GPU

“So it's designed basically, you can sort of think about it, you know, as your general purpose sedan, that's sort of the CPU and the GPU. All it does, it gets on the F one track and all it does is go fast on very specific workloads.”

“A GPU, instead, is really built for a very specific class of workloads, and generally those have been called throughput workloads, so it does lots of floating point processing and matrix operations, You know, and so it's very dedicated for things like graphics…”

“Jensen said in the earnings call this week that somewhere around 40% of our data center GPUs were going for inference, which I think is you know, pretty amazing and definitely a shift from where things have been a few years ago.”

“The model is less important than the data. And the compute that goes into training the model. If you have a model that has really excellent compute properties that allows you to scale really well, efficiently to, you know, many thousand of GPUs, the kinds of r…”

“It's not CUDA that's keeping, I think, keeping a lot of us. It's actually that there is nothing really dramatically better than NVIDIA's GPUs. And so if there's nothing dramatically better than, I mean, the reality is the cost for training and inference are so…”

“It plans to amass 1.8 million GPUs by the end of the year. And it plans to triple its data center capacity, which is already at five gigawatts, roughly what it takes to power all of New York city each day by 27, 20, by 20, 27, it plans to spend more than a hun…”

“Over the last 10 years, we've had essentially a million x performance increase.”

“Clearly there's a lot of GPUs that have gone through Singapore and then go right to China.”

“These models use a lot of GPUs. They are really expensive to train and they are really expensive during inference on today. So that's another area that, that we really question sometimes the ROI of some of these things because of the cost of all of it.”

“So far we've always been able to use our GPUs, even ones that we deployed multiple years ago for different use cases and get positive ROI from it. And that's why our depreciation cycle for GPUs is what it is.”

“If we are here two years from now and there's hundreds of billions of dollars of debt, And the demand for AI stabilizes, or we built enough data centers to support the demand we have at that point in time, and the price for leasing pieces of AI for renting acc…”

“GPUs, as you know, are still not elastic. In fact, they're often, you know, availability itself is, can be a challenge.”

“The idea that these things burn out in two or three years, like, it's kind of bunk, right? And from a physical perspective, Right. Within three years, these things are all still under warranty. So if they break, they get replaced. Right. But from a, like, this…”

“Nvidia sells GPUs. They don't sell proprietary models. And if they have great open models that everybody can use, then everybody is going to buy GPUs.”

“Now if you invest one thousand billion dollars today into GPUs for data centers, this means that within five years you are going to lose nine hundred billion dollars, you know. Somebody is going to lose nine hundred billion dollars in the near future because t…”

“The current status quo of we're spending two hundred billion dollars a year on CapEx. And, you know, yes, we're renting our GPUs, so revenue is accelerating, but fundamentally there's no There's not a new business line coming out of it. That is not sustainable…”

“So we have some GPUs that are failing inside of modern data centers on an 18 month cycle, some that are failing on a much longer period.”

“The failure rates of GPUs used so intensively for training purposes are much higher than inference specific usage.”