Episode 705: Pete Sacco on AI Infrastructure and the Human Work of Leadership
Calvin Hendryx-Parker hosts Pete Sacco, a 30-year veteran of data center building, founder of Gray Wolf Data Centers, and author of The Bridge. The first half is infrastructure. Pete argues the next trend is enterprises training small language models on their protected data and running inference inside their own walls, questions a neocloud model built on 20-year leases and GPUs that depreciate in a few years, and predicts data centers will get smaller, not bigger: a data center in every town, locally owned and switched, with identity, governance, and security run centrally. He splits inference into light, on the device, and heavy, federated in sites from a 300-kilowatt container to a 20-megawatt facility. The second half is leadership. When agents take the rote 80% of the day, what's left is one important decision after another, and Pete explains why that's exhausting, why the anxiety lands even harder on employees, and why discernment, wisdom, and community are what leaders have left to offer.
Transcript
Welcome back to the CTO Advisor. This is Calvin Hendryx-Parker. I am your host today. And today we're joined by a good friend of mine, Pete Sacco, who is a 30-year veteran of the data center building space. And he's actually dived in a little deeper on neoclouds. He's got a new startup called Gray Wolf. And he's the author of a new book called The Bridge. And it's kind of something we're going to talk about in the second half of this episode.
In this current rate of expansion, are there some physical constraints around power, density, and cooling that enterprise CTOs are underestimating right now? Let's just go right into it. Yeah. Okay. Data centers. Let's define what we're talking about, right? Because there's the IT data center, which ran our information technology. And then there is the space that data centers forgot, AI. It caught the data center industry napping. And so, everybody's using AI, whether it's agentically or just at the chat box.
But that's a whole different model of large language model that the new trend is going to be, hey, look, I don't want all of my private processes and all of my private information out there on the public sphere for, you know, ingestion by the large language models of the foundation models, the closed weight models, nonetheless, the open weight models. So, how do I protect myself? And there's lots of companies that are defining their space in there. But the new trend is going to be, how do I build a small language model on my protected information and then inference against it?
And I'm going to do that in my IT data center. Okay. So, data centers. You know, 20 years ago, we all had a data center. And it sat in our people space. Right, or the closet. And then people space went away. Right, exactly. And then we went to co-location because we didn't want to own the space anymore. And then we went to cloud because we didn't even want to own the infrastructure anymore. And then the gloves came off.
And today we find ourselves in an ever-expanding hybrid space. With always in there, there's always some segment of protected data that I need to have the ultimate control over. And that's still sitting at the edge. S. federal government, all the labs, or the NASAs, that's probably my biggest customer segment for whom I build data centers for. But also people like Pfizer. And then the data center operators, people like CyrusOne. The trend right now is how do I build that new edge, that new AI, and retain it for myself?
And there's lots of different ways to do it. And we can explore where that is. But that's the new trend going on, is how do I deploy AI within my organization so that I can actually train my model and inference against my secret sauce to give me that greater differentiated edge in the world? And now how do I do that becomes the question. Right. So, CTOs used to be, like you mentioned, going to the cloud, getting out of the closet, moving out of the colo.
That was a discussion that those CTOs had to have. What is different about the discussion this time for CTOs about, say, like neoclouds, for example? Because that may be the space they can go host privately the AI workloads. Yes, and they surely can. I think the problem that the neoclouds are going to be, is the overall architecture of it. I mean, there's been a lot of news right now. Elon Musk, you know, he had a not so flattering thing to say that the neocloud industry was going away.
Let's just talk about the model of how they do it. What they basically did is they rented, bought 20-year leases in existing co-location facilities. And we all know the names, right? Whether it's, you know, CoreWeave, $50 billion in debt, or Lambda, or, you know, some of the other, you know, the guys that are out there. But they signed 20-year leases on space. And then they bought and financed tremendous amounts of GPUs, right? And then they're renting those GPUs by the hour.
It doesn't take a rocket scientist to see the miss. What happens when the GPU becomes no longer is the favorite son as far as compute goes? What do I do with a 20-year lease? And I've depreciated that asset over an extended period of time, something that really is only going to have a, you know, a lifespan of two to three years. And I know I just lost a whole bunch of people and everybody at NVIDIA is yelling at me right now and everything to say.
No, but we have to be realistic. Be realistic. GPUs are not the sole foundation of compute that is going to carry us into what AI will ultimately deliver. Free energy and free labor. That's the direction that we're going. The post-scarcity world that both you and I are enamored with in the abundance community. That's the end delivery. The roadmap there, that's the part of the, that's the book stuff. It's going to be difficult and it is going to be our discernment that makes that journey either difficult or faster.
Okay. We can discover that in the second half. Right. And so, I mean, up front in like this next three to five year period, what do the, how will data center architectures evolve in this three to five year period? Right now. So, I've built that vertical stack like Elon. So, I own architectural engineering, construction, commissioning companies and power companies that I could build power behind the meter. And I think that the next bastion of data centers is going to get smaller, not bigger.
Because the next great massive AI market is inferencing. And inferencing does not require 100% GPU training stack. So, what I project is we are going to have a data center in every town. In fact, I just gave you that. That is the mission for Gray Wolf Data Centers. Data center in every town. And what I mean by that is we need to have community ownership, people ownership of the data center. Now, decentralization. I don't think it could be completely decentralized.
Let's say Bitcoin, right? I would argue that Bitcoin is almost sometimes too overly decentralized because there's no ruling body that when it does have a problem with the DAO, how does it control itself? It's going to be more like a franchise. Appropriate decentralization. Like the human body, for instance. Core brain and nervous system, that is the central coordinating thing. And every cell in your body is decentralized. I want to build the same thing in the form of data centers.
Tens of thousands of sites that are actually in their independent control. All of the local switch to turn them on and off in order to have sovereignty to know where my data is geographically. Not AWS is, hey, somewhere on the East Coast. Yeah. I'm talking real sovereignty down to an actual hard drive. But then the control of it, identification, governance, cybersecurity, all of that is highly centralized. The approach will always be the same. And we all share that same approach.
That's what I'm building with Gray Wolf Data Centers. I think that's going to leverage the community. And we're going to have two forms of inferencing. One, I call it light and one heavy. Light is going to be inferencing that's done at the AI device level. The robot. The AI, you know, the tractor, the car, the this. It's going to have enough compute that it could do some level of inferencing locally to itself. However, we're also going to have an environment in communities where we need to federate a certain amount of AI devices and robots at the local level.
And that's going to be done in a as small as a 300 kilowatt container and as large as, let's say, a 20 megawatt data center in every town. But that's well under the threshold of the one that's consuming all the resources that everybody in the public is complaining about. Yeah, that's true. So you mentioned the nervous system. And let's dive into the second part of some of the things you mentioned actually in the book is that, you know, these high performing executives probably already filled with like a bundle of anxiety and nerves.
I mean, how is the current rate of AI transformation accelerating the executive anxiety and burnout across IT leadership? Yeah, that's a great question, right? Because it is huge. And as bad as we might feel as executive, let's hold in our heart that it's worse on our employees, right? Because right now, let's remember AI, right? 10% utilization in the United States and 1% uses AI to the level that you and I use AI. 90% of America can't spell AI and it's how you spell it, right?
I mean, so there is the problem. So you want anxiety. That's the anxiety. That's when I'm writing the third book right now, which is coming out in 27 called Freed for What? What is AI freeing us for? I'm going to write that book for the general masses that, look, we're going to be all okay. But I get that and I'm going to tell you why in the book. But here for the executive leadership right now, here's the problem befriending our people.
We're telling our people we need 10 times the capability out of you and we're going to make you no longer manage just processes. We're going to make you manage agents to do the rote part of your job. What I'm learning already amongst myself is that when I strip away the 80% of my day that is rote stuff and I hand it off to the agents, you know what I'm left with, Calvin? I'm left with important decisions, one after another, after another. The hardest part.
The discernment part of me. And you know what that is? Exhausting. It's exhausting because, you know, those little easy tasks were how we unwound our stress and our anxiety, right? Now, all of a sudden, I'm stacking important decision on important decision amongst many companies, amongst many parts of my life. And I find that I'm exhausted. And so I've had to resort to other things in my life that enable me to build that reservoir of things, the things that you guys have come to know me on.
Look, I'm a 30-year veteran, you know, of data centers and AI, but I've got three little girls under the age of 10. I'm a devout meditator. I'm a devout understanding what the limitations of my human body are. And look, everybody knows that I'm the guy that talks about psychedelics all the time in order to be able to free ourselves to understand. But that being said, we have to understand the anxiety that it is building within our people and ourselves. And we have to have an outlet for that.
Because if we don't, we are not going to use the mirror the right way. We will just keep compounding ourselves as to the stress of all of this. When really, it is going to be us, beings of discernment, of wisdom and intuition and insight that allow us to get through what is going to be the most tumultuous time in human history as we transition into a world where our meaning was no longer set by how smart we think we are and what our skills were and how much we could sell for them in the marketplace.
And an understanding that the ultimate is our meaning is going to be something completely different. Faith, love, and community. Yeah. Well, you just mentioned it, too, that you suggest that human value is going to center on discernment, context, and wisdom. And what is going to separate tech leaders who merely survive the shift from those who build resilient, thriving teams over the next decade? I mean, that's going to be the real trick, right, for these CTOs who are listening right now.
A hundred percent. So remember, just 10 years ago, the model was fire your bottom 20% in order to scare the shit out of the top 20%. That's no longer. Tell me the last time that you did that. No. You're the master motivator, the cultural leader. You're the wisdom maker. You're the person that stands in the trenches with your people, and you are the representation of everything that your people need to be. That's what the new leader is.
It is no longer that guy. It's the guy who walks the walk, talks the talk. You have to be there for your person, which means that you have to be there in every facet of your life, for your wife, for your children, for your family, for your employees. So stands the man. So stands the leader. And so that's what we need to be. We need to become that person for our people. Represent. I want to be a part of communities whom I can elevate and whom can elevate me.
And I want that to be my businesses, my employees are just another part of the community in which I live. And so I want to represent all of the things that they could, dare I say, aspire to, right? Because I'm working on myself. I'm fallible. But they see me working on myself every single day, and that's the best they can do, whether that's my children seeing me work hard or whether it's my employees seeing me work hard. Yeah. You have to do it with the space to recognize that, look, we have limitations.
It's interesting because I feel like this has applied as well for using these LLMs and models, the tools. You have to kind of go back to first principles. You can't write great software without having a plan and brainstorming the specs and having the test-driven development. And all the pieces that were true in a pre-AI world are even more true in this post-AI world. And it sounds like you just said the same thing about how you perform as a leader. Yeah.
Yeah, I think you're exactly right. I mean, you know, I would say the biggest where I'm leaving the AI is the people who are saying that, and this is my just pure, utter spiritual belief. The people who are saying that AI is going to have human rights and unalienable rights that we do as a human. Here's where I give it. The new AI Netflix thing that just came out that was talking about Hugging Face, you know, every doomer that they put up there, right?
Yeah. Talked about the human, they talked about the language that they used to describe what it was they were doing. And let's remember, the bots used emotional language, not because they are emotional conscious beings, but because they are trained in using the emotional language that we use as human beings. So when we hear an emotional response from a bot, we think immediately that it has emotions. It does not. It's executing a code. It's just using emotional language to do that.
So, look, I think that we think that AI is taking over thinking completely. I don't think that is correct. I think thinking is actually two things that we represent as one. I think it's 85% perspiration, 15% inspiration. And the perspiration is everything AI does well. Search, research, collation of data, putting in a tabular, pattern recognition, right? All the things that we are never, ever going to do again faster than it. But the 15% that is left was the 15% that was the most profound.
And I like to refer to it as discernment. That is human wisdom, human intuition. The moment that we sit and we look at the data, all the important decisions that we've made in our life that have nothing to do with a calculation. When we're going to have children, when we're going to get married, who are we going to spend our days with? And they all come down to our intuition, our insight, and our wisdom. We're going to use AI as the mirror to make it better than that.
Look, you look at the youth. They're looking at AI right now, and they're already using it as their therapist. And that's okay to a point, as long as they recognize it, it doesn't emotionally understand you. It's just spitting out in response to what you are telling it, everything that's ever been written about what you just said in human language. Yeah, yeah. You understand that in that context, it's okay to do that. But just understand it does not sit in consciousness like you do.
Yeah, well, I think this episode was not only for the CTO, but for the CEO. All those words, I think, hold really true for folks who are seeing the wow and confusing it for, like you said, a conscious being or a person. You know, we've lost the right to separate that blend right now. Yeah. We are at the forefront of the connection of the deeply technical world and the world that the ancient mystics said all along, that we are all deeply connected.
The machines will never share in that deep connection that we stand as a biological entity in a universe that we can't quite yet explain. And I'm perfectly good with that because I think the AI is going to help us answer all of this once and for all. And this is the hockey stick moment in life when I'm grateful to be alive right now more than any other time. And I'm grateful to give me the platform to do it because I want to be the positive voice.
I want to be the guy out in our industry saying, look, the journey ahead is going to be difficult, but it is going to be courage and wisdom that gets us through this faster rather than slower. So thank you for the platform. That is perfect. And I think that's a great way to wrap up the episode there. Pete, I want to thank you again so much for joining me today and hanging out with me and spreading that level of optimism. I think there's a lot of people out there who are very worried and that anxiety is real.
And I think that you have the right attitude and I want more folks to share in it. com. com. Help me spread the word. Thank you, Calvin. Yep. Thanks, Pete. All right. That'll be it. Thank you all. We'll talk to you next week.