원문 보기
Outside of that particular category… Okay, so we just talked about four major model manufacturers, maybe five. We talked about Open AI , Anthropic, you mentioned Gemini, we talked about…we talked about… we talked about…and we talked about Meta. Yes. How do you see them all in the long run? I think you're seeing convergence.
It used to be that when a model was dying out, it was so superior to everything else that you thought, "Oh my God, we've gone from kerosene to jet fuel, right? To jet fuel. So of course I'll go to jet fuel ." Instead, I think the analogy now is that we've seen the nth version of the iPhone. We all kept improving it.
We still use it for more or less the same purposes. My elbow hurts, what should I do? And you wonder, "Why am I spending so much money?" And I think that's where we are today. So if you're going to control all this behavior, why not just spend 50 cents or a dollar? This is a quite rational question.
The question is whether a metamodel like Grok is as good or close enough to approximating what you think is needed to achieve with Anthropic. The answer is: definitely yes. In some very narrow use cases, no. But for those narrow use cases, I would say spend $50 per barrel of oil. Spend them. That makes sense.
If you're a cybersecurity company, if you're Palo Alto Networks, and you can book billions of dollars of additional revenue by locking down the entire infrastructure of large companies, use Fable. But the panic over the weekend, even at Anthropic , you know, we were talking about this yesterday morning.
Anthropic, as you know, planned to end the free experience of Fable for some paid users. This would sort of end the expansion, so you would actually have to upgrade or pay to access it. And then suddenly they say to themselves that we ca n't afford it. We ca n't afford to do that because if we do that, people will cancel their subscriptions and switch to Open AI.
Exactly. And the Open AI model is excellent. Soul is perfect. Correct. Mythos, fable, is perfect when you can use it. All these people face the same limitations. They have huge power constraints, huge data center constraints, and therefore they are limited in terms of utilization. Meanwhile, those who have all this capacity, Google, SpaceX, and Meta, are finally getting back on their feet and starting to consistently release model after model after model.
As I mentioned, you've seen over 100% year-over-year growth. You have an impressive backlog. What changed in the company this quarter that makes this particular moment in time so important? I think what you think and thank you very much for inviting me today. I appreciate it. What you're seeing is simply an acceleration of our operating leverage, right?
We see the business reaching enormous scale, and that scale translates into profitability, margin growth, and importantly, it also translates into fantastic operating leverage for our clients. CoreWeave's customers are some of the world's leading AI companies, and the value they derive from our platform drives growth rates across the sector.
This allows you to increase your forecasts for the entire year. Currently, it ranges from 12.4 to 13.2 billion. Your third- quarter forecasts point to about 150 percent year-over-year growth by mid-year. What gives you confidence that CoreWeave can maintain this level of acceleration without hitting physical capacity limits?
These are our achievements in achieving our goals, right? There are currently over 51 data centers in operation. We've been doing this for eight years. The amount of supplies we launched amounted to over 500 megawatts of active power in the last quarter alone. This just goes to show the speed and depth of expertise we have in delivering these calculations to our clients.
And the truly impressive number was the $104 billion in backlog, plus the more than $25 billion in new commitments that were already signed in the third quarter, not included in that figure. Brandon, looking at this, how much confidence does that give you in this ongoing debate about the need for AI infrastructure that we're seeing, in terms of sustainability, versus just short-term growth?
Inference is the economic engine of AI, right? We have spent years of intensive training supporting our clients in getting their models to the point where they could monetize their business models. And inference today is just a straight line. There's no other way to put it. This demand point changes from quarter to quarter, only higher and higher.
So the contracts we're signing today are what we're seeing being used in the AI loop. The AI loop means our customers will train, tune, infer, and then simply continue in that loop as they receive new information to continually improve their models. The CoreWeave platform is simply unrivaled in the market when it comes to enabling this type of AI development process and staying at the cutting edge.
And that is progress beyond these frontier laboratories. I mean, you have Caterpillar, you have Lidos, Flow Trader - all of them, as you mentioned. How different are the economics and physical requirements, such as geographic location, network latency, etc., when hosting real-time enterprise workloads versus training these massive fundamental models?
That's a fantastic question. Ultimately, we see little difference in the way contracts are configured . Our AI Labs clients are more focused on long-term contracts than five- or six-year take-or-pay contracts. I think companies are focusing on short-term contracts, say two to three years or so. Hi everyone, welcome back to my channel.
And if you're watching this video so far, subscribe to our channel. So, without wasting any time, let's start our video. Something extraordinary is happenmpanies are demanding more computing power, and hyperscalers are increasing their capital expenditures. Demand for memory is skyrocketing, and Nvidia is now doing much more than just selling GPUs .
It helps build, finance, and structure the infrastructure necessary to use these GPUs. This raised a fascinating question. Could this be the start of a multi-year semiconductor supercycle? Or are investors getting too comfortable with the idea that demand for AI can grow indefinitely? Let's take a closer look at this.
The strongest argument for the semiconductor boom is the huge amount of infrastructure planned. Major cloud companies continue to spend huge sums of money on data centers, networks, power supplies, and AI accelerators . Importantly, the industry is not just talking about today's demand. Companies are already committed to future efficiency.
And that 's why some analysts believe the current AI cycle could extend much longer than investors expect. The bottleneck is not necessarily demand, but supply. Developing the production of advanced semiconductors takes years. Data centers take years to build, and power infrastructure takes time . Even after the physical building is in place, connecting it to services and electricity can become another bottleneck.
The market may remain supply-constrained even as demand continues to grow . This is an extremely favorable environment for semiconductor companies. Why is Nvidia different now? Nvidia was once seen primarily as a company selling world-leading AI accelerators. It's still a core business, but something much bigger is developing around it.
Nvidia is increasingly involved in the funding and infrastructure surrounding AI computing. The company has partnered with a major financial institution to help raise hundreds of billions of dollars in capital for the AI infrastructure ecosystem. The basic idea is simple. A data center operator needs billions of dollars to build infrastructure.
Investors have capital. Nvidia already has the technology and insight into the future. Nvidia put these elements together, and suddenly funding became another mechanism to accelerate GPU development. This is potentially a very powerful competitive advantage. This is the most interesting part of the whole story.
Traditionally, investors finance things like houses, planes, cars, and other assets with measurable economic value. Now the industry is exploring whether computing infrastructure can be treated in a similar way. However, if the revenue generated by GPUs remains useful for years, can be redeployed, and has significant residual value, they become much easier to view as financial assets.
And NVIDIA has the advantage here. Its GPUs do not operate in isolation. They are connected to CUDA software. This software layer makes NVIDIA hardware much more useful and flexible. If AI customers disappear, another customer could potentially use the same infrastructure. This flexibility is extremely important to lenders.
Comments 0