Microsoft CFO Amy Hood said capital spending in the third and fourth quarters would remain around the $22.6 billion, a level seen in the second quarter.
Growth rate slowing means instead of going from 80 billion (in 2025) to 120 billion in 2026, it will be 95 billion instead.
I know what growth rate traditionally means.... However, in this instance, I just think Amy is being coy.... We'll see in Q1 FY26.
I highly doubt it. Rubin has been tape out and brings another big gain in AI performance over Blackwell NVL72, both in training and inference. The leading AI models will benefit tremendously from it, so they won't lower their investment in the next hardware cycle, unless they want to loose over the competition...
Wait, who’s Nvidia’s competitor? It’s not AMD, it’s not Intel…. Im not an expert in the AI field but leading AI models use transformer architecture. If we choose to believe the experts then that will certainly not lead to the creation of AGI which everyone is chasing. Further more, transformer architectures are reaching their scaling limit.
Nvidia competition is themselves, like in the consumer GPU space. They must continue to innovate to grow and please their shareholders.
Transformers are far from their limit. We are just at the very beginning of code optimization. Just look at Deepseek and what they did to lower the hardware requirement. And unlike uneducated opinion thinks, this kind of breakthrough won't slowdown the market but will accelerate the AI adoption, ie Jevons paradox.
There's very little competition going on with Nvidia especially since they released a 5070 masquerading as a 5080. You can't even in good faith propose that argument after Nvidia released the worst gpu generation they've ever released in over a decade.
Deepseek is less hardware intensive but the outcome in terms of accuracy is still similar. There are several videos and papers discussing how the transformer model scales and how irs reaching its limit. I'll take the word of people who worked on and designed these models over yours.
1. You should read more and 2. it's not my words but from the IC industry that I'm working in for more than 30 years.
Optimizing and increasing the performance of the transformer model will not overcome the inherent flaws of the transformer architecture. I don't have time to watch the whole video you linked but I imagine that you didn't watch the video you linked at all. You were just looking to confirmation bias yourself. The video author just in the intro section calls this paper "5% ideas, 95% smoke and mirrors". If you even watched this video in any length, you'd realize that this does not actually significantly change the fundamental structure of the transformer architecture in terms of outcome. The inherent limitations still remain. As for further AI discussions with you, I'll be bowing out. It's not my field of expertise and it doesn't appear to be yours either. I don't have the time to invest in a conversation that comprises of you posting youtube videos that you don't even bother to watch.
The “less educated masses” don’t move markets, market movements are almost always from institutional players. For a mega cap like Nvidia having it drop 20% in one day means an institutional selloff, not retail nonsense.
Well, to be frank, I've been calling this the worst Nvidia generation for a while and I raised several concerns about the GPUs based on the configurations prior to the release. Hub calling it a 5070 is just another way of verbalizing the concerns I had raised immediately when I saw the specs. Hub is not wrong and the data backs it up.... Imagine a new x80 series not beating the prior 90 series but instead being multiple percentage points slower...
How would you describe the 5080?
