00:00 Speaker A
I was seeing this this clip of Michael Dell make its way around where he talked about on-prem AI, which is exactly what you’re talking about.
00:06 Speaker A
Uh why does it matter? Like, why can’t a company just run its stuff through a data center that is being run by a hyperscaler? Li like what is the advantage for them? And and when we say on-prem, does it mean like physically in the same building as the company that they they have their own or adjacent to, that they’re that they have their own servers?
00:27 Speaker B
Uh, all great questions. So I’ll I’ll tackle the second one.
00:31 Speaker A
Sorry, that was a lot of them.
00:34 Speaker B
No, it’s okay, it’s all good. Uh, so onprem is the traditional, it’s your corporate data center. Sometimes it’s physically in the same place, but increasingly we’ve seen a lot of companies use what are called colos, uh colocation centers where uh a company like Equinix and some of these other companies, they basically host the servers, they physically manage them, but they’re basically owned and or leased depending on the arrangement by an organization like a Coca-Cola or whomever.
01:04 Speaker B
So that’s the notion of on-prem. It could be physically there or in other places, but but sort of under their control. Um, and the reason they’re starting to do this, again, there’s the capacity questions, there’s the uh, how much can I even get access to this. But the biggest thing is what people like to call tokenomics, which is uh, how many tokens am I going to be generating and how much does it cost to generate those tokens.
01:33 Speaker B
And we’ve seen the ability now, one of the things that Dell also announced and uh Google talked about this at IOL last week, which I was also at, that some of their largest Gemini models, you’re now going to be able to run on prem. So, there will be certain types of workloads you can do completely within your own on premises environment.
1:53 Speaker B
Most of the biggest frontier models and capabilities, you’ll still go to the web, but the idea is what I call a hybrid architecture. I think you and I have talked about this when I was on last time. And that is this idea that you do some of the AI workloads in the cloud, you do some on prem and even and you start to do some on the devices.
2:09 Speaker B
The other point about Dell’s earnings, by the way, was not only was it the servers, the PCs crushed it and everybody was shocked by that. Of course, they followed up HP had HP Inc had just also announced a really strong quarter on PCs, ads did Lenovo. What’s happening in the PC market is the ASPs are going up because of the memory shortages and everything else. So the unit numbers are down, but overall revenues are increasing. And both of those companies in fact, predicted even a bigger year than they originally thought, which that definitely caught a lot of people by surprise because everybody’s like, well, memory shortage, you know, all these sales are going to get crushed. And in point of fact, they’re like, yeah, units may go down but you you know, they’re going to make it up on revenue.
