Kirstin Burke:
Welcome, everybody, and we are going to wind down the year with something that we usually do either in December or January, which is really queuing you all up for things that you ought to think about as you enter 2025. What are some priorities to be thinking about, what are some trends that really are going to shape the coming year, and so we’ve identified three for 2025. Shahin, I’ll let you lead off.
Shahin Pirooz:
To be fair, we didn’t identify them. One of them is actually, one of them is actually also the source of the top three.
Kirstin Burke:
Yes, yes.
Shahin Pirooz:
We went to ChatGPT and we said, what are the top three initiatives or strategies for 2025? And those three were AI for automation or augmentation. The second one was edge computing or distributed computing. And the last one was sustainability, which I thought all were kind of interesting because they’re kind of in conflict with one another. So as you think about AI and we had a TECH Talk four months ago talking about what is the impact of AI in the cloud and why should people consider not putting their AI architectures in the cloud? And one of the big factors is the cost associated with it and the sustainability impact of using AI. And so it’s kind of interesting that the initiatives for ’25 coming out of ChatGPT include both AI and sustainability. And edge computing obviously is one of those things that is becoming more and more relevant because as we get to more and more IoT devices and more and more devices that need to be able to communicate information back as our services, our products, our people become more and more distributed, all of those things need to be able to talk to edge computing rather than core computing inside the network. And then obviously securing all that is a factor we worry about. But yeah, those, for me, the conflict between how do you stay sustainable but also build all this very high compute and resource heavy infrastructure to support it.
Kirstin Burke:
Well, I think that’s really the story of ’25, right? And I think we started seeing it this year, when AI really is kind of becoming more than a concept. And you see more and more people really putting initiatives out there, applying it to workloads, applying it to business processes and things like that. I mean we see the horsepower that that requires and the big story out of that is not only, hey, this could be transformational for some businesses, but hey, what are we gonna do to sustain this? Because it requires a lot of everything. And we then have kind of our green and sustainability priorities that just as human beings we have, which is how do we take care of the environment? How do we make sure we leave a smaller footprint? And AI is not footprint friendly. And so I think going into the year, really organizations are going to have to think about how that balance works for them.
Shahin Pirooz:
You have to prioritize one over the other to some degree. There’s got to be a give and take. You can’t get something from nothing. And, you’re right, AI takes an awful lot of compute between the massive amounts of storage that are required to store the data, and that storage has to be up and running all the time so that you can access it. So it’s constantly consuming. It’s not just active when you need it. It has to be there so that when a search happens against that data, it’s available in there. And then to the massive amounts of compute with regards to continuous education and training of the models, and the learning process that happens to continue to improve and evolve the AI, as well as all the interactions that are doing the dynamic searches and interactions and natural language processing, there’s a lot required to make a system support a simple generative AI like ChatGPT.
Kirstin Burke:
Well, you just even go to the basics. I mean power, equipment, water, cooling, I mean there’s just, it trickles down just to the simplest, basic things of those types of things that we’re trying to conserve and be smart about.
Shahin Pirooz:
Exactly. And you start to now look at if I’m putting it in the cloud, is the cloud I’m picking, are they doing all the right sustainability things in order for me to feel comfortable? And does that also apply to these very highly costly compute resources that are used for AI HPC infrastructure? And then lastly, if you’re not doing it there, if you pick a colocation and build your own, which is what we were suggesting as a good way to keep cost down, does that colocation have a sustainable model? And finally, if I’m going to build everything myself and build a data center, put it in my facilities, now I need to think about how am I going to generate power and cooling and am I using air, am I using redundancy, am I using solar power? So it really becomes a very conflict heavy balancing act that you have to play between sustainability and technology enhancement and advancement.
Kirstin Burke:
So we’re not the only people that have noticed this, right? As brilliant as you are, there are other people that are observing this issue too. What have you heard from the market or from any of these high computing folks? Have we been hearing anything out of vendors that are recognizing this and that are maybe trying to think of how to still attain the advances that AI is offering without, you know, burning up the world.
Shahin Pirooz:
Yeah, cooling is a major factor. So cooling is one of the biggest impacts there are. And there’s a lot of players that are building the HPC platforms that are focusing energy on the liquid cooling. Liquid cooling allows you to pump liquid through the systems and cool down the processors without having to use HVAC systems to cool the ambient air. And heat exchange is another aspect that people are looking to do. So, you know, there’s a lot of conversation about data centers that are going into places that are very cold so they can take the ambient air and process it into the facility and just it’s movement of air as opposed to cooling of air. So there are a lot of activities trying to figure out how best to take advantage of our environment as it is. But, you start thinking about that and there’s so many, every decision we make is as a species ends up having trailing impact that we don’t realize. If we start taking cold air from the outside, running it through servers and pumping hot air outside, are we not reducing the ambient temperature or increasing the ambient temperature outside by pumping hot air into it? I know it’s big, much slower impact, but know all those things come into thought as you start thinking about what are the best approaches here. So liquid cooling is a very interesting idea. The under underground data centers are another factor. So trying to use the coolness of the earth to cool the facilities. But, I would say in terms of manufacturers, liquid cooling is probably the most forward thinking at the moment. But, it all depends, we’ll see what happens when it all comes to bear.
Kirstin Burke:
Right, right. Well so if we back off a little bit on the kind of conflict areas that these three topics generate. As you think about AI, as you think about edge computing, as you think about sustainability as we kind of go into ’25, what are the things that in each of those that you would recommend people either think different about, or anything to consider that might be new or fresh going into ’25?
Shahin Pirooz:
I would say since it’s been a focus of ours this year to help customers with it, I wouldn’t say that it’s new or fresh. There’s technology decisions to make about whether you consume cloud services or not. And that decision is really based on how you’re accessing data, are you having to exfiltrate data from the cloud or not. Because data exfiltration still continues to be a very expensive aspect of cloud consumption. So if there’s a lot of data exfiltration then you probably want to look at on-premise resources. And if you’re building on-premise, then you want to think about the most efficient systems for being able to squeeze the most power out of with the least amount of sustainability impact. So there’s a lot of factors that go into architecture and design in picking out where you need to go, what you need to do based on the use case you’re trying to accomplish. A lot of the AI implementations so far, even generative AI, have all been using public models like ChatGPT, which is bringing the data into your AI as opposed to pushing it out into something. But those people who are creating their own large language models that are very specific to their industry now your applications and customer applications are going to be coming and getting data out. And in those scenarios then it becomes very costly to use cloud. So really understanding how your application is used, who’s consuming it, what the data flow diagrams look like for that application, will help make architectural decisions about where to put it, how to build it, what technologies to use, and is cloud a viable factor or not?
Kirstin Burke:
Where does security play into things? So, I’m an organization, I’m developing my own, you know, maybe it’s in the cloud, maybe it’s not, but I all of a sudden have a lot of data. A lot of it might be tangential, but I do have a lot of probably IP or customer data or whatever. How do organizations think about securing such massive amounts of I would imagine unstructured data?
Shahin Pirooz:
There’s a combination of structured and unstructured data that goes into these models. But the traditional pillars for security don’t change. You have to have proper you have to put controls around that data so that it is backed up, it is protected, access is restricted, there’s encryption where encryption makes sense. So, you know, managing. we’ve talked about the CIA triad in other conversations. You have to make sure the data is confidential, it has integrity, and access is restricted. So, if you take that triad from a security perspective and apply it to your data, that’s the first step in protecting it. Where we start to get into challenges, the concept of model poisoning, which is as your models are educating themselves and learning and they’re going out to the Internet to pull data in, how do you prevent poisoning into your model? And there’s not a great answer for that today in regards to how to protect your large language models from external influence and hallucinations and all the other things that come from developing a large language model that feeds an AI. So, I don’t have at this point, it’s something that we continue to look at and evaluate to see if and how we can add value in that space. It’s really difficult to protect that, other than looking, at the end of the day it’s going to be looking at the lexical analysis that it’s doing and seeing if there’s something that is malicious in nature.
Kirstin Burke:
And awareness and ability to inspect.
Shahin Pirooz:
Imagine a large language model that is applying security to a large language model. That’s really the end state we would probably end up going to, but we’re not there yet. And then in terms of protecting the actual model itself from theft, or ransomware, or whatever, then you go back to that you got to lock it down, you got to limit access, you got to make sure it’s encrypted.
Kirstin Burke:
Its a lot. It’s a lot. And in addition to all of the security and data structure challenges that everyone already has to add to the table.
Shahin Pirooz:
Exactly. And then with edge computing, you know, the obvious, easiest answer for edge computing is leveraging cloud microservices. whether that be one of the major public hyperscalers or companies that are cloud companies that are providing edge services. And then sustainability, as you’re looking at implementing in colocation in cloud and whatever, understand how that particular provider’s approach, what their particular approach is to sustainability and if it aligns with your own. As you’re building infrastructure within your facilities, as you’re adding facilities, as you consume power. You know the simplest things are what kind of light bulbs are you using? What kind of HVAC systems are you using? So if sustainability is an initiative that you are going after, you need to start looking at not just where your compute runs, but also what are the places you put people into. How are you, if you’re providing charging for electric vehicles, how are you powering those? So there’s a lot of factors that go into sustainability that go above and beyond just traditional IT.
Kirstin Burke:
We’ve talked about liquid cooling as something that’s forward thinking and probably to come soon. Are there any other things that you’re seeing across the industry that you look at and say, wow, that’s really smart the way they’re thinking about working sustainability into whatever it is they offer.
Shahin Pirooz:
Not more than what we discussed a few minutes ago. It really is trying to look at how we take advantage of natural environmental influences. For example, being in colder locales where we build data centers. Building data centers under ground to be able to take advantage of the natural cooling of the soil. Doing heat exchange, whether through water, like liquid cooling or some other medium. Those are all the things that they’re unique ways of looking data center builds instead of the traditional hot aisle, cold aisle that we used to traditionally look at data centers. We do a lot of work with data centers, so we spend a lot of cycles trying to help with sustainability and building designs of cabinets that work well for that. But, it honestly is one of the more challenging things as we move into if in fact ChatGPT is right and sustainability is the next big thing for ’25 for us to look at. It’s a heavy lift.
Kirstin Burke:
Right. Well, it’s an interesting conversation for business leaders to have, right? Do you mean what you say in terms of sustainability being part of your values? Or is what you need to do to compete or to add new value, and it doesn’t mean, it doesn’t have to be a one or the other, but there does have to be flex.
Shahin Pirooz:
Yes.
Kirstin Burke:
And to what extent is the business willing to flex on one side or the other to do what makes you unique?
Shahin Pirooz:
Is it more important to have political capital and be more sustainable, or is it more important to have a customer experience that relies heavily on AI?
Kirstin Burke:
Interesting. Well, with that we will sign off for 2024, and we will next see you in 2025. And it’ll be an interesting year to see if all these things come to fruition. And obviously the things that come out of ’25 that neither of us would have expected at all. So, we’re off to a new year. We wish you all safe and happy holidays and we’ll see you in ’25.