are involved? How do I find the investments of other new companies on this year? In particular, we may have been overlooking the very relevant role of Taiwan in the AI work year . I Henry is a pig investor who operates certificates, my goal is to do investment knowledge, this video is personal sharing and use Financial advice said investment recommendation, I am not self-licensed financial questions, any investment decision please research and professional investment risk Past performance does not mean future gold may lose.
First of all, we need to know where the big brother even used over a billion dollars to start using the so-called development of AI actually used to go, mainly actually use all the starting center, inside this track should not be, only one and another AI vision, day and night people, constantly transporting training their own AI models such as GoogleGermany or renting this building to the AI OpenA or Mom, to help my customers train their AI without type, so this one has a building of AI servers this, it is in practice, over a hundred billion dollars of capital every year starts to go places.
The most core and most valuable of these AI volts is likely to be NVIDIA's latest super- crystal chip GB2, which combines two state-of-the-art chips inside , designed for AI modelless transport. So we go back to the most basic question is why AI typeless training and operation must be done with GPUs, but also must be done by NVIA?
If you buy a computer to play games, you will certainly have heard of CPU and GPU. In fact, they are the brains in the computer that are responsible for helping to process data, but its working mode is completely different. CPU, you consider a good mathematics professor, he can handle extremely repetitive work, such as calculating those calculus or manage the whole computer operation, but this professor one, there are shortcomings directly limited right.
What if you gave him a simple number like one plus one ? Even if he counts, he counts one by one , one by one, so with good time, he can count these 10,000 one plus one such simple numbers, and when he is 10,000 primary school students, he does not know how to trace the score, knows the simplest addition and subtraction, but his advantage is that there are many people, there are thousands of deep cores in GP.
If I give 10,000 one plus his GPU that is these 10,000 to one person count one together while counting below, then maybe within a second this should be the same time sun these 10,000 one frame. So why must AI operations use GPUs? Because AI is in the exam and learning, it is not like your outside human body research relative so, it only needs to process millions or even tens of millions of very simple data at the same time learning this kind of extremely large number, difficult to return to work, if you find me Stan this immediately CPU, he is very ਧਰ and Yu enter that black this can not drop the cow, but if you synchronize it, you can pay 10,000 GPU together Okay, so before the promotion of AI must be to rely on GPUs to conduct unmanned training So, why can AI be like this ?
In fact, it is by rote memorization of Kailiang's data, AI, is by the body exposure of all human beings all the picture alliance learning For example, you rely on AI to recognize what, to pick up 10,000 cats to see it so that it recognizes that the ears are pointed, no cat 's so CPU what is a cat, a person can sit one by one photo, oh this one is a cat good next one this one , so I'm a cat.
Cat is divided to 10,000 people at the same time to see this photo so he 10,000 bodies together to work, will hold a photo sound this second finished frame, because AI learned knowledge is so calculated, in the face of such a huge data, CPU 4 six, but GPU this crowd tactics. NVIDIA is definitely the most recent overlord, AMD and because now at the moment, the series NV's rear single light can not be seen.
Feed, no one can draw, today's draw less right. Make NVIDIA can market , even Elter to table car clubs than point owners 3 reasons. The first is the chip performance is far beyond its rivals , the craziest is MVIA's own fine chip, each generation is more crazy than the previous generation, its 2024 bedwar reasoning ability is thirty times the previous generation brand, coming this year will release a generation of bed 5 times, but also save 10 times The second is IA, he has opened this newspaper office a few years ago to put together a system called uda, all AI-related programs, basically use it to write, which is the official language of the AI industry , so someone may ask, feed or not use MV, can not, but all A willing to re-year, like today , more than suddenly announced, tomorrow, Hong Kong all cars become all left face and roundabout ah originally look down, you look down , say you drove Farah your Benz all immediately greedy re-buy that step, not the old tattoo D 囉 so Work City for a thought of the millions of red rewritten again, of course not so many brothers, rather, row to NVIDIA .
So thirdly this is a GP2 very powerful right can help you turn 36 GP2 into a complete view, these 36 GP2 is completely connected, into a super invincible ONC tank. The luck of your MicrosoftAmazon is talking about the parameter meter, my Neder technology can help you turn the whole datacenter, maybe say a GP2, all into a single shadow.
Its chips shipped to is fast, and can grie, in F-B together with the horse flew the big guy, and then after the big lining of technology, it just said tied to NVIDIA body change. This is the reason that created the status of the NVIDIAAI era. Moreover, Nvida can wake up, it is only responsible for the structure of the grain , and then uses programming languages to technically weigh the world.
He puts out the most troublesome production and makes the most easy and biggest money himself. So you look at the latest GB 200 that structure, the core is two BetrillGPU, and then take another, is the design battery made, the world only TSMC can be so mature so new production of it, they are the only kitchen in the world can inject the most crystal chip, NVIDIA is very strong ?
It's very powerful, he designed the strongest chip out, has a strongest recipe, but, every time he himself is without a kitchen, must rely on the left kitchen to help him fry stack sent out, when GoogleMeta these rich road this, carrying a robbery a dollar came in and said, feed and I take all out , Nvidia can only ask how to kitchen, help OT or pull back two partners Come with it first, so NVIDIA is the design, but TSMC is the world to master the most advanced technology, the only mature ability to speculate.
We can speak well. Feed feed, but the whole chip, to operate to still have many very relevant parts of the , one of which is open split note h, that is, your brain is good very fast, wow so good after this you do not not low, the brain, can not remember anything of so no meaning. HBM This is to note the sun is being shipped, or the information after shipping.
This HBM is SK Hiness 3 and MU Meiguang from the three major suppliers in the world. Each GP2 2 needs to be matched. So we understand SKHin Samsung exploded all, connection and CPU first communication chip control computer yeah, cooling is good parts of together put all the hair on the motherboard to make a seal can become an original GB2.
But before the wind this has a very important and important step, is to test, hey, I do not know what we have crystal chip energy rate chip this thing is very complex, is extremely high-tech process, a little bit of the department, if you have not tested the can not use the next pen to find can not use then other pieces must be reported Level test equipment, for you TSMC such a big thing, said the world's leading only its a state-of-the-art 2 cubic meter 3 cubic meter chip, but I 'm sorry, before its chip was invented, it was to pass three hurdles, find Moonlight to do Sharp Electronics using a technology called Cows, is leading the all-crystal chip GB2 parts, using 2.5G 5G technology can do the whole world.
After the wind is completed and have to do another test the world is the most advanced and most important crystal chip production base , it holds the world's most advanced manufacturing most of the production capacity NVIDIA all the top AI fine chips are electric, and pre-defined design, because a fine chip is not just the core and CPU.
Inside and CPU paper communication fine film and external communication, power switch film many of these are Taiwan companies designed and production is patented, such as joint development, testing, etc., so the MED most powerful GB200, in addition to the face and CPU, is their own design ? Other things can not yeah and not very familiar, straight this in back to Taiwan, alas open D yeah let us come.
So Google right Amazon, they arrived so early to match so much before Nvidia earn, no one wants to let Nvidia drive the spring bag, and have to be small to buy your video, for you nvia does not care the rate to 70 more than a fine video you buy, 70 is profit frame, guys banknotes what is the difference, feed you NVIDIA design , we Google anyway you know your own design NVIDIA in addition to वाटGP own design , other fine film yeah, power chip design of the series generation, production of TSMC, testing of the red disaster days and months over what we do not accumulate Taiwan?
So Google 2013 has started their own research and development has gone to the eighth generation TPUT and 8I latest generation TPU is Taiwan old design electrical products, also use technology installed , chip testing is Taiwan old detection technology Amazon Microsoft have developed their own call to invest in the Amaz built a large number of their own traum sperm made of the number center, inside the chip this NVIDIA does not have, all developed traum2.
And there is a very important thing about Google, very funny, [ __ ] found that the CPU they developed this training AI and AI pushed to this, not only cheaper than much, but is much faster. The video mentioned it , CPU, very lack of what, it's a doctor , but he only respects to do one thing at a time.
GPU this is different, it can be distracted and do more things , you consider it the same price, and hire a PhD form hundred turned over , although not so beautiful, but I am not to do some very strong buried feed, research what guys? Training AI to keep looking at data count, TPU kind? Feed, brother, anyway, it's a heavy repetitive counting thing, no matter how stupid it is, I directly do not want this big burst of life, the same not bud no, I please save 10,000 hours back.
That is actually non-stop one squeeze, do not have to buy something DSD-specific all Formel 380 dollars a counting machine, one step, Japan is 30 dollars. Well, that's all you are, a couple of sleepy eyes, no good. There are reports, Google's TPU push you cost much cheaper than Nita, Met this big innovation this has begun to use Google's TPU to find short- term services, Google find this group to help NVIDIA do Taiwan cooperation, from the original NVIDIA buy, become directly to the unit to find people to do, it not only obtain the chip and deploy AI said the cost, there is a chance to decrease, but also the opportunity to use T to lower the cost of the people NVIDIA's chips will not disappear in the future I do not know, maybe the first generation Rubin continues to build practical chips, but whether Nvidia continues to dominate chips mazon share more marketshare is good production of it is looking for investment and other advanced buy, chip design chain this is very important, Investment Semiconductor is the most familiar, the most leading advanced film production factory.
Whether Nvidia or Google, Amazon is very advanced chip production this does not open its , joint release this is involved in the development of Google TPU. Chip power management, whether NVIDIA or Google must use it, moonlight is it is doing the world's largest wind and testing company, TSMC's order , often will not produce enough, sent ASE road done, powerful equipment related new technology, it has the most advanced data center optical communication technology in trying to break the speed and loss limit of money transmission.
The original number of center chips between this , to and to transmit, but the same line, the speed is not fast enough , the selection number is too fast, so this has begun to use the fiber frame. And the brand new technology possesses some of these fiber optic technologies. Recently this year rose, 3076 stars left even stronger than SOX, 3076YTD stars 86%, so left 66%, against the month 3076 rose 60%, SO6X rose 21%, today we see a short-term concept, and is a global hundreds of billions, even trillions of dollars capital began to promote the AI industrial revolution What about the long-term investment in the entire supply year?
Which ול give D's returns this may be very strong, just like I have been sharing over the years that I like to buy ETFs so can the whole theme whole sector buy up all, 30765 is not just buy a company, but the most verified core Taiwan in the AI era , of course, not asking you to buy 3076 , but if you still really worry about when AI will disappear or feed the industry , let's think about whether you were involved in the whole world of the AI industry.
As the saying goes, you are not enough, you can join them . That those mentions are the same, that is package one, so I think now something says must. Limited editor, I have prepared, 3076 this Taiwan see Eff's detailed introduction in this top ten constituent stocks, daily business doing what , relying on what to make money, what competitive advantages, as long as you subscribe to my channel, the following message 3076 will be sun to you With two financial institutions the cooperation of the free financial investment course, inside 61 topics, salty economy, investment, completely free , and the following message 3076 at once ga, today said so much first, like, remember to help, share and most importantly to the website 25 degrees I have YouTube not said things, we see you next time, bye
Comments 0