저자: JimCramer · 외국인
요약 보기
엔비디아의 AI 추론(inference) 시장 대응 전략을 분석한 영상. 구글 TPU·아마존 트레이니움 등 ASIC 경쟁이 부각되며 추론 시장 중요성이 커졌고, 엔비디아는 그록(Grok) 인수(약 200억 달러 규모 기술 라이선스/인재 영입), 인프라브릭스 네트워킹 기술 인수, AI21 랩스 인수 협상 등으로 대응 중이라 주장. 엔비디아 GPU 점유율은 약 93%로 제시. 주가는 10월 고점 이후 약 6개월간 횡보 중이며, 잉여현금흐름 약 800억 달러(연말 960억 달러 전망)로 M&A 여력이 크다고 언급.
다듬은 글
This is a very big inflection. You know, the last few years have been all about building this pre-training. We've seen Nvidia dominate the market, more than 93 94% of this GPU space. But then over the last part of 2025, we heard a lot about Google and its TPU and Amazon and Tranium. And Jensen had sort of brushed it off at different times saying, you know, these XPU projects are, oh, they're a year behind.
They're two years behind. I think the Grock deal was the first time that you really saw Jensen acknowledge. you know, he he's always acknowledged the importance of inference, but acknowledging Jim Kramer describes a turning point for Nvidia. This is a significant inflection. Building this pre-training has been the main focus of the past few years.
You know, Nvidia now controls over 93% of the GPU market. However, we heard a lot about Google and its TPU, Amazon, and Tranium in the latter half of 2025. At various points, Jensen had dismissed it, stating that these XPU initiatives were, "Oh, they're a year behind. They're two years behind." I believe that Jensen's first meaningful acknowledgement was during the Grock deal.
He now consistently recognizes the value of inference while admitting that it might be an architecture game or a talent game. We are aware that memory is quite limited, and Grock's method of explaining why inference alters memory in a very special way. I believe Jensen. You often hear Leslie say, "Oh, Jensen is playing 40 chess.
He's playing it with talent." In my opinion, I believe he is manipulating the supply chain. He understands he has a sizable market share, but he still needs to protect it. Therefore, I believe he's playing it on the inference economics. Although it is short game and a long game, I believe these moves are good.
Additionally, I believe that he is currently playing a longer game. Yes, the talent component is particularly intriguing since R&D is frequently the reason why a corporation tries to create a competitive, you know, mode around a certain business. The product is to blame. Talent is a bit more difficult.
Since you are working with people, you must people are free to depart. People need to be paid a lot. Furthermore, it's frequently challenging to determine the potential value that a human could add to your company. As an investor, how do you evaluate this approach? Particularly when you're discussing hundreds of employees and their skills and how they'll deliver those skills for a huge corporation like Nvidia, the trillion dollar companies that are investing enormous sums of money to build out their AI data center networks because they believe talent will become Nvidia's new moat and that's where the opportunity is.
I trust Jensen Wong's judgment, so I won't refer to it as a bubble. He is the CEO of Nvidia, which is perhaps the originator of accelerated computing, which is very important. It's not discussed as his generative AI. Every step of the way, Jensen has encountered skepticism. And despite his continuous reputations, the skeptics never seem to learn.
And I believe he's doing it once more. Despite the fact that data center expenditures are indeed enormous, you must remember that the majority of them come from the enormous cash flow of the wealthiest businesses on the planet that are taking out loans. Yes, as far as we are aware, Mark Zuckerberg has, in my opinion, sort of set the stage for how much money can be invested in the proper talent.
Now, one of our better ideas was meta. I believe we discussed this last week, but those investments don't always provide a profit right away. Therefore, the possibility of spending2 to3 billion on AI21, which is bringing in $50 million must be predicated on what we believe to be the subject of 2026.
Additionally, Aenic AI will be the focus of this theme. The actual way to unleash revenue is through enterprise AI. We're trying to figure out where the revenue comes from, which is one of the main reasons we had this bubble argument throughout nearly the whole second half of 2025. like the capital expenditure.
We are aware of all the limitations and that every hyperscaler will keep spending astronomically to get their first for aisi whatever you want to call it today. However, we also know that 95% of the data will essentially allow businesses to use AI to increase productivity, efficiency and revenue. This data is located behind firewalls and enterprise tools and agents must be developed.
Only a small percentage of people in the world possess the knowledge and skills necessary to accomplish this. Therefore, whether it's Google with its 10-year investments in deep mind, Mark Zuckerberg, etc., Jensen is aware that in order to be able to play in every area of the stack, he must bring in the greatest players.
We believe that he is making the right decisions regarding the supply chain, securing the wafer and packaging capacity with TSMC. And you know what he has been able to accomplish with memory and moving forward, possibly even slowing down Google's XBUS? Jensen is doing a great job of playing a game on all these many fronts.
But to put it simply, what is inference? That's what it is. Let's imagine you are a doctor who attends medical school and receives training. That is the portion of training. I'm saying, "Oh, no. In the infant section, I had either a broken arm or a broken teeth. For that, see a dentist." Yes, I am aware of it.
I got lipstick on my teeth because of you. That's what you were politely telling me. You are the physician. Then you're giving the answers inference, which is its condensed form. What's the matter then? Why is this spending happening? Huge [snorts] wager. Thank you. is investing a significant sum of money, possibly up to $23 billion, to take control of Nvidia's enormous $20 billion AI inference bet.
The chipmaker has completed two significant transactions in less than 4 months and may pursue a third. Inference was the main focus. Thus, the technology that executes learned AI algorithms. I hope the doctor analogy was successful. However, in late December, Nvidia made its biggest ever acquisition of Grock with AQ, its inference processor technology, for about $20 billion.
They introduced Jonathan Ross, the company's creator and the man behind Google's TPU chips. These are the proprietary semiconductors that performed admirably in the latest AI models. Gemini second Nvidia hired its CEO and other key personnel in September and spent more than $900 million for Infabrica's networking technology which is crucial for connecting AI chips.
Third, local Israeli media reports that Nvidia is in advanced negotiations to purchase Israeli AI firm AI2 Labs for between $2 and $3 billion. The transaction is valued at approximately 10 to 15 million per employee for their 200 person team while OAI 21 only labs generates about $50 million in sales.
Why Nvidia's dominance is threatened by inference. A lot of PhD holders. It's a play of sheer skill. Regarding the agreement, Nvidia declined to comment. Although Nvidia controls over 90% of the market for AI training, inference is rapidly disintegrating. Because inference is less expensive to operate and requires less memory and horsepower, it makes room for rivals like Google and Amazon.
Well, a lot of businesses are attempting to produce customized chips. They are appearing everywhere. Even so, the AI21 contract is still pending. The clear investment plan is to acquire inference talent, particularly expertise in technology, before rivals can. All right. And we're not sure if they're paying too much for this thing.
Here in New York, there are several excellent eeries in our immediate vicinity. Even while I could spend $30 or $200 on the same plate of spaghetti, it wouldn't necessarily be any better. I worry if they're paying too much. You don't need to deal with that. However, we can talk about the stock price.
Bring up that chart again. The filthy little secret is right here. Why the stock of Nvidia has stagnated for the past 6 months or so. Nvidia stock hasn't done anything. That was a lot considering it was the October high. Therefore, what are people? However, the stock has essentially remained flat. Yes, it did make an appearance in October.
Why? Why is there no movement on that stock chart? Regarding that, what are people saying regarding the stock over the past several months? There's been profit in many of these AI bets, but what will be Nvidia's next major catalyst? They are therefore reverting to other plays such as memory correct that has witnessed an enormous I am aware that you recently discussed the enormous spike for many investors.
The upside is let me move into the other picks and shovels of the AI trade. Regarding your remark that Nvidia is overspending, the company has around 80 billion in free cash flow which are projected to reach $96 billion by the end of the fiscal year, a 58% increase from the previous year. They must take action because they have a large sum of money on hand.
to assist shareholders. They were only able to spend around $52 billion on share buybacks. What more will they do? They are in fact purchasing businesses. You mentioned it just now. In order to acquire businesses, they're investing billions upon billions of dollars. What investors need to know about Nvidia's long-term thesis.
Due to their size, they are unable to simply purchase it. Thus, they're engaging in such fraudulent transactions as this Grock business. Although they didn't actually purchase Grock, they effectively did so by bringing the CEO over. Nevertheless, Grock will undoubtedly continue to exist. Shall we consider it a deal?
We refer to it as a merger. What do we call we refer to it as an alliance or a partnership which seems to be the standard, right? If you discuss Open AI and all these agreements with AMD, Broadcom and other companies, everyone is essentially just friends today. Don't call a comeback since they've been here
Comments 0