100%
1/1 predictions confirmed
0 pending · 0 expired
7
companies · 12 data points
We're already seeing it with llama models, like multi-modal llama models. We're not rolling out in Europe because of the regulations they put in. So I think that's the net of this is that the consumer suffers, right? They have fewer choices and there are fewer ways for them to modify, customize these things for their purposes. So I think it's a net negative if you start to restrict what people can do, especially this early.
in that same timeframe, can we double the price? And I think that doesn't sound crazy, right? In three years, can I double the price?
the valuation and stock price of Databricks is strictly less risk than Mosaic.
There are really three places in the world that can build state of the art silicon. Uh, TSMC, Taiwan Semiconductor. That's the biggest one. And that's where NVIDIA has a deep relationship. Samsung is another one that NVIDIA also fans on and then Intel, uh, as a fab and Intel primarily focuses on CPUs for their, um, for their fabrication capabilities. And then beyond that, there's like a something called high bandwidth memory, HBM memory. Yes. Packaged within the, the same physical package as the
There's really three places in the world that can build state of the art silicon. Uh, TSMC, Taiwan Semiconductor. That's the biggest one. And that's where NVIDIA has a deep relationship.
a customer has a relationship with AWS. They believe they can get AWS to give them GPUs. We can run our software stack inside of their tenancy without ever seeing their data. That's, that's something that people like because of the security and privacy. Uh, but as he said, the shortage of GPUs starts dictating how people go here.
that model took nine and a half days on 440 NVIDIA a 100 GPUs. And it costs about $200,000 to build from scratch.
there's really three places in the world that can build state of the art silicon. Uh, TSMC, Taiwan Semiconductor. That's the biggest one. And that's where NVIDIA has a deep relationship. Samsung is another one that NVIDIA also fans on and then Intel, uh, as a fab and Intel primarily focuses on CPUs for their, for their fabrication capabilities. And then beyond that, there's like a, something called high bandwidth memory, HBM memory. Yes. Packaged within the, the same physical package as the GPU,
The new H 100. That's the latest GPU from, from NVIDIA. Uh, it's going to help a bit in the sense that each one is faster than the previous generation. So you don't need as many, but I think what will happen is it's sort of like goldfish. You know, you grow the pond you have like as the capabilities of the hardware gets better, people just want to use more of it. And, uh, I anticipate us routinely using a thousand GPUs for, for customer workloads. So, um, it, that crunch is going to continue.
we are in a GPU crunch, no doubt about it. And that's not going to alleviate for a while.
Nvidia clearly had a huge bump recently. Um, they are the backbone of all of this, both training and inference right now.
Intel, uh, as a fab and Intel primarily focuses on CPUs for their, um, for their fabrication capabilities.