원문 보기
CrowdStrike is showing incredible growth, setting a new all-time high yesterday. It looks like we're in for a new record today. Almost 180% growth in the last 6 months is not bad. The company continues to benefit from the demand for AI-powered cybersecurity. With us now, we are honored to have CrowdStrike CEO George Kurtz.
George, first of all, thank you for coming, because I have to tell you, we're going to put some rumors to rest today. Good. We will prove that this is not Stephen King, not "Confrontation", this is an engineering problem that can be solved. I'm talking about agent-based AI, which some companies are so concerned about.
What do you think about security solutions like NVIDIA Open Agent Safety Platform, which I was excited about because I realized that this was becoming an engineering challenge? Well, we are certainly a part of it. We work very closely with NVIDIA, and when we think about what an agent does, it's a program that runs, interacts with files, data, and performs actions.
And importantly, model developers are focused on creating barriers around exactly how the model thinks. But when you implement it in production, you need to see what it actually does. You must understand her actions. And one of the things we focused on was creating a "black box" for the actions of these agents.
You need to be able to track their actions and link that to control tools, which is what we've been working on with NVIDIA . So this is a problem that can be solved. It just takes the right partnership and combination of technologies to ensure security for the companies that use these agents. Okay, there are people who say this is a solvable problem, and there are others who call for a pause because we don't know what this will lead to.
Why are supporters of the latter view so popular in society now, not in your circle, but among the general public, especially before the elections? You know, I think there are a lot of different motivations here as to who wants to slow everything down and why . I think if you look at AI as a whole, it's a fundamental technology.
This is the biggest change I've seen in my lifetime and certainly in my entire career in technology. So you must be able to tame it. This is not true. In a race, you don't slow down. You want to achieve maximum safety in order to move faster. And I think this is an important topic. Of course, safety and security are important, but it has to be done right, and that doesn't mean we have to slow things down, giving our adversaries like China and others an advantage over us.
So, I think that's why people on both sides agree that this can be solved. We just have to do it right and without too many adjustments. What exactly? What can be solved? I want to try to understand this, because you hear different things from different people. Bill Gates gave an interview that I listened to last week where he said that we need a global oversight body, looking at open models, not just the closed ones that we're all talking about here.
So, what would actually work? Well, the right standards actually work, right? If you look at what organizations are focused on right now , it's self-regulation, but there are also things like NIST standards that come from the US government, right? They are working on updated standards for AI that focus on how to build, train, and safeguard models.
And there is also the oversight of the board of directors and external auditors. These are all great steps. Fantastic. But when you run it , you have to see what it actually does. Think about an airplane. You build an airplane to certain standards, then you put it into production, it flies, and when something happens, you turn to the " black box." What did he do?
What really happened? Therefore, control is needed during work. You also have a federal regulator, the FAA, constantly monitoring you. Yes, it is, but ultimately, we can wait the next 5 years for AI models to be released or be competitive and have self-regulation. Yes, but we don't want the plane to crash, George.
Well, that's why there are rules for airplanes, because people are involved in their creation . But from the perspective of an aircraft manufacturer, it takes years to certify something . Do we want to wait 5 years for the model to be certified? Well, yes, because in 5 years there are people who have been in the thick of it and are saying that we could all die, and that's not a very optimistic situation.
Jim, there's a 10% chance we won't die in 2030 . Elon Musk just announced the cost of surpassing everyone else in AI: $40 billion, and that money won't go to Mars. They go straight to Nvidia. And while this deal was becoming known, the open-source AI model we talked about yesterday continued to receive coverage in almost every major publication.
Let's look at both topics. The Financial Times reported today that SpaceX is in talks for about $40 billion in financing led by Apollo Global Management with one goal in mind — buying more Nvidia chips. The structure consists of about $10 billion in direct , that is, debt rated safe enough for conservative investors like pension funds to hold.
The deal is expected to close sometime in 2027. That's why it's such a huge amount. Last month, Musk said SpaceX's Colossus 2 supercomputer the end of this year. During SpaceX's August earnings call, Musk told investors that the company had decided to build its infrastructure exclusively on Nvidia, specifically highlighting the Vera Rubin platform—Nvidia's next-generation chip and CPU combination —as the best available architecture for AI.
SpaceX CFO Bret Johnson also noted that annual capital expenditures are approaching $65 billion. So this attraction of 40 billion is not a secondary project. That's a huge chunk of SpaceX's total spending for the year, and it all goes to a single supplier. And the relationship between Musk and Nvidia works both ways, which is worth dwelling on in more detail.
Nvidia isn't just selling chips to SpaceX and its AI startup XAI , it's actively helping to finance these purchases. Nvidia reportedly invested up to $2 billion in equity as part of XAI's $20 billion funding round earlier this year, specifically to help XAI acquire more Nvidia GPUs for the same Colossus 2 complex.
Additionally, Nvidia separately owns a stake in SpaceX itself, which was recently valued at around $21 billion . So you see: Nvidia invests in SpaceX, invests in XAI, and now SpaceX is turning to outside lenders for tens of billions more, which ultimately translate into chip orders going back to Nvidia.
The market reaction shows how investors perceive it . SpaceX shares, which only began trading publicly in June after an $86 billion IPO, fell about 1% in over-the-counter trading on the news, while Nvidia shares actually rose half a percent. One trader on StockTwits summed up the sentiment succinctly: “Bearish trend for SpaceX’s balance sheet , bullish trend for Nvidia’s backlog.” That's the essence of this compromise.
SpaceX takes on serious debt, and Nvidia gets a guaranteed buyer, securing billions of dollars in future chip sales. It's also worth comparing this to SpaceX's overall financial picture. The $40 billion raise is equivalent to about 1.8 % of SpaceX's current puts the company on the verge of investment grade.
One analyst also calculated the burden of interest payments. Every percentage point per annum on $40 billion in debt is $400 million per year in interest payments alone, assuming the entire amount remains outstanding. This is the true price of computing leadership for AI. And it's part of something much bigger than just SpaceX.
This year, Nvidia has built an entire financial ecosystem around itself, partnering with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to mobilize over $500 billion in outside capital specifically for AI infrastructure. SpaceX's request for $40 billion is just another element being built into this mechanism.
It is worth noting that this is not the first time Apollo has participated in such deals. In June, the firm led a separate $35 billion funding package, although it involved chips from Broadcom, one of Nvidia's real competitors. So Apollo is playing both sides in the “ chip wars,” funding Nvidia’s biggest customers while simultaneously supporting its competitors.
A quick addendum to yesterday's story as it continues to spread. Reflection AI's Beam open-weight model continues to appear in major publications today, and the wording is becoming clearer. The basic message has not changed. A model with 501 billion parameters that activates only 23 billion for each task, making it three to four times cheaper to operate than the Chinese counterparts with which it competes.
But it is remarkable how many separate publications now use the same title . An American open-source model backed by Nvidia, designed specifically to challenge China's dominance in free AI tools. This repetition matters. When one publication covers a story, it's news. When five or six independent publications present it in the same way within 48 hours, it's a signal that the AI industry and policy circles see this as a real turning point in the open AI race, not just the launch of another model.
Reflection positions Beam less as a product and more as a statement: proof that an American company, backed by money and Nvidia chips, can match what Deep Seek, Qwen, and Z.ai were giving away for free. So today, SpaceX is borrowing $40 just to keep filling its data centers with Nvidia chips, and Nvidia doesn't even have to lift a finger to make a profit.
She has already invested in the buyer on both sides. Meanwhile, the open source counterattack against China is gaining momentum in headline after headline. No matter how you look at it, it all comes down to the same company. Let me know what story you want me to follow next.
Comments 0