AI Safety Rules and the Future of Data Centers | TSG Ep. 924

Hey, everyone, is AI Safety and oxymoron. You're watching Textron Gang. All right, moving on a block in three, two.

Hi, everyone. Happy Tuesday. Wow.

I hope your week got off to a great start yesterday. We are, of course, back at it now as we get into the meat of the week, and it promises to be another interesting week with lots of news and the tech world and, and the greater world around it that affects the tech world, or maybe it's the tech world affects the greater world around it. Either way, we've got some good stuff to talk about and some great people to talk about it with.

Let me introduce you to our gang for this fine Tuesday morning. First of all, joining us, uh, as JP Morgenthal, jp, uh, relatively new gang member, Chaya au Chaya. Welcome.

Thanks. And of course, two old Reliables, Steven Foskett and Mitch Ashley. Hey, gentlemen, how are ya?

And Lee and not, Not so happy being called Old Man. Well, who's old and who's reliable? Was he taught?

Was that too different or were we both What? Or I'm not sure what you meant there. Well, we'll discuss it off camera.

Here we go. Everybody's a comedian on it. Tuesday.

Keep your day. Um, so, So guys, let, let's jump right into it 'cause we've got some meaty stuff to talk about today. I wanted to first, uh, talk about AI safety, which is what it kind of teased out in the beginning.

You know, we've come a long way from the, the letter signed by the hundred people that we should stop all AI activity until we figure out how to make it safe, including some of the people who are pouring money into it and going after it. But, you know, AI safety still is a thing. There's a, uh, another pass out coming outta California legislature, um, talking, you know, some of the providers are talking about putting guardrails in how well they'll work.

I don't know, but Chaya, why don't you kick us off on this? What, what are we dealing with? Sure.

So first of all, the bill is a step in the right direction. We do need cartilages around it, so it's definitely much needed. But if we go into the detail side of it, it's like essentially saying that the companies who have a valuation less than 500 million or 50 million has to regulate themselves and expose themselves in terms of what is their source of information.

And they call it as a frontier model or frontend model. So that is something needs to be explored on. Uh, it's, it, it's looking like, like we are trying, so the bigger players are trying to control the smaller players.

That's what I can think about it. And it requires more due diligence when the policies goes in life. Um, when I say I like the step, I am also concerned with, it shouldn't slow down the pace at which AI is developing, right?

So when there's so much of guardrails comes in place, it slows down the momentum of it, and we are in a changing era. So that's something we have to be careful about. Agreed.

Yeah, I, I would agree with Chaya there that, um, this, uh, some of the unintended consequences of these bills can be, as she said, um, having the larger players cement their lead and locking out smaller players. Uh, that certainly is something to be concerned about. For example, with the anthropic financial settlement, uh, that we talked about last week on, on the gang, as well as, uh, you know, bills like this one at, at the same time, of course, there's a huge political dimension here where you've got, uh, essentially two Democratic, uh, presidential hopefuls, uh, or at least political hopefuls, uh, going head to head over this, where one is trying to prove his, uh, progressive, uh, bonafide days by saying, you know, I, we need AI safety.

And the other is trying to, uh, basically, uh, prove himself to the Silicon Valley establishment and say, wait, wait, wait, maybe we don't. And I think that unfortunately, the, the world we live in, it, it often is more about politics than about, uh, the intended consequences of actions like this one. Yeah, considering I don't, so many, so much of the donor money is coming, of course, from the, the same companies.

There, there is a comparison between the last time, last year when they took a shot at this. This has got some of the teeth removed in it. You don't have to have a governor that you can shut it off automatically.

You just have to disclose whether you have one. Um, and there are less strict reporting requirements. There's still some of the same safety requirements.

So it's, it's a little bit lessened, not quite as sharp teeth, and we'll see if that's more palable to Newman. I, I, I view this equivalent to the SEC, right? Where, you know, there was an option, either you police yourself or we will institute, uh, a rule codify the policing actions out of the government.

I don't think politics is the right place to attempt to police, uh, emerging technology. Um, but what they're saying is they're not seeing the industry police itself enough, which indicates that maybe what we need is a third option here. We need an organization that is supported by these, by the, the big tech companies that they're willing to delegate and submit to, uh, in order that, that's equivalent to the SCC that's going to, uh, ensure the openness, uh, and the, and the, and the safety of what these companies produce.

Uh, and I haven't seen that emerge, and I'm surprised that that hasn't been an option put forth and accepted it. It's something they could drive very quickly. If you could get, you know, Microsoft meta, Google Anthropic, uh, open AI to, you know, accept and adopt an a, a particular organization, um, to act in this role for them.

Well, I, I've got some views on this, not surprisingly. First of all, everything's political, right? You're not the amount of money at stake here, right?

Let's be real clear, the amount of money at stake here is, is literally the pot of gold at the end of the rainbow, right? And so with that amount of money, there's gonna be a lot of, a lot of players, right? Political, nonpolitical, corporate, individual, what have you, uh, social justice warriors and everything else.

So to to, you know, it's naive to think that we could somehow keep politics out of this. We're not. Um, what's interesting is, as in so many other things around technology, policing, environment, environment policing, et cetera, we lack the will it seems, or the consensus on a federal level to get something done for the us right?

The eu, they passed something almost a year ago, right? It was when they started their AI safety bill. Um, so in the absence of a national political will, certain states take the initiative and take the lead.

California has been a state that's been doing that for years. California privacy laws, the California laws on, on fossil fuels and car, uh, mileage and so forth. You know, they, so California has a history of, of being out frontier, right?

They, they have the will to do it. What we don't often see is, is I think is Steven pointed out the two facts in California, for all intents and purposes, is kind of a one state, state, a one party state, right? But you have two powerful players there.

And, and, and it's playing out. It plays out on the background though, of the tech bros, right? And, and what do they want to do?

And of course, they probably don't want formal government regulation, they prefer self-regulation, but they can't even get their act together to do that, right? They're all busy suing each other and everything else. So don't look to them.

Now in income's, the white knight here, if there is such a thing, and it's AWS they're saying, hold on, hold on. We could put some guardrails in here and, and make it safe for everyone. I don't know.

Is that, is that putting the fox in charge of the hen house? Maybe Everybody has guardrails though. I mean, they all have guardrails.

The the question is a do they really, or do they just say they do? No, they, there are guardrails. I mean, there, the, the challenge, and this is why I think a third party organization is required, that it's, you know, how, like anything else in software engineering, you have to test it, and you have to be willing to break your own code, so to speak.

Um, biggest problem in software engineering is QA and, and unit test engineers at unit test don't know how to ba break their own code or aren't willing to break their own code. So no, no, But, but jp, when you put out bad code and your app is bad, there are consequences, right? You're at, no one wants to use your app.

You've got, you get sued for security problems, et cetera, et cetera. Who, where is the teeth of a th how is a third party gonna enforce AI safety? What it would, How Do they enforce That?

They would have, yeah, so it's what I said earlier, right? It they have to be the SEC of this industry. They have to be accepted and contracted that Those, so you, but let's, the SEC is a quasi-governmental Yes.

It's chartered agency. And that puts The federal government isn't chartering an AI commission. I don't see it happening under this administration, that's for sure.

Yeah, it's required, but yeah, to Steven's point, not happening. Well, yeah, just like it should happen doesn't mean that it will happen. If you look at what Amazon announced with bedrock guardrails, those, that's at a much, much finer grain grainer of, of guardrail.

It's things like, uh, the example they gave in their block blog was, here's the document policy document that you use to say whether someone can get a mortgage loan or not, right? That, that's, so it's like no knowns of, of safety, if you will. This is really policy enforcement guardrails more than safety, safety guardrails for you to create something in bedrock using that facility.

I'm guessing that's probably not a comprehensive enough abil capability, or you'd have to cover so much, so many areas. It's really the model creators that have to put in the, the guardrails into the safety, into the model, as opposed to every third party they can check it, but making it safe, that's a tough, tough nut to Crack. Well, oh, Mitch, you raise a great point.

And actually, this is a question I had in my head before we even started segment. Let's define safety here, right? To me, I think with regard to this technology, the most important safety, uh, mechanism I want checked here is can the LLM be used to hack?

Can it be used for cyber crime? Can it be used for nefarious purposes? I, I, I think that is one important thing, but I think, believe it or not, jp, there are more important things.

We discussed it on a show last week about the mother who wrote the essay in the Times about their daughter who killed herself, SU su committed suicide. And, and I, you know, I put that, I wrote an article on that with Bill Brenner, a good friend of mine who's big in, uh, mental health and tech. And we got a tremendous, you know, response from the community.

This evidently happens a lot, and there's a lot of lawsuits pending right now, because people are starting to rely on AI for therapy companionship, right? You know, this is on a personal level, we're talking, I'm not just talking hacking, I'm talking life and death. Yeah.

But isn't that just gonna be re resolved by a disclaimer that pops up on the, on the chat, on the gp, on the chat interface? That's is, Hey, anything, well, anything you learn here is just the purpose of the entertainment, blah, blah, blah, blah, blah, you know, and that, that's it. Now I move On.

I, I, no, because they're actually creating AI therapists and AI companions, and, you know, and, and, but you bring up an excellent point, jp, I, you know, looking at it from a legal angle, right? There's two ways that we get kind of Clarity on, on, on the legalese here. One is by statute, you know, by regulation, by, by some sort of governmental action that clearly defines what the liability is and what the penalties are, right?

What's allowed, what's not allowed? And, and what, what are the repercussions In the absence of that, we fall back, or at least here in the US except for Louisiana, we fall back to common law, and it's, it's basically tort law, right? With, uh, uh, you know, reasonable mis test.

So is it reasonable, jp, if I clicked on the click that, you know, that stupid thing that we click on every site letting us know that cookies are involved and they have their own policy without reading it, and then, you know, some AI therapist tells my kid it's okay to off yourself or whatever. Yeah. Is, is that enough to, is that reasonable?

Well, also, I think it's important to recognize that a lot of these guardrails are themselves AI models. Yeah. It's turtles all the way down.

And so if you have an ai, like I, I applaud Amazon's bedrock guardrails on a, on a technical level, because essentially you can say, look, if it gives me some information, verify that information, you know, if it gives me a fact, verify that fact, I like that idea. The problem is then we need a guardrail for the guardrails, and then we need a guardrail for the guardrail guardrail so that we can make sure that, you know, 'cause that that one's not hallucinating in the next one. But I actually want to, you know, kind of circle back this whole conversation to Chaya's first point, which is, does this lock out innovation, whether it's guardrails or an SEC for AI or, or whatever it is, i, is that going to make sure that no company can compete with the magnificent six who have built large language models, Right?

So what I can add to that is if you say, why do we even need guard rails? We need guardrails. Because as users, I want to know the information that the model is giving.

What is the source of it? And is that source legit? Is that source valid enough to certify that here is information before AI error?

You can go to the a website and see where you are reading from, and you know, if it's Wikipedia, is it something else? Now we don't know. So we definitely need guardrails.

And here Bedrock is saying that there is an agent world where you can automate lot of things, but there are knowledge bases and policy documents, which we need to have a manual interaction to set up those rules. And that rule, that is where the critical piece comes in, that companies are allowed, the users are allowed to say, here is what the source of information that I can trust on. Here is what my model is going to read on, and here is the information.

But like Stefan said, that we need to make sure this shouldn't slow down the pace of development, because again, we are adding a human angle here. You know, I think e essentially, aren't we making a decision? Are we going to, basically, to your point, Alan, about tort law.

This is gonna be solved in the courts and, and all the model makers will put every, you know, liability disclaimer that they will into their policies, of course, if possible. And when harm happens, it gets resolved in the court. Or there's some things, so potentially egregious with ai, which I think some of us think that's true, like gaining power, physical safety, things like that, um, that might benefit from some either oversight, maybe regulation.

So I think you start with oversight, like what are the kind of things that are happening? Or does it warrant that we need some regulation? Maybe it's a light touch, maybe it's a little heavier hand.

But if you put a big heavy hand regulation out there from the beginning, that's potentially going to, uh, you Know, impact the innovation and The speed of the market. I, I, I agree with you, Mitch. I, I think in the absence of regulation, in the absence of industry, uh, oversight, you know, taking care of their own, so to speak, in the absence of all of that, the court fills the vacuum plain and simple.

And, um, that's, that's just the way of it. It's not an AI specific thing. I think that's kinda how our society functions, right?

The absence of everything else, the court becomes the arbiter of, of reasonable, of allowed or not allowed. Um, it'll be interesting. And, and, and of course this does, as Steven pointed out in the beginning, I want to come back to that.

This plays plays off against the backdrop of people jockeying for the next presidential election. And, you know, depending who wins and what policies are jp, we may very well have an A-I-S-E-C, right? We could hope.

But, um, it remains to be seen. It remains to be seen. Let's take a break.

We're gonna come back and talk a little bit about AI infrastructure. And Steven recently had AI infrastructure Field Day. We'll get a report on that and some more of what's going on around super intelligence.

You're watching Textron Bank, Discover Textron Group, the epicenter of tech innovation. We are your go-to for reaching IT, leaders and practitioners worldwide. Our secret impactful content that sparks awareness, engagement, and top quality leads with us.

You'll access editorial websites, streaming videos, virtual events, custom content analyst research, and more. Join our satisfied clients. Let's revolutionize your tech journey.

Contact us today and tell your story to the world in the most powerful way with Textron Group. Welcome back to Textron Gang. Uh, I'm Steven FoST and I run the tech field day business unit, uh, sort of sister company for Techron.

Uh, you, if you were watching last week on techron tv, and really, if you weren't, why weren't you? Uh, you might've caught some of the presentations from our AI infrastructure Field Day event that was live streamed here. Um, I, I wanted to bring some of the takeaways from that event to the, the gang this morning.

Now, first off, I, I should point out, um, you know, we had, uh, six companies present, uh, Broadcom Hammer Space, Rafa Mortis, Satera, and HPE. Now, I don't want to necessarily focus on those companies specifically. What I'd love to do is try to pull together some of the threads from the conversations that we saw last week.

So let me put a couple of things to the panel here. So three things first, um, one of the unifying messages throughout those presentations was the difficulty of moving generative AI applications from experiments into production, and specifically how to physically infrastructurally, you know, nuts and bolts, make these things run in a way that enterprises want to be able to make these things run. This is especially critical because all enterprises are investing in AI and trying to figure out how AI transforms their industry, and yet we don't really have a consensus on what that infrastructure is going to look like.

And that leads us to the second point, which is build versus buy. Do you have the resources? Do you have the time?

Do you have the interest in building your own AI infrastructure, or does it make more sense to just use a service to run it in the cloud or to find some sort of SaaS that can run it for you? Uh, and, and also, what are the implications of doing that? If you do decide to run it as a service, uh, does that sacrifice some of your ability to control access to your data and so on.

Which leads us to the third point, data management. Now, we all know that it's garbage in, garbage out when it comes to AI models. And, uh, this is especially true when you're looking at smaller AI models, because most enterprises aren't going to build their own chat bot.

Instead, they're going to leverage a foundational model with their own data and sort of a rag situation, or they're going to build a special model. I actually wrote about that on LinkedIn last week after talking to another company, uh, called Articulate, um, which hopefully we'll see at the next field day. Uh, and if they build a specialized model that makes the data even more important, because that is going to be a very limited data set for, for a model.

So lemme throw this out to the panel here. Um, uh, jp uh, you know, you're somebody who I very much respect on this AI topic. What do you think about the challenge of enterprises trying to put generative AI models into production?

I I think it's a spectrum. I actually have written about this too. Uh, it's, uh, probably, I think it's a LinkedIn post I have out there, right?

I, I think it's a spectrum. I think there, what I, the way I like to think about it is that you, your organization and, and businesses are gonna learn this very soon. Your organization is so unique in certain areas that to get to the quality outcome you want from a model you are going to have to use, you train it on your data, right?

It, it, it's not, I mean, while there is some amount of training you could do on the larger ones, your data's still getting lost in the corpus. That is the universe of data that is, you know, used by those models. So if you have stuff that is very specific to your domain, then it makes sense that you would want to build and, and host that model for yourself.

You could host it in something like Azure, uh, or AWS, right? And in a safe, private, you know, compartmentalized way. Um, so you don't have to go out and get the infrastructure to necessarily build that out per se.

Uh, you know, obviously for some organizations operating at the edge is important. Getting small models operating at the edge is in architecture, it's required in certain manufacturing and other areas where they're starting to introduce ai. But let's just for now say that you go to the, um, that the cloud is suitable.

You, you, so your domain specific model, but then there makes, uh, there's a point at which you want to add in the universe of data, right? And that's where it's almost, it's almost analogous to being able to burst out, like we've always talked about in cloud computing. I have my local resources and you know, VMware, and then I want to be able to burst out when I need it to the cloud resources, right?

And it's similar kind of architecture. I have my domain specific stuff local to my business, and then when I need the corpus, uh, that is the world like, okay, but why is this pattern happening? Well, it could be geo, uh, you know, socioeconomic, it could be something in the news that happened.

I don't have that data in my model. I need to go to the larger corpus to go get it right. And that to me is what I see as a complete intelligence model that needs to be understood and built by companies.

You know, jp, I think that's both you and, and Steve are direction and correct. Uh, my sense is that for most organizations, building models is at least, uh, LLMs is, uh, you know, out of their reach, right? The cost and the effort and the skills to do that.

Probably the next step down or step to that is of course, we all are familiar with the rag, and that is a way to provide information to the model that it, it you ingest into the model and have it analyzed. I think we've entered a new era where MCP servers are the new rag because I mean, you can, anybody can create an MMCP server. Matter of fact, there are online services that will built on for you, and that's a way, if you start to put in filtering and sort of guard rail rails or logic of how to use that MMCP servers to your information, your data, et cetera, so you can curate what it has access to, it really gives you the access to anything.

Now, you don't have to move data into some rag location, and you can maybe use a combination of both to get access to your data and then use the general reasoning or processing capabilities of the model. The thing that I want to add on the infrastructure side is there are a shift in the technical skillset that is needed to maintain this infrastructure, right? The typical skillset of a DevOps and SRE team, which been managing the infrastructure versus managing the AI infrastructure needs some additional skillset, which gives us people an opportunity in terms of directions of career opportunities to explore, right?

There are a lot of, um, thought around the industry that AI is going to take away job, but here, there is this direction where every company needs their own infrastructure or maybe some sort of, if they go to a cloud provider. So this gives them an opportunity to see if new skills can be leveraged to maintain this, um, infrastructure. There are two aspects of this.

One is where, um, the model is getting learned, and the other aspect is where the user experience of it, how slow or how fast your internal tools work versus the outside world enterprise level solutions, right? So that is something I wanted to add on because the thing is, um, we built services on CPS and rags, but this field has been changing. Now I just read this, something came out as fast MCP last week.

So we really need to make sure the skillset that we are hiring is up, up to the mark in terms of the things needed to manage this infrastructure. So, so I have two points. One is you jaya around, uh, skills management.

I, I, I, I think you're onto something really interesting there, uh, because I think that's part of, uh, the cost differential be in, in using these resources and getting these resources as in an outsource model like the cloud, right? It's, uh, if you were to compare a general virtual machine that you might take from Amazon, um, a against what it costs for the same amount of time for a GPU based machine, it's significantly higher on the GPU machine. And I think part of that is both the hardware is expensive to, to maintain, and that there is additional skills and labor that are required in order to build and maintain and operate that cloud.

Now, Mitch, as to your MCP point, I, I, I, I don't necessarily agree MCP is your data. Think of it as just a, a, a way to tap into other data sources, but that data just gets pulled into the context window. And what I've learned working with these things is that, um, you know, right now there is a, a huge problem with context windows as they get larger.

I have found that stuff later in the window is obviously, uh, paid heavier attention to than stuff earlier in the context window. So pulling in additional data doesn't necessarily get you to, um, where you want, if you were to have, if you have a domain specific requirement, um, a a better solution would be take an existing small model and then, you know, and then retrain that using, you know, uh, techniques like Laura, uh, which allow you to change the weightings. It's not as expensive as building your own LLM, you're basically just take, you know, affecting the weights of an existing model, much more effective, much equally effective and much less cost, uh, to, to be able to, to do that.

But then your, that LLM is specific, or we call them sometimes s SLMs, small language models are very specific to your task and your domain, whereas I, I, again, MCP is gonna allow you to reach out to a world of data, but a, you have to know what that data is, and you have to know how it it fits into you. You're basically providing it as a tool, and you're saying, and, and the LLM might use the right tool, it might not use the right tool. You sometimes you can force it to use your tool by saying, use this tool to do this, but then you are basically just, I mean, all it's doing is an a, a, a natural language app program that you're writing, right?

It's just basically, uh, use this integration component to go get me data and then do something with the data and, you know, and then, and you end up with the same result as if you had put that data in a CSV and uploaded it or attached it to your message. So I think, I think the viewpoint I'm trying to express JP is, is it's a matter of degrees, right? Maybe the largest organizations could, could, you know, modify the weights and have the sophistication enough to know that what they're doing and changing the model in that way more or less go through the much more expensive process.

What I, what I'm saying is I think the market is going to innovate and create ways for us to not have to train models specifically on our data. So for example, um, I mentioned MCP, it isn't, isn't just giving a model access to, uh, services that are tools that are available through an MCP server. It can be a, it can be a specific agent that's built, that's trained that is d directed on a specific kind of task and how to use the data that's coming from an MCP server versus a general model.

I, that's one of the resources or tools I can decide to use, and whatever's in that, in, I can do whatever I want with it. Mm-hmm. So I think there's ways to direct how those resources are used, whether it's MCP or through a RAG or some other option.

I think we'll see more innovation in this space. So maybe there's a, an advancing CP server that's got more, uh, direction and guidance to it of, of how you, and the ways that you want that information, uh, to be used, right? And or some other way like that.

Right? And just to add to that, your point, Jason, jp, when you said that GPU versus your own machine can have performance issues, I do want to add with the SLM, the small models you can even develop on your own machine even to get started for A POC with the tools like MCP. So there is a fast pace coming in in terms of, do I really want this?

You don't need to first buy expensive GPUs, then experiment around the infrastructure, then build your servers. It's, it's can go in the other direction as well, whether you first figure out, this is what we want to build as an organization, these are the tools, MCP versus RAG versus any foundational model. Let's see if this is what we wanted.

And then go and evaluate the cloud providers where you want to get a GPU based infrastructure. You know, listen, listening to the three of you talk, I, I have one thought, what an opportunity, What An, what an opportunity to go, this is what people want, but they're not gonna be able to do the full Monte. Let's figure out something that gets them that easier, cheaper, and I'll go start a company around it.

And isn't that what makes tech go round and round? You Got it. Absolutely.

And that's, It's plenty of opportunity, Steven, That that's the most interesting part of this, is that, you know, whether it's a conventional infrastructure company like HPE or a more DevOps focused company like Mortis or a new startup like Refa, um, they're all trying to figure that out. And I love to see that kind of innovation coming from these companies. And, you know, that's the excitement of this space.

So, um, as we wrap up here, I'll just point out that we are doing another, uh, AI Field Day event. Uh, we'll actually be back, um, with our full on AI field day in, um, October, the end of October, the 29th and 30th. So, uh, keep an eye on this space.

Uh, obviously we'll be talking about it here on Techron Gang, but we'll also be live streaming more, uh, tech Field Day focused on AI in the, uh, next month. So thanks, uh, thanks so much, uh, uh, y'all for this, uh, really interesting conversation. Absolutely.

All right, with that, let's take a break and we'll come back and talk about, we're going to get local, local, and, you know, all politics is local, right in Steven's backyard, or maybe not in my in Steven's backyard. com is the leading resource for news analysis and education on challenges facing the cybersecurity industry. com covers all aspects of cybersecurity, including data security, DevSecOps, cloud security, application security, network security, security threats, and more.

com has the largest selection of security content featuring breaking news, blog posts, podcasts, and more. com, home of security bloggers network. So the Columbus NIMBYs are pushing back about data centers near them, what, why, you know, they could bring jobs.

Maybe it glows at night with the gigawatts of energy that these, uh, centers use, perhaps. But what else? Steven, you know, because we're talking about Columbus, you are our resident, what's the word?

Ohioan? Ohioan. Ohioan, yeah, sure.

Okay. You're our resident Ohio. And why, why, what's the matter with the good people of Columbus?

They've had data centers there forever. Yeah. And they got a lot more now.

Um, yeah, so first, just a little geography lesson. So, uh, Jerome Township is the one that has voted on this, uh, that is, uh, right, uh, basically it's the township outside of a town called Dublin, Ohio, which is one of the richest and most, uh, prosperous suburbs of Columbus, the capital. And so it is really no surprise that this is where the nimbyism would start.

Um, but I, I don't really blame them. I don't, you know, NIMBY always sounds so negative. Uh, the truth is, they're building data centers like crazy in Ohio, um, in, uh, around Columbus especially.

There's, uh, you know, the intel fab that we've talked about a lot to the east, but then there's also, uh, just a whole belt of data centers, especially north of town, which is where this is. And those data centers are really causing some havoc. Now, we talked last week about the fact that it's driving up energy prices, not because it is, uh, necessarily outstripping the ability of utilities to provide energy on a regular basis, but because of the extremes, essentially they, that last watt, when it's, uh, extremely hot or extremely cold, the data centers, uh, are bidding up the cost of that tremendously, which raises everybody's utility bills.

Uh, Ohio is a, is a state that has a lot of, uh, a lot of water and a lot of land, and a lot of people and a lot of electricity. But, uh, all those things are being tapped out by these data centers. And the, uh, the voters are reasonably starting to say, wait a second.

Um, is this really gonna bring the economic benefits? Because again, these are data centers that are often subsidized in the form of, uh, tax reductions, uh, subsidized in terms of allowing companies to come in here with, uh, lesser regulations and, and, and hopes that somehow they will contribute to the local economy. But people are starting to realize that in many cases, just aren't that many jobs in these data centers.

Um, at least not that many, uh, you know, good jobs that are gonna be taken by locals. So the question is, why are we building all these things, especially in, uh, leafy suburbs, uh, outside of town? So, uh, you know, this is not just an Ohio problem.

I think this is coming everywhere. It is. We we're seeing this across the nation for the most part.

I haven't heard a lot of pushback in West Texas yet, but, um, it's, you know why? I mean, why you, one is Steven, you're right. The, the economic benefits maybe are not all they that we were sold, right?

There's not a lot of jobs being created. These, these data centers are largely on autopilot, and they could be managed remotely. The jobs don't even have to be local.

Secondly, it, it, it puts a tremendous burden on our energy infrastructure. Our water cooling in, they don't really use water as much, but our cooling infrastructures and, you know, just all of our infrastructure, um, that, you know, depending, again, you know, I only get into a red state, blue state thing, but, but depending on your political persuasion means drill more, import more shale oil, or heaven forbid, use more solar and wind power or, you know, renewable sources. And, and this is, you know, this is state by state, county by county.

We're gonna need to make decisions there, you know, and, and, and it thrusts the utilities into it too, right? The electric utilities become key players here, because if I'm selling you all of my bandwidth, in essence to run your data center, do I have enough to keep people's houses lit up? And that, that's, that's when stuff gets real, right?

If you're telling me I gotta do a brown out, because that data center down there is sucking up all my power consumption, well, I could tell you where I want that data center. Yeah. Well, I, I should point out too, that Ohio's, uh, electrical utility is notoriously corrupt.

Uh, if you wanna look up first energy, you'll see some h hilariously brazen bribery schemes and attempts to influence elections and things like that. Uh, they're the ones providing the energy. What Could go wrong?

What could go wrong, You know? And I think what I want to add here is, think about it, data centers is the source for startups around the ecosystem to be built. There is a reason why there are startups in places like Bay Area or in Seattle because the accessibility to the data centers and the performance that they get relatively better.

So, uh, this, this location such as Ohio gives the opportunity to build that ecosystem. And also it may not require a large number of technically scaled workers. It still needs specialized knowledge visit.

So the technology get diversified across the country to manage the infrastructure. It gives a central point for the startup because it's harder to build companies around in states like California or others, which are expensive in terms of taxes, in terms of other compensation or the minimum wages companies has to do, versus states like Ohio. And others would give that benefit in terms of performance on the technical side and also the, uh, the technical resources they would need.

Uh, Ohio has, uh, couple of number of good companies. You say GM is there, aviation is there. So there is a good amount of technical talent which hasn't explored themself around those basic companies.

Yeah, this is probably a very naive statement, so I'll admit that upfront, but it seems like some good zoning laws might help with this, right? If you're building your data center next to my neighborhood, that's, that's a problem. If you're building it in the, in, in the industrial area, you've created an industrial strip wherever, where you have the target distribution center in the Costco, whatever, and the, you know, Amazon Center, et cetera, and you putting data centers in those areas, then I think's gonna be more acceptable to people.

Now, we may have that flexibility either because of cost or availability of land, et cetera, but maybe that's not the right place to build anyway. But is this in sensible zoning gonna help with this? Well, the problem is that our zoning rules are all local.

Uh, in the United States, there is no such thing as even statewide zoning, uh, let, let alone national. Uh, and so that leads us to some really strange situations. For example, Houston, where I used to live famously has no zoning.

So you can build anything anywhere. Um, places like Dublin, Ohio have incredibly strict zoning. The little town that I live in, you know, regulates what color you can paint your house, let alone where you could put a data center that's Boulder, Colorado.

Yeah, we have one of those too. So yeah, it really matters, um, on a lo on a locality by locality basis. And I honestly, I feel like this is one of those things where, you know, America, America is famous for, um, devolving decisions to local authorities.

And that seems to be one of the things that the, uh, uh, the Republican party has long espoused as well, that all decisions should be local. The problem is there's an awful lot of localities, and all you have to do is find the one that's gonna allow it in order to, uh, really kind of change the whole map. Well, not only that, you know, you can go local, local, local as, as we, we are doing in this country, and then you wind up with this patchwork of inconsistency.

Mm-hmm. Right? And it's also, I will point out getting back to First Energy, very easy and cheap to bribe local government officials.

Um, yes, they were convicted of that. We'll leave it at that guys. We are right on the, uh, mark here to wrap up today.

This is a great discussion. My, my purpose, I'm gonna end it with this. My personal opinion is progress.

No man can stop progress and no local township can either. Data centers will be built where they make sense, I guess. But, um, Steven, Mitch, JP Chaya, thank you so much for joining today.

Thank you for watching. As Steven mentioned that, uh, tech Field Day, ai, tech Field Day, it is available on the, uh, tech Field Day YouTube site. It's available on Techstrong tv, as well as the tech strong TV OTT channel.

So you can get it on Apple, Amazon, Roku, uh, iOS or, or Google. So check it out there. As usual, we have a full text from TV lineup immediately following today's show, so stay tuned for that.

We'll be back tomorrow with more great talk and more great gang members. Until then, this is Alan Hummel. We're out.