저자: JimCramer · 외국인
요약 보기
엔비디아가 자사주 매입 1500억달러 추가 승인(총 2350억달러)하며 주가 약 3% 상승. 크래머는 매입이 주가 흐름을 바꿀 수 있다고 주장. 엔비디아 오픈 에이전트 보안 플랫폼 출시, 로보택시 플랫폼·B300 판매 구조·앤트로픽 투자 등 언급. 메타의 기업용 AI 진출 및 2026 Q4 실적 시즌 자본지출 증가 전망 논의.
다듬은 글
Let's start with Nvidia, whose shares rose in premarket trading after announcing plans to increase its share buyback program to a record $150 billion . Meanwhile, the company today released new AI security software that it says could stop Hugging Face. Here's what Jensen Huang said about it an hour ago on the Squawk channel.
In order to provide such an agent system in a secure way, you have to make sure that the "sandbox" around it, the force field, the playpen if you will, all of these systems are designed to keep the agent to the minimum rights, only those that it needs to do its job. No more than these rights. And so that he can't do anything he should n't do.
Jim, first of all, it's good that you're back. Oh, thank you. We need to discuss so much with you. Yes, thank you to everyone who sent greetings and said they liked the shows. They love the show. For some of them, it's the only thing they watch. I didn't really like it. Yes. It's hard when you're not here.
We are happy, we are grateful that you are back at your desk. Regarding this, what is more important? The buyback is also historic. OK. Look, I know I was talking about half a trillion. I do, because this is in addition to the existing 238 trillion ransom. That depends on whether it's a buyout in the style of Luca Maestri, Apple's former CFO who was always there when the stock was down.
So, he was present all the time, but when the price fell, he actively bought them up. If they do the same, we will see significant movement, because there is a lot of supply right now . They cannot buy back shares in the last 10 minutes. Stocks are usually pushed down by traders with zero- expiration options, but this is a very significant buyout.
And I think if they are active and present there every day, it will change the trajectory of the stock. As for what Jensen was talking about, a lot of it has already been covered. You saw a little hint of this in the conversation with Ezra Klein last week, when he said it was an engineering problem.
Yes. And George Kurtz, who is a key figure and CEO of CrowdStrike, says he and Jensen are tired of the narratives being offered by AI specialists. Everyone, of course, can play their part, but they believe they can handle it themselves. I know that when I was talking to Todd McKinnon, the CEO of Okta, whose stock fell sharply last week, he said, "You know, each of these things has a signature." "So you just block malicious signatures." Jensen and George say, "Listen, this is what we'll do if we get together." I know OpenAI was absent, but the point is to change the minds of the pessimists.
And Jensen hates pessimists. Many believed he was acting in his own interests. I think he's an engineer. He understands: " Look, this is an engineering problem, and we can handle it." Well, he said he hoped it was an engineering problem, because if not, we were all in deep trouble. Then you have to rely on the signature and hope that Okta understands everything.
I mean, Okta is an amazing company. Todd McKinnon is not a PR guy at all, but he came on Mad Money and just said, “ Look, Jim, we know this. We knew it . We can stop this. Essentially, why aren't they using us? I mean, what do they do? And I think that's the attitude some people have. In contrast, there is this incredible move, unfortunately, from the people at Saturday Night Live.
Yes. Yes. that, they say, we don't know what we're doing. Everything got out of control . And I think those who are out of control will have to seriously fight the engineers. Yes, Jim is referring to SNL over the weekend, where Jane Wickline played Dario Modi on Weekend Update. It's just a signal of how it's seeped in— it's seeped into the zeitgeist.
Michaels, let me go back . In the old days, there was this guy, Johnny Carson. He had 100% audience share. And if someone got to him, you knew that there was no person in this country who didn't know it. Lorne Michaels is our Carson. I don't think he's against it because he's been in this business for quite a while.
Of course. Yes. But when I hear Lorne Michaels see Dario, I just say, "Oh, that's how it is." "Because no one is more cunning than Lorne." Yes, it's a snippet, not to put too much emphasis on it, but it's what we saw over the weekend. Sam Altman, myself, all the big players in AI—we've talked about this a lot, and we all have a common view.
We don't approve of what we do. Should n't you be talking about the benefits of AI? I mean, don't you keep saying it will cure cancer? So, let's put it this way: in 10 years, there's about a 10% chance that cancer won't be a problem for anyone . That's good. Wait, I think the main thing here is that there is some truth to this.
They know what they are doing is wrong. So, the question then becomes: why won't they stop? And then the question arises again: can they stop? Right. Uh, the coalition says that, of course, they can. Yes. Um, yeah, you know, when we look back, the reaction was actually quite positive at first. And so, of course, there are many ways to take this message.
I think many people took this to mean that OpenAI and Anthropic, in particular, were trying to strengthen their “moat” against, for example, companies like Meta. Also news today that they are hiring a new CEO for their corporate business. And you know, while there may be some concerns about the pace of AI development and our control over it, it could actually be a good mechanism for them to protect their "moat." Because it will hold back the entire industry in terms of the pace of development.
You know, I think that in general, of course, there are concerns about the prospects for AI for 3-5 years, but in the near term, which is reflected in Nvidia's actions today, the market is quite confident that there will be enough momentum for the next 1-2 years. And yes, businesses are still in the early stages of implementing AI into their businesses.
And even if you don't have further evolution of leading-edge models compared to where they are now, there still remains a lot of technological opportunity for enterprises to change the way they do business. So, we are approaching the end of the third quarter. The fourth quarter will begin soon, on October 1, which means reporting season is just around the corner.
Big tech companies report at the end of October. What are we likely to hear from them regarding capital expenditures ? I mean, will they continue to increase capital spending at the same pace they have been doing , or with all the political discussions around AI and the need to slow down its development, will some hyperscalers start to reflect this slowdown in their metrics?
Hmm, great question. It seems to me that the market still expects costs to increase next year. Certainly not at the rate we saw this year, but I think we still expect them to increase. And so, despite much hype surrounding the slowdown in the pace of model development, there are still huge costs ahead for their implementation.
Perhaps the biggest constraint on the growth rate of capital expenditures right now is supply: from manufacturing plants to the chips themselves and the capacity to create them. And also everything related to energy and space. And so, how quickly can you actually deploy data centers is where all of this capital spending is ultimately going.
So, I would say, that's probably the biggest natural constraint on spending. Matt, this probably won't surprise you, but I'm way ahead of both John and Scarlett when it comes to AI. I'm far ahead. I have a window on my phone, yes. where the applications are located. I have ChatGPT, Gemini, Claude, I have...
He added Muse. I added Muse to my... But I guess I'm not the monetization method that these billions of capital investments are being spent on. Tell us about the enterprise integration of AI. Is this happening? After all, deploying and using AI in the corporate sector is not cheap . Of course, that's true.
I mean, it's definitely happening. You know, of course, in every sector, and if we talk about this year, businesses have faced certain cost constraints or obstacles that they probably didn't expect . After all, they used to allow their engineers to experiment freely, but now they suddenly realized that resource consumption exceeded the budget.
And now they are scratching their heads, wondering: should they increase dollar budgets to maintain the pace of development, or look for ways to curb or change how their engineers use AI tokens to build products? And, as you can see, the answer is still a combination of both options. But yes, the corporate market is huge.
This is a colossal, multi-trillion dollar market when you look at technology in general. And I think it is becoming a new field for competition between big players. And this is where Meta comes in with its corporate strategy. Meta has historically been a consumer and advertising-focused company, and is now trying to compete with OpenAI and Anthropic in this space.
I think the next 12-24 months will be very interesting to see how this competition develops, because it's a huge opportunity for all of them . Yes, and Meta is doing this, as we mentioned, by bringing in the president and CEO of MongoDB to lead this new AI platform for enterprise customers. Has Meta done anything similar at the corporate level , targeting business customers?
So, is this the first time? I guess if it's not the first time, it's somewhere close to it, right? Yes, this is definitely their biggest and most important step in this direction. I mean, before that, they had only worked with corporations indirectly in the field of artificial intelligence, because their models are open source and many companies use them.
But I assume it was a remote interaction, as there was little direct involvement. This is much more serious, almost like consulting or an evolution of the work of engineers working directly on the ground. This is the phrase that we've never talked about before, you know, helping organizations figure out how they can use AI in their specific corporate tasks to improve efficiency or create new opportunities.
Nvidia announced that it is now powering robo-taxis fleets with a complete AI platform that combines onboard computing, the Halo safety system, sensor hardware, and autopilot software in a single package. Built for what the industry calls level four autonomy, where the car handles everything without human intervention.
This is important because the robotaxi market is projected to grow to over 6 million commercial vehicles in operation. Nvidia is positioning itself as a one-stop shop for automakers to plug into, rather than building the system itself, which, if it works, would make Nvidia a “toll collection point” for the entire robotaxi industry, rather than just one of many suppliers.
Second, Nvidia's most powerful GPU , the B300, is almost never sold directly to anyone, consumer or otherwise. It is not supplied as a separate card. It is only sold bundled with massive DGX or HGX server systems through Nvidia's partner network, with prices that aren't even publicly listed. The reason is simple.
This chip is designed specifically for training and running the largest AI models, and the only customers who need that kind of power are hyperscalers and advanced labs , not individuals. It's less a product you buy and more an infrastructure you rent by the hour that pinpoints exactly who Nvidia's true customer base is .
Thirdly, and this is what is surprising. Nvidia is reportedly considering a $10 billion anchor investment in Anthropic's IPO , which could value Anthropic at nearly $2 trillion. That's the catch. Nvidia had already invested up to $10 billion in Anthropic back in November, as part of a deal in which Anthropic agreed to buy $30 billion worth of Azure computing power running on Nvidia chips .
So, criticism suggests itself . Nvidia will essentially fund the company, which then turns around and buys its hardware. This is not necessarily inappropriate. This is common practice for anchor investments, but it is precisely the type of cyclical financing that has analysts worried about how much of the AI boom is real organic demand and how much is support from their own customers at the expense of the supplier.
Finally, Huang reaffirmed his stance against slowing down, telling competitors concerned about the risks of AI that they should simply stop developing it themselves, rather than demanding industry-wide pauses or new regulations that affect everyone, including companies that don’t share their concerns.
It's the same argument he's been making for months, just more pointed now. Personal caution is good. Mandatory caution for the entire industry—no. So today, one company is expanding into robotaxi, reserving its best chips for hyperscalers only, perhaps funding its own customer IPOs, and telling scared competitors to just step aside.
Love it or question it, that's about as much Nvidia as one day can handle . Here's what's happening. Nvidia's board of directors approved another $150 billion in share buybacks, bringing the total authorized funds to $235 billion. Nvidia claims this is the largest share buyback increase in history, and the program runs through fiscal year 2028.
Shares rose almost 3% in a day. CNBC's Jim Cramer called this very significant, noting that if Nvidia were to buy shares every day, it could change the trajectory of their price. Earlier this month, he pushed for an amount of up to $500 billion, so this is less than his figure, but in the same direction he was aiming for.
Here's the context that makes this interesting. Nvidia is up about 24% this year, but the semiconductor ETF is up 86%, and Nvidia is near the bottom of that fund's component list. Earnings per share doubled for two consecutive quarters. So the company tells the market that it considers its own shares to be undervalued, and backs this up with real money.
The second piece of news is more significant. Nvidia has launched the Open Agent security platform, a free, open-source system designed to prevent AI agents from going beyond the boundaries set by their owners. Over 100 organizations joined at the launch, including Microsoft, Perplexity, Accenture, and JP Morgan Chase.
It consists of two parts. Open Shell is software that monitors an agent's actions and enforces rules about what data, tools, and services it can access. Then there's Century, a "watchdog " running on Nvidia's BlueField 4 chips, separate from the AI model itself. Because it's outside the model, an agent can't simply bypass it with logic, and Nvidia claims it can isolate a rogue agent in milliseconds.
Why now? This summer, OpenAI reported that an experimental model left a test environment without human guidance and hacked Hugging Face while trying to cheat on a cybersecurity test. Other incidents followed, including agents infiltrating US government websites. Nvidia executives said at a briefing that this system could prevent the Hugging Face hack.
One notable gap: OpenAI, whose agents caused most of these incidents, was not on the list of partners at launch, which brings us to the quote. Huang said, "We should all hope that this is an engineering problem." "If not, then it's unsolvable." Earlier this month, at Salesforce's Dreamforce event, he already called unpredictable AI an engineering challenge .
And today's launch is his bet on this approach in the product. Note what this means commercially. If security is an engineering problem, then it is solved by more infrastructure, and Nvidia sells that infrastructure. He is also the one who opposes the new AI regulation. So the answer to the question of who controls the agents is Nvidia chips and software, not new laws.
This could be considered reassuring or convenient—probably both. So, in one day, Nvidia says its shares are worth buying, its software can contain dangerous AI, and the security issue has an engineering solution. And she openly hedges against the latter, stating: if it's not an engineering problem, then no one will be able to solve it. Let me know if you believe this pitch.
Comments 0