저자: TickerSymbol:YOU진행자Alex · 외국인
요약 보기
트럼프 행정부의 중국향 AI칩(H20 등) 판매 재허용으로 엔비디아 중국 매출 회복이 예상되며, 메타 하이퍼리온 등 멀티기가와트 AI 슈퍼클러스터와 구글급 데이터센터 투자 확대로 엔비디아 GPU 수요가 크게 늘어난다고 주장. OpenAI의 AI 브라우저가 검색을 CPU에서 GPU 기반 추론으로 옮겨 GPU 수요를 더 키울 것이라 분석. 엔비디아가 5조 달러 기업이 될 것이라고 낙관. 영상 중 VPN 유료 광고 포함.
다듬은 글
Holy moly. I wasn't going to cover Nvidia until their next earnings call. But there's been so much huge news that this really can't wait. And after everything that happened, I'm convinced that Nvidia will be the world's first $5 trillion company, even sooner than we think. Don't say I didn't warn you.
waste your time. So, here's the bottom line up front. The Trump administration just reapproved AI chip sales to China. Meta Platforms is building an AI supercluster almost as big as Manhattan and OpenAI is launching an AI web browser to compete directly with Google Chrome. And of course, I'll explain how all of this impacts Nvidia's bottom line and what it means for the stock.
There's a ton to talk about, so let's start with the elephant in the room. Up until a few days ago, the Trump administration had tight controls on exporting advanced AI chips to China, focusing on any chips supporting AI workloads in data centers. This included Nvidia's H20 chips, which were designed specifically to meet earlier versions of these export controls.
Then, back in April, the government announced a new set of export restrictions that blocked selling these chips to China altogether, unless Nvidia had a license for every individual order. These restrictions prevented Nvidia from shipping a huge portion of their most in- demand chips to Chinese companies like Alibaba, BU, Tencent, and Bite Dance.
China accounted for about 13% of Nvidia's revenue last year, but because of the ban, China couldn't fulfill these existing purchase orders or take on new ones. As a result, Nvidia lost around $2.5 billion of revenue in quarter 1. They wrote down another $5 billion in unsold inventory and they removed China related revenues from their profit forecasts, all of which I covered in my video on their most recent earnings call.
Then earlier this week, the Trump administration announced that they would grant Nvidia the licenses they need to resume sales to China. Each shipment still requires a government license, and licenses will only be granted for the H20s and RTX Pro GPUs that were specifically designed to comply with these export controls.
So, not Nvidia's highest performing chips. But even with these restrictions, there's a ton of pent-up demand from China's biggest tech companies and cloud service providers. Since Nvidia's hardware and software are much more powerful than what Chinese suppliers can provide. As a result, analysts are estimating that around $8 billion in deferred or canceled revenue will be recognized over the next few quarters.
And total revenues from China could reach 15 to 20 billion by the end of this year. That's around 15% of Nvidia's total annual revenue. But this isn't just about today's revenues or profit margins. The global artificial intelligence market is expected to more than 8x in size over the next 8 years, which would be a compound annual growth rate of over 30% through 2033.
And around 50% of all AI research happens in China. So the Trump administration's trade restrictions were locking the world's biggest AI infrastructure company out of the world's biggest AI market, which forces China to build their own solutions. While that might delay their AI progress in the short term, it also forces China to innovate and build their own independent AI and semiconductor supply chains over the long term, which makes them much more resilient.
So, this isn't just a huge catalyst for Nvidia, but for the entire AI landscape since it brings more research and advanced computing into Nvidia's ecosystem, adding more AI developers, solutions, and innovations over time. That's why paying attention to AI policies in every major market is becoming more important for investors who want to get rich without getting lucky.
But China's AI chip market isn't the only thing we should be paying attention to. Just one day later, Meta announced plans to spend hundreds of billions of dollars building massive multi-gawatt AI superclusters as they race toward super intelligence. The first supercluster is called Prometheus and it's expected to come online in 2026 with another one called Hyperion coming online by 2030.
This Hyperion data center is one of the largest infrastructure projects ever attempted in humanity's entire history. Its planned area is almost 23 square miles, which is roughly the size of Manhattan, and it's expected to reach 5 gawatt of power when it's fully built out. That's enough electricity to power around 4 million homes.
Meta Platforms is deploying around 600,000 Nvidia H100 GPUs. And both Prometheus and Hyperion will rely almost entirely on Nvidia's AI infrastructure for model training, including their Blackwell B200 GPUs, their Grace Blackwell superchips, and their Envy Link and Quantum Infiniban networking solutions.
While Meta's training and inference chips and AMD's GPUs will have smaller roles supporting much more specific workloads, that means most of Meta's huge AI spending spree will go directly to Nvidia. And I wouldn't be surprised to see orders worth tens of billions of dollars in the coming quarters just to support these two superclusters.
On top of that, Meadow Platforms increased their data center capex to $60 billion in 2025, which is a 50% increase year-over-year, with most of that money aimed at supporting powerful AI models trained on the mountains of data from their 4 billion monthly active users. And speaking of data, making it look like you're somewhere else.
But did you know that VPNs can also save you a ton of that and your entire family. But some . So, you can also use Surf Shark to . And since VPNs change your digital location, Surf Shark lets you access content, websites, and services that are usually blocked in your region altogether. I do this all the time when I travel.
since Surf Shark offers a today. All right, this isn't just about Mark Zuckerberg spending an insane amount of money on AI. It's about Meta's race to solve super intelligence, causing other tech giants to enter the race as well. For example, Google just announced plans to build $25 billion worth of AI data centers over the next 2 years.
And Microsoft, Amazon, and Oracle all increased their data center investments as well. All these massive companies rely on Nvidia's tech stack to power their AI workloads, which means the competition between them will turn into even more revenue for Nvidia. Another trend this is setting is the shift to multi-gawatt AI superclusters, which means tech giants will need to secure long-term energy contracts and invest in their own sources of power.
So, one area of the market that I definitely want to do more research on is power generation for data centers and high performance computing. I also think this proves that the massive rush to build AI infrastructure isn't as temporary as most Wall Street analysts have been suggesting over the last 2 years.
In fact, the opposite happened with most AI infrastructure investments actually increasing year-over-year. That means we should expect Nvidia to report record-breaking data center revenues, not just on their next earnings call, but for many earnings calls to come. Meta's Hyperion and other planned superclusters will use hundreds of thousands or even millions of NVIDIA GPUs over the next 5 years.
And it seems that top tech companies are willing to spend unprecedented levels of capital to dominate the AI market or at least not lose the race. The global AI data center market is expected to almost 9x in size over the next 9 years, which would be a compound annual growth rate of 27% through 2034.
And as AI models become bigger, more complex, and more specialized, data center operators will need to upgrade their hardware faster than before if they want to stay competitive, which will drive up this growth rate even further. And the fact that they're all turning to Nvidia's Blackwell architecture means that Nvidia still has serious pricing power, which wouldn't be the case if there was any meaningful competition from the hyperscalers's own custom chips or from companies like AMD.
But it's not all about the AI race between the Magnificent 7. This third catalyst shows how even small companies can change the entire AI landscape altogether. OpenAI is launching their Aura browser to their 400 million weekly chat GPT users, directly targeting Google Chrome and the quarter trillion online search and advertising business that makes up almost 3/4 of Google's total revenue.
Google Chrome status as the internet's default browser means that it covers more than 65% of global desktop usage and runs over 90% of its search and ad-based revenue to Google search. And Google Search handled over 5 trillion searches in the last year, most of which drove ads and dollars by retrieving information on the internet and returning it to the user.
That part is obvious, but what may not be so obvious is that information retrieval is a task that's mostly powered by CPUs today. On the other hand, Aura will keep their user activities like search, web navigation, and shopping inside a chat GPT- like interface, which means they'll shift from Google's CPUs to Nvidia's GPUs.
But it wouldn't take a lot of traffic to move from Google to Aura to make GPU demand explode. Let me explain why. Generative answers come from large language models that run trillions of parameters every session. Which means that the computational cost goes up for every search that moves from Google search to OpenAI's Aura.
For example, Google's initial AI powered summaries drove up their compute costs for each search by 7x compared to classic retrieval. And generative AI agents use even more compute resources than that. Back when I covered the breakthroughs in Deepseek R1, I talked a lot about something called test time scaling, which is where reasoning models spend more tokens on inference by breaking complex prompts into steps, thinking through them step by step, using chain of thought reasoning, double-checking their work, or even working with other models to find solutions.
So, if just 10% of last year's global search traffic moves from CPUs to GPUs and they take 10 times more compute, the demand for Nvidia's GPUs would go up by the total compute of all of Google searches. And I think that 10x compute estimate is actually conservative because OpenAI's browser is expected to tightly integrate AI agents that can handle much more complex workflows right inside the session, not just basic search.
Think about multi-step tasks like filling out forms, doing shopping comparisons between different items and vendors, and scheduling appointments and making reservations. Each added step takes more inference tokens for the AI model to complete, which exponentially increases the size and cost of the total workload.
And of course, all of these workloads are designed, optimized, and deployed on Nvidia's Blackwell and Hopper architectures today since Nvidia has over a 90% share of the data center GPU market. So, as web browsing and search move from Google's retrievalbased model to generative AI browsers and inapp assistants, the cloud services that they run on will need to increase the ratio of GPUs to CPUs even more to support all these new inference workloads, which will turn into tens of billions of dollars for NVIDIA that analysts weren't expecting.
And this isn't just about OpenAI's Aura browser. AI search competitors like Perplexity are also gaining traction, while Google's online ad share is projected to fall below 50% for the first time in years. The global AI browser market is expected to almost 13x in size over the next 9 years, which would be close to a 33% compound annual growth rate through 2034.
That includes AI powered web browsers like Aura, browser extensions that can do things like compare prices across existing websites and integrated AI agents that can do specific online tasks for the user like make purchases, fill out forms, and make reservations. So, all of these AI powered browser tools really just need to take a few% of Google's market share combined to skyrocket demand for AI infrastructure in these markets thanks to those workloads taking so much more compute.
Nvidia stock popped by around 6% this week so far. But I think a lot of Wall Street analysts are still missing the bigger picture here. American AI chips re-entering the Chinese data center market. American companies spending tens of billions of dollars on city-sized superclusters and AI browsers shifting search from CPUs to GPUs aren't one-off events.
They're massive long-term tailwinds for Nvidia and honestly the rest of the AI market. which i content like this. Either way, thanks for watching to the end, even though I gave you everything upfront. .
Comments 0