Techstrong Gang – August 14, 2024

Hey guys, I'm Mike Baard. Today we're talking about AI sticker Shock. We're talking also about the implications of data gravity.

And finally, an AWS Misq. You're watching Textron Gang. We'll be back in a minute.

All right, folks, and we're back in. Our guest turn today are David Nicholson, who joins us from the Futurum Group, and also a member of the Wharton Business School there. And he's out in Northern California.

David, welcome to show. Happy to be here. All Right.

Also joining us from Denver, as always, our own. Mitch Ashley, who works for both Futurum and Techstrong group as our CTO. Mitch, how you doing?

Hey, everybody. Good to be talking with you, Mike and Dave and everybody. All Right.

And finally we have Amanda Razani, who's still down in Texas. How are you, Amanda? Doing well, thank you.

All right. Let's jump into this because a lot of folks are now starting to get down the path far enough for Gen ai that they're really starting to figure out the cost structures here. And when you rely on a SaaS application or a SaaS platform of some type, um, you run into this thing called tokens that is the pricing mechanism.

And you get charged on the input and the output. And a lot of folks are starting to say, Hey, this adds up pretty quickly. In fact, we now see, uh, IBM is offering people an option for an inference engine that they call lightweight that you can deploy on the, on your own cloud or in your own data center.

The idea being you don't have to worry about the token based pricing 'cause you'll have more control over your infrastructure. David, are you starting to hear from folks about the cost of ai? Is this starting to resonate where people are going, Hey, maybe we can't afford this?

Absolutely. Yeah. It's, it's, uh, it's a concern.

And, and I think sticker shock is an interesting way to talk about it because this is almost like unanticipated to the nth degree because people have no idea how these things are going to be priced. Um, one, one development I've seen in the market, however, is, uh, taking a page out of the whole idea of multi-cloud and arbitration between clouds for cost and performance and things like that. There are companies out there that are positioning themselves already ahead of the game as an abstraction layer between the end user customer and those LLM models as an example.

But those AI capabilities so that you can then, uh, subjugate all of these tools that are being developed now. And you can pick based on cost, based on fit for function. That's the promise.

It's the universal remote control that subjugates all of these other things. We'll see how that works, but just the fact that that exists is an indication that yeah, people are trying to get their, their, uh, their arms around cost containment. Mm-Hmm.

Mitch, we've talked in the past about, um, the revenue flow for a lot of companies that were supposed to be making a lot of money in this space and the disappointments in Wall Street. Um, do you think that that has a lot to do with this whole issue of the cost structure and people are kind of sitting back on, I can't really launch 20 of these projects. Maybe I gotta pick two or three and hope for the best?

It's, uh, the AI exuberance is, uh, definitely hit us all right. In some more than others. Some people, some companies laid off people thinking, I won't need those people anymore.

And now they're having to lay off regrets. It's, we're, we're at the brute force stage. You know, when you think of the large LLMs, the chat GPEs of the world, those are trained on massively different topics.

Everything from code to what's, you know, written, written documents, et cetera. And they take many, many resources to be able to process that kind of data. Tokens essentially are whatever you're feeding into an LLM gets tokenized.

And then that's what's connected to say, what are the relationships between all this stuff? And let me pull it back together. So it takes literally thousands, not tens of thousands of tokens to, for a response for some of the responses that we ask from a model.

Um, David's point is a good one because even some of the technology vendors, I know Salesforce does this, some others where they put a broker in front of, uh, multiple models, LLM models, and take parts of whatever the request is and hand it off to the appropriate model that can answer that question or provide that information. Um, so my, my point about the brute force sta uh, stage is yeah, if we stayed where we are today and ran those large models for everything, yeah, it, we, we, we will never afford it there. It's just a, you know, a sledgehammer when we need a, you know, a pick for a specific task or specific need.

And, and that's what a lot of the, let's go with, you know, small language models. But I think we, we also don't want to ignore machine learning. Um, I was just at Black Hat last week, and there's definitely a turn to going back and looking at how we can use, uh, machine learning to mine the data caches that we've been gathering in the security world for, for a SecOps and many other purposes.

A lot of companies are turning back to that problem and trying to use AI to do some fruitful things, which isn't nearly as, as compute intensive as gen ai. So I, I was gonna mention, we've posted so many articles on this topic, urging business leaders to take the proper amount of time for planning and research and speaking with various vendors, determining what do they need, a large language model, a small model, a boutique model, and, and not just going with the first vendor really doing their research and understanding the problem they're trying to solve with each different solution. To that end, um, we see all these reports and, um, David, I'll throw this your way about, you know, all these data centers that people are building out there, and there's this whole level of investment in this space.

And yet, to Mitch's point, if we're gonna run smaller, uh, language models or what some people call t-shirt sizes now, small, medium and large, um, are we over investing in the infrastructure and the assumption that everything's gonna be these huge, large language models that may not manifest themselves on the infra side? Yeah, that's a great question. I, I would, I would, um, uh, Amanda was mentioning something that kind of brought a visual to mind, um, along with what Mitch was saying.

When you, when you think about generative AI and the resources necessary to do something, um, not everything you want to do requires that kind of work. Look at it this way. Um, what, what do I need?

I want a picture of a cat. Well, I can type the word cat in and get a million pictures of cats, and one of them might be perfectly fine. Whatever the resources are involved in that retrieval, you know, just doing a Google search, uh, versus I want you to create a photorealistic image of a cat.

Wow. Look at it. It's almost as good as a picture of a cat.

Well, how much did that cost? Well, that cost me $3. Well, what was the, what was the, what was the old way?

How did, well, how did, how much did the old way cost? Uh, oh, that was, uh, nothing. Essentially, nothing so small that we can't measure it.

So I think what we're gonna realize is that, um, frankly, there isn't, as, isn't as much unique stuff in the universe as we like to think. Not everything needs to be bespoke and generated out of nothing every time. At least based on resource constraints today.

And, uh, and frankly, I go back to this over and over again, it, this is all gonna be about the cost in terms of energy to drive these things. There's a dollar cost associated with these things, but really the do, the dollars are going towards the cost of energy. And at 6:30 AM West Coast time, I checked just because I wanted to see, I live in California.

We have something called California Independent Systems Operator. They're the broker for everything electricity, uh, in, in the west. And I check to see 20% of the energy going into the grid was considered renewable energy at six 30 in the morning.

Why? Because solar is not up and running quite yet, because the sun is not high enough in the sky. What's my point?

We're in California, we charge 63 cents per kilowatt hour for electricity, um, 10 times as much as some other states. Yet 80% of our power right now, which would have to, which would be going into a data center, is non-renewable. So there's a huge misalignment of, of sort of energy policy goals and the idea that we're gonna have power for gen ai.

So, might, might seem like a little bit of a drift, but I see all of this, every time I hear data center, I think electricity. Where's it gonna come from? I'm gonna have to choose between charging my Tesla or running the Gen AI program, is what you're saying.

Cover your roofing panels. That's what I, There's a whole nother part to it. You, you bring out some good things about energy.

Um, we, we tend to apply things on the intel chip model, you know, how long this current platform that Intel's been leveraging for CPUs and, and its lifecycle. Um, you know, Nvidia comes outta the gaming world where chip sets evolve every six months and they're coming out with new platforms. I mean, they have D-G-X-H-E-X, the Blackwell platform.

So I think one of the overinvestment risks is in two years what you've invested in is, you know, seriously out of date, at least for the kind of state of the art. And also it's very expensive to buy the kinda latest, highest priced performance chip set. So it's, that's a, that's a delicate balance.

How much you're gonna spend on the best technology or the current technology, and how long, what's the life of that which drives up the cost? That's why some of these costs are so much, 'cause the, the lifetime value of those chips are shorter and they need to replace with the next generations. 'cause that's what people want, right?

They don't want on, they don't think they wanna run on the old stuff until they get the bill. Mm-Hmm. Mitch, to that point, there was a, there'll be an article on tech strong AI about this with talking to IBM about, um, this very issue.

And they were pointing out that what they're seeing now is that the responsibility for the inference engine is moving over to centralized IT and DevOps to kind of help reduce those costs because they have a better sense of what, uh, models actually need a GPU versus some other processor class. And that's one way people are reigning in their costs. Mm-Hmm.

Inferencing versus training and, and using kind of other parts of the GP or much sort of the use use of it versus the analysis and training of it. And you bring up a good point, which is it organizations are used to tuning and getting the most out of their resources and lifetime value of their investments. Um, and they don't even have to be on GPUs.

It could be on CPUs as part of what people are looking at is, is this really a GPU problem? Or can I get the same kind of performance on current generation, uh, CPU technology so that, you know, we're at such a learning stage about how to, how to you both apply ai, especially generative ai, and also what are the, what are the non brute force ways, you know, smaller models, models running on our phone, things like that, running at the edge. So there's a lot of, lot of, um, I said a lot of learning to go and a lot of experimentation to go, um, and to, to, uh, Amanda's point, part of talking to multiple vendors is learning from them, right?

What do they all think the strategy is that you should be employing? And of course, they're gonna have their version of it that, uh, is their products, but you wanna learn from them. 'cause they're, they're spending a lot of money learning it, learning the market themselves.

David, there was a time when the data science people said that they were just too cool for school, for those IT people, and they wouldn't need them, and they would just kind of take care of all this themselves. And lo and behold, it starts to seem like they're knocking on the door of the CIO with hat in hand. Is there some irony in all this?

Uh, a bit of a bit of irony, but, uh, you obviously, you're referring only to the data science people who were able to reinvent themselves as AI people, because in a lot of large organizations, um, the, uh, AI sort of leapfrog over data science and, um, and, uh, the folks were, who were rewarded with the coveted AI positions didn't have any, uh, data science background. So, so there's a bit of a, a bit of a dilemma there. You know, uh, Mitch, Mitch Min mentioned, you know, the ever present difference between, uh, inferencing and training.

And, uh, we talk about the, the, the latest greatest, most expensive, most resource intensive, um, AI hardware. And it's largely thrown at training because of this whole idea of time to market, how quickly you can train a very, very large model is critically important. Someone quoted to me recently that 80% of AI spending was essentially directed towards training now, um, 20% inference at the edge, and that that's expected to flip relatively quickly over time.

What does that mean exactly? What does that, what does that look like? And I came to Mitch's point about these deployment cycles where we're used to call it an 18 to 36 month refresh cycle.

Um, I know some big OEMs are still selling n minus two generation servers as their main product. They're selling because people are saying, Hey, I don't need the latest and greatest. But when it comes to GPUs, people are graph taking.

So what's Nvidia going to do? Are they going to push, I think they, my prediction is they're gonna push sort of this cascading redeployment model where, yes, you must buy the latest and greatest for that time to market value of really, really quick training. And then you use the stuff that is 18 months old, 36 months old to do all of your other stuff.

That's gotta be the only strategy they can employ. 'cause they can't tell people to refresh their tech every, every nine months. Mm-Hmm.

That's not why. Yeah. And to your point, I think Nvidia knew this when they tried the buy arm, right?

They were thinking about this whole process and the relationship between these things, and there wouldn't be these light away processors. Um, so do you think as we kinda look at the rest of the processor lineup up there, I mean, Intel is getting beat up, but AMD's out there, and are all these companies gonna make a comeback because people are gonna figure out this inference engine issue and they're gonna go, I don't need GPUs everywhere. I, I, I think there's, there's a huge opportunity.

And that opportunity is the margin. The margins are the margins of the profit that Nvidia is sucking out of every room they walk into. So there is an opportunity for incumbents like a MD, uh, the Broadcoms, Intels of the world, if they execute well to go after that margin in the AI space.

Um, a MD seems better positioned today than Intel does, but Intel's got gaudy three coming out, and I know that they're doing a bunch of work to demonstrate that Gaudy three is, uh, a, a, a great substitute. If you need GPUs, uh, they're gonna make the case that, hey, you can save money doing it. The Gabi three open way, uh, as opposed to the walled garden approach from Nvidia.

So, so yeah, the gold rush is on, uh, uh, the, the pick and shovel and pan vendors. I'm in the hills of Northern California where gold was discovered. Um, all of those, all of those folks are getting ready to, to sell their wares.

It's just the beginning. And, you know, the other thing I'm starting to see, and it's just like little startup companies or spinoffs of other larger companies, is people are putting together, um, like memory platforms to optimize GPU access. And it seems like there's gonna be a whole school of, um, accelerators and things that are used to, uh, maximize the utilization rates of gpu.

So is that another way to get after this? Uh, Mitch, I'll throw that at you. Yeah, Absolutely.

Yeah, I mean, we talk a lot about accelerators, um, uh, which are kind of companion or sidecars to, to CPUs and GPUs as ways of helping, knowing when to apply. Um, you know, AI chip sets to certain problems there. There's a whole nother part of this conversation.

We're not having though, Mike, and, and, and we talk about training. Um, I know around, uh, machine learning and, and gen AI are different in terms of how you train the models. There's a lot of work right now that goes into machine learning applications and interesting, I've been doing some work with a couple companies and it, it has its own lifecycle, kinda like software development has it where you're coupling, uh, a data engineer, an AI engineer, and a domain expert, and you create, they call 'em feature features or feature sets, which is basically when you've conditioned the, the data you've built, the algorithms, you've refined it enough that you're getting the kind of results that you want out of your machine learning algorithms.

And you go through that cycle many times to get enough together to say, this is what we're gonna release into production. Um, but it's very inefficient. Um, so there's some technologies coming on the market to kind of help the development process of that to what David was talking about accelerating that I, I suspect we'll see the same thing coming out for Gen AI to make that much more efficient for tokenizing and putting the probabilistic information to LLMs and medium and small, medium large.

I think it's more French fries than t-shirts, right? Would you like to large fry large size that order? Mike, My question is, as all these companies are scrambling to get this latest and greatest technology and the best chips in everything, do we have the resources to provide them?

Are there enough companies to provide this technology to all these companies? 'cause we've seen the bottlenecks in the supply chain. So how is, how are these, um, technological orders gonna be fulfilled?

David, wanna take a crack at that one? Yeah. So, um, that's, uh, the big question.

When, when people are looking at companies by their stock symbols, which I try to stay away from 'cause I'm known Nostradamus, um, uh, that's one thing that, one thing that's asked, what about supply chain issues? Um, there are companies that have their orders filled for the next 24 months where they're at capacity, it's all they can ship. And, uh, and so that's one end of the spectrum when it comes to headwinds and friction.

Um, the other though, and now this is, this is me being, you know, the world outside of my ears is very different than the world between my ears. I have to remind myself that every day. So inside my head, I, I gotta tell you, this starts feeling like the savings and loan crisis in a, in a, in, in, in some ways, I don't wanna say AI hype is just hype or that, or that or that hype is involved because I'd be excommunicated from the tech community at this point.

But look at it this way. Back during the savings and loan crisis, what did we have? We had everyone with an incentive to do one thing.

People writing loans, got paid to write loans, people doing appraisals got paid to make sure the appraisal was high enough so the loan got written. Uh, the people bundling the loans wanted to make sure the loans were written. And everybody in government was happy to tout the latest statistics on home ownership.

There was not a single adult in the room saying, wait a minute, are we being irresponsible here? By any stretch of the imagination right now, here's the issue we face, especially in the US, 40% monetary inflation over the course of about five years. The savior for our economy globally and in North America is ai, it's technology.

That's what, that's what's being sold. It's the idea that the only thing that can deliver us from this is this pushing out of the production possibilities curve to use the, the, the econ way of looking at it. And that is by increasing the efficiency of technology.

So the promise of ai, we're putting all of our eggs in we the AI basket, in my opinion. And so on the one hand, it could be we can't keep up with demand. On the other hand, it could be three years from now, there could be data centers that are built full of GPUs that aren't being used.

I know that sounds crazy. Like well, that's impossible. We're headed, we're headed into a future where, where, uh, well Dave, didn't you just say energy was gonna be the problem?

Now you're saying, now you're saying, uh, we're gonna have to decommission windmills 'cause we don't need the power anymore. Um, that's, that's another one of those parallel universe possibilities in my mind. Alright, I think we're gonna close this here.

However, in defense of loan officers, of which one of them is happens to be one of my best friends for life, um, they were screaming their heads off about this issue. They just got overruled by the So-called adults in the room, so, Right, so yeah. Yeah.

The So-called, yes, All right, one thing or another. What we can be certain of is hardware issues are cool. Again, whether we have too much hardware remains to be seen.

We'll be back in a minute. All right folks, and we're back. And we're talking about another issue that's closely related to AI and it's data.

Gravity turns out that, well, all these models need a lot of data and everybody's kind of trying to figure out how to manage that data. And truth be told, we were never all that good at it in the first place. But let's start with David here.

We have a study here from Nutanix where they're talking about, well, everybody's tried to move data to something from one platform to another. But when I was much younger, people would tell me, nothing good ever happens when you move data. Just, it's, it's costly, it's a security problem.

Um, nothing ever fits in the format and you wind up buying more hardware. So, um, do we need to go back to the old days where we just bring compute to the data? Uh, yes.

And get off my lawn. Uh, uh, yeah, I think, I think that, uh, look, the data, data gravity is something that's real because of the, just the sort of physics involved with moving data around. Um, also there is the emotional gravity associated with, you know what, I kind of like the idea of the core of my business being under lock and key in a physical facility that I have control over as opposed to being in the ethereal cloud.

So, um, it, it, it is an issue. I think it's going to, uh, in, in the AI space, it's gonna manifest itself the way that we've sorted three of us, the four of us have alluded to here, which is this idea of some things happening on premises or, uh, in now what we call edge data centers, uh, because they're not the mega cloud data centers. So I think that data gravity is going to make a lot of those decisions easier to say, Hey, look, I don't want my data going over a national border.

I don't want my data. Um, I want my data being air gapped from, uh, from general use. And so I think a lot of these smaller, lighter weight models, um, are gonna be deployed on premises, and they're gonna be, they're gonna reason over data that is in people's data centers.

So this whole idea of when are we a hundred percent cloud remains the same. It's like the vanishing point on a highway. You look up ahead and it's like, look at that in the distance.

The two sides of the roads converge. The closer you get there, the further the further you are away from that point. So that's, that's how I see data gravity manifesting itself.

Mitch data engineers, are they the cool kids in the, in the neighborhood now? Because we need these folks to figure out what to do with all this data. And, and I don't think, you know, the storage administrator is necessarily up to the task.

Oh, it's a lot more than just storing it, right? Not that that is an important and performance for applications, but how you organize data and different applications as well as technologies, AI versus conventional applications, how they use that data makes a big difference. So we end up with all kinds of things, caching and vector database and things like that to accelerate them.

But they, you know, if you were gonna go into a computer science degree today, I would say go into data or data AI and engineering. 'cause that problem is not going away and it's only gonna increase. Um, as far as location we talk, you know, I used to talk about data gravity, now I talk about data, the stickiness of data.

It's really sticky and hard to pick it up and move it. It's even worse when you step in it. 'cause now you're, you get all sticky trying to mess around with that data if you don't know what you're doing.

Um, and it's easy to, um, for, for someone who's maybe less experienced or doesn't have data engineering experience, it's easy to construct models, uh, spreadsheets, algorithms, whatever it might be. Um, and not actually be getting the correct answers. 'cause you don't understand the, uh, the contextualization of the model.

And one of the things that we have that's helping with that is graph databases, which connects data from a lot of different sources and gives the context of here's Mitch and here's, here's here who he is and here's the applications he's using. But what is he doing in those AppSec that, how does that data relate to each other, um, when you go across data sources. So I think that's an, an interesting area.

We see it, especially with AI as you try to do things like automate tasks. Well, I need more than just knowing this is the task. How does it fit into some workflow or Process?

Correct me if I'm wrong, but I think one concern business leaders would have in this area is if they're spreading the data out among all these different sources. Some of it, um, you know, in the cloud on premises, it, it's everywhere that security is a, a big concern. Huge.

Yeah. Spot on. Amanda Is for no other reason than psychologically it's a concern.

Uh, I think that that, you know, hyperscale cloud vendors would argue that their security, it's, it's like, I feel more comfortable with my money under my mattress, but is it really safer there than in a bank? And, you know, the, the, the cloud providers would, would argue that actually despite hearing horrible stories every once in a while, statistically you're much safer that way. But that doesn't matter.

We're still, we still have humans in the loop and a lot of people are more comfortable, um, just saying, no, no, no, we're not. We're we're, we're not gonna play that game. Well, the FDIC will not insure your mattress.

So that's the first issue, But only up to 250,000. So be careful. You Put in your mattress.

Anyway. David, you mentioned that you were living near where the original Gold Rush was in California. Yeah.

AI has a certain feel of it to be the great gold rush of our times. And yet if we look back in time, the people who made the most money in the Gold Rush was selling DRIs and shovel. And so is storage the, and data gonna be the shovels and dungarees of ai?

Uh, is, are you asking storage? Yep. Storage systems and data management.

Is that where the money is gonna really be? Yeah, I, I'm almost incapable of being objective on that subject. I spent 20 years deep in the storage industry.

Uh, you know, my kids are, my kids are enjoying, uh, fine educations thanks to these data storage industry. So, um, so I think I, I I, I, I think it's, um, I think that the feel to, to Mitch's point, it isn't so much about the storage systems themselves. Why?

Because we've gotten to a point where it's sort of expected that these things aren't going to lose your data. So we take that for granted, and the real value add is in the data science layer. So I would say that the Gold Rush really is in the, um, in loosely termed the, the kind of the data science layer, the idea of prompt engineering and people who can connect the dots between the bits and the bikes and the business value of data.

That's where I think the gold rush is. That's what I tell my kids to, to go after, be that person who can translate between the deep tech and the business value. And you're gonna be fine forever.

Weird days. I'm like having a flashback. 10 years ago when I went to this conference, and I remember the developers would be hanging out in like some cafe, lucky if they got a beer and a pretzel, and all the storage guys were staying at their Ritz Carlton on the other side of the tap.

Right, right. Same kind of thing. Yeah.

It's not glamorous, high tech plumbing is what we used to say it is. But, but, uh, plumbing's important. There you go.

Amanda. I would swear on tech strong ai, every fourth article now is about data management issues, but maybe I'm just biased in that direction. But, um, do you think that this has become a focal point conversation?

Oh, of course. As you mentioned, uh, every other article seems to be centered around data and managing the data. It's a, and when I have interviews with different business leaders, half the time it comes back to data, Right?

Mitch, what is your best advice? You've been doing this for a while, um, and when confronted with a large amount of data and somebody says, we're gonna move this somewhere. Should I just say no in a kind of a, a Nancy Reagan kind of way?

Or is there some other, is there some other way to like, think about this in an intelligent fashion? I, I'm gonna give you an answer you aren't expecting. I, I kinda learned that by ob observing earlier in my career.

You mentioned, I've been around for a while. Um, you, you, you always wanna be the third project manager. You don't wanna be the first or the second that's tried to move the data.

'cause expectations were inflated. First one fails wildly, second one comes in, tries to pick up the pieces, just not enough to get it going. Third project manager comes in and save the day.

So don't be the first one to try to lift all your data to the cloud or, or back home or everywhere. So what I'm really trying to say is there, there's a lot of learning and experience you gain from trying to do this. And you can learn from others.

Don't go it alone. This is not an easy topic to take on. Be the third project manager.

David, you've done storage, um, and data. I've seen more than a few CIOs lose their jobs because of some massive data management initiative that got launched and just outright failed. So is this kind of a career threatening to see it?

Well, CIO stands for career is over. So I think that's sort of a given, uh, a given in some circles. Yeah, it's scary.

And, and, and, uh, you talk about, you know, being proud, you know, it's like we're number three. We're number three, you know, be the, be the third project manager. Um, CIOs are hesitant to take on projects that might go beyond, you know, where the, where the positive ROI is, beyond the average tenure of A CIO.

Uh, there's a psychological hurdle, hurdle to get over when something, if something realistically is going to take 24 months, and there's no way the business is going to tolerate that kind of time horizon, what do you do? Uh, so you see a lot of, a lot of companies with twine and twigs and gum and, you know, putting patches on things instead of having a real, um, strategic plan because that strategic plan is gonna take too long. So, Or hire the consultant so you can blame them when it goes wrong.

Right, Exactly. No, that plausible deniability, I mean, it's amazing. Hundreds of millions of dollars worth of plausible deniability to be able to point to a Bain and McKinsey or an Accenture and say, but they said, uh, uh, that, that seems to be worth its weight in, in gold.

To go back to the gold rush analogy, It's real, well, well, maybe you should, you know, at the end of your career, launch that three year project with two years to go to retirement. So, you know, you kind of like just bail out before it comes to fruition or not comes to fruition. Yeah.

Yeah. Good advice. Good.

You have a future in career coaching. There you go. Yeah, I would just point out one thing, right, as we joke.

But David, the volume of data continues to increase, but it's more diverse than ever. It's getting harder to manage. There's, and it's moving faster than ever.

And some of this data is at rest, and some of it is in the cloud, and some of it's being analyzed, um, in flight using various platforms like Kafka. Is this whole thing just, uh, taken on a new era? Are we, have we entered a new phase of data management and analysis?

I think it's an extension of what we've been going through for a long time, but yeah, the problems, the pro, you know, you, you look at the rate at which networking bandwidth and throughput has increased versus the scale of data increase. And we're, we're not keeping up. It's not, it's, uh, it's like, yeah, we're, we all think we're cool because we can stream 4K videos, you know, uh, but, but, uh, try moving a peta bit petabyte of data over the internet.

Uh, good luck. Uh, that's why that there, why there are physical devices that if you, if you're doing a wholesale move into the cloud, they will send a semi-truck full of essentially hard drives to load your data up and bring it over. And then there's the, then there's a question of data synchronization.

Um, you know, active, active, active passive, uh, business continuance strategies. It becomes a very, very, very expensive, and the frustrating thing is, 80% of that data is probably garbage, but you just can't. But, but, but, but, but the cost associated with pruning and tuning sometimes exceeds the cost of just saying, you know what?

We're just gonna have a storage facility and we're gonna keep all the kids' toys there and we're not gonna get rid of anything because you never know. We might need that Barbie. Uh, and so there's a, there's there, therein lies the challenge.

And I don't think it's going to be solved, uh, in my lifetime. I think that inefficiencies are going to exist in the system and they're just gonna get worse. There's So you're saying there's caravans of data traveling across the data, across the, the world we have to watch out for Yes.

Caravans. Well, there's, there's an auditor that's gonna come looking for that piece of data in five years from now. It's gonna be that one.

You were the CIO that deleted it. Yeah, Yeah, exactly. And now you're going to jail.

I know that sounds like a joke, but it's serious. These people, you know, it can be serious consequences for not, for not protecting data, but, uh, but yeah, look, I I, I'll tell you that my colleagues from the data storage industry, we sort of look at it from an AI perspective, and you look at all of the traditional ways of storing data and, um, and many of them just don't apply. So there's, there are new models necessary for storing data that's going to be accessed in the world of machine learning and AI that are, that are different than some of the traditional ways that most data is still being sold, sold, stored.

Yeah. I think another thing to add to what you're saying, David, is we have so many more things that emit data now, and we have such a easily available storage 'cause of the cloud, right? I don't have to go have to go, I don't have to go buy a rack full of a right, a NA or something to store it.

But, you know, we used to just let logs coming out of any piece of software just kind of roll off. We kept them in case we need 'em, you know, after two weeks they rolled off. Now we keep that stuff and we put it into the Splunks of the world and use it for analysis, our IOT devices and that tons of data.

We had telemetry information to applications to understand what parts are people using, what's performing well, all that kind of usage, uh, information. And that's just, you know, three examples of, of a thousand or 10,000 things. So part of the, the, the growth or rapid growth of data is that we're, we're emitting more and we're keeping all of that data, not necessarily always using it, but that, that's some of the promise I think of ai.

ML is using more of that. But the, to your point, the world has changed in that way. We have the things that got us here and then we have all these new sources that are, have taken us to where we are.

I think it's pretty clear at this point we're gonna end this section here that if you wanna succeed in it, as always, stay close to the data. 'cause that's where the money is. We'll be back in a minute.

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 Techron Group. All right folks, and we're back for our third segment. And AWS kinda surprised a lot of folks, they, uh, announced that they would no longer be taking on new customers for a slew of services that they launched, including a cloud-based IDE, and a GI repository that took everybody.

It, it took everybody back a little bit on two counts. One is the announcement itself seemed a little abrupt. It was just kind of put out there and a lot of the customers were surprised.

And secondarily, AWS has kind of been positioning itself as the backend for DevOps for a lot of developers. And, uh, these services that they're turning off were at least considered by some of us strategic, but maybe not. But Mitch, what's your take on what's going on here with this whole thing and, and is a w is kind of shifting a gear in your opinion?

Well, I think it's, it's several factors. One, if you were kind of entering the cloud era, maybe you didn't have the tools to, to develop for the cloud. That's when a lot of these services, even though there's more recent ones, you know, kind of made sense to use.

Now we live in a multi-cloud world, right? So do you want to be stuck with your development environment and in one hyperscaler that maybe is even more attuned to that, or you already have these things you already use GitHub, you, you know, or you have whatever you're using for, for source code management, you have DevOps platforms, um, and automation tools, CI/CD platforms. So, you know, for AWS to compete in all those different parts of the development tool, DevOps tool, security tool markets, I think it's just a recognition to say, you know, know people need, um, more than what an AWS would provide.

And, and AWS would be better to partner with those companies and make that extremely easy. So if you're already using a tural CI for your C-C-I-C-D or your n Jenkins and CloudBees or whatever it great, come, come in, we have a great environment for you to work in and, uh, we'll work with your GitHub or whatever you're using for your, your code management, et cetera. So I think it's just a recognition that people don't want to get it or aren't looking to get their IDE from a cloud.

Right. David, do you think that from a IT leader perspective, this is a cautionary tale about not jumping in too early with something and wait and see if it actually is gonna be a wrap? What it means is you should always be aware that change is, is going to come around every corner, and you need to be prepared for it.

So to the extent that you can hedge to the extent that you cannot, that you can avoid vendor lock-in. Yeah. Yeah, that's great.

Um, from an AWS perspective, um, I see this as the sign of a healthy, well-managed business. Frankly, you don't hear about things like this from companies that are fat and happy and bloated and ignoring the ruthless optimization of their business to stay ahead of the disruption curve. Um, it's, it's hard for, for some of us, for all of us really to get our emotional heads around the idea that AWS could go out of business eventually because they'll be disrupted by something else it seems impossible to conceive of.

But I guarantee you there are people at A AWS, just like there have been people at Microsoft and other companies who wake up in the morning thinking, oh, oh, the end is near, what do we need to do to make sure that we don't fall prey to what happened to all of these other companies before us? So I think I see this as a healthy thing. They're identifying things, they're like, look, there's gonna be some pain associated with it, but ultimately it's healthy pruning, uh, in order to better serve customers.

So I I, I don't think there's anything negative here. I'm sure there are gonna be, there's you, you could anecdotally find people who are really upset by this because they, because their whole workflow is based on some tool set. Um, but they're not gonna cut, they're not gonna cut off at the knees.

Uh, this is like any, any end of life cycle product sort of thing. They're gonna maintain it. There will be people three years from now using every single one of these tools.

Mitch, how far down multi-cloud are we? 'cause when I talk to folks, they'll say we're multi-cloud, but they're 90% in one cloud, and then they're on premise and then they've got a smattering of other stuff. Are we moving to a model now where developers and are building applications where they're invoking resources that exist in different clouds?

And, um, if I'm running it, I need to kind of think about it in those contexts these days. 'cause all those things are just an API call away if I think about it. Well, if for, for a lot of folks, multi-cloud is kind of like, um, the relatives you gain when you get married, right?

So you do an m and a transaction with many, many companies go through, you know, you're not gonna, you're probably not gonna, let's don't buy those people 'cause they're in Azure and we're in AWS or something else. You know, you acquire and then you have to, to make the decision to, to centralize and move things is a tough one. Um, it usually doesn't happen unless you've got a really good process for just acquiring, uh, companies and moving their stuff in.

You know, that for somebody that's doing like a rollup or something like that, so multi, a lot of multi-cloud is just, yes, we do 90% of our stuff in Azure and we bought these other three companies and they're in three other clouds too. So guess what? We're multi-cloud and we have to figure out how to work with that.

So I, I think that's a, that's a big, a big part of it. The other, you know, we were talking about, um, the development tools from a WSA lot of things have changed in the cloud providers. Uh, you look at a CloudFlare and all the development capabilities that they've added to their offerings, whether it's security tools or, or workers at the edge of the network.

You look at, uh, Akamai who acquired LA Node, which is a very friendly, um, developer friendly, uh, environment, uh, to add to their CDN now, now kinda larger cloud. So developers working with your cloud through APIs and tools and things like that is much more commonplace and easier to do versus kind of being centralized in one cloud is just, that's this sort of a fact of life for all of us. David, have you seen any effort to kind of go multi-cloud in a more proactive manner versus, uh, what Mitch just described, I would say is more reactionary?

'cause you know, I inherited it somehow. Yeah. Um, I think that a lot of folks have been down the path of thinking that multi-cloud was going to be substantively different than a multi-vendor strategy.

Um, but, uh, setting aside for a moment, uh, the idea of acquisitions and, you know, we were Azure people and we just, you know, I just married into an AWS family, what are we gonna do? You know, Thanksgiving uncle, uh, uncle S3 comes over and gets drunk, you know, it's gonna be embarrassing. Um, I think the idea of cloud bursting and, and you know, on Tuesdays we'll run this in AWS but on Thursdays and Fridays, we're gonna run this in GCP because we're gonna save 10%.

I think. I think a lot of that isn't even on the table anymore. I think a lot of it is, um, uh, fit for function, the perception that some providers, uh, fit certain categories better than others.

And largely it's, it's, uh, Hey, hey, um, AWS I'm thinking about doing this. How much are you gonna charge me? And, uh, when you have, when you have enough power, I'm talking about large it, uh, shops, um, because Azure's telling me that they'll do it for, uh, half the price.

And so you get that sort of, you're not locked in and you have that as a hedge. So I'd say the best multi-cloud strategy is frankly, you know, 60, 70, 80% of your stuff's running, of your stuff that's gonna be in hyperscale cloud, 60, 70, 80% of it is with one cloud. And then you have your second provider that's doing maybe 20, 25%, 30%.

And then you have a third provider who's out there in the wings waiting to get a taste of your business at some point. That's, that's how you keep things honest. So I think if, I think if most of multi-cloud is multi, is a multi-vendor management strategy, um, but uh, that's just, that's my perspective.

Back in the day, if I just wanted to threaten a vendor, I would just get a coffee cup from their competitor and have it on my desk anytime. The sales rep table, same, same thing. Same thing, right?

Just, Just have a login screen to the competitor on your monitor when they come to visit you. It's amazing. It's the, yeah, it's like, it's the, it's the million dollar GCP mug.

All you have to do is take a sip out of it and all of a sudden you save a million dollars Really good coffee, tastes really good out of this Customer. Got So back in the not, well, recently actually, I was talking to this person and I said, you know, what is ultimately the difference between your IT team and the IT teams you see at Google and Microsoft and uh, Amazon and these other places? And, and without batting an eyelash, he basically looked at me and he said, well, they have engineers and I have administrators.

And his point was that that, uh, the talent that is building those services out at those platforms, it tends to be much more, uh, sophisticated, shall we say, than the average IT administrator who know is basically running things historically through a EY on, say, VMware vSphere, right? Takes some level of skill, but it's not quite the same. His point though, and, and one of the things that he illustrated was they use software engineering to go build actual control planes for those platforms.

And that's what allows them to manage these clouds at scale. And the average enterprise still hasn't quite mastered the, a single control plane, nevermind a multi-cloud control plane. So David, let me start with you.

Do we need that? Maybe if we wanted to get to this hybrid cloud thought process, rethink the way the control planes are structured and IT organizations are gonna go have to go build one or find one somehow that works against multiple Clots. I don't know, I may be showing my age here, but whenever anyone, what I hear when in this discussion is, wouldn't a single pane of glass be nice?

That's what we would call it. Single Is a variation of that data. Single, yeah.

Yeah. Single pane of glass, universal remote control. And I think yeah, every universal remote control.

Because, because yeah. Okay, great. Change the channel.

Adjust the volume. Fantastic. But what about the deeper functionality in the devices?

You know, they all, it, it just starts to fall short and then the single pane of glass, well, who, what person in the organization needs to see everything and needs to have the level of granular control that a single pane of glass would deliver? And what is the complexity that's involved Now? Yes, if you are a, if you're a uh, GCP, you're an Azure, you're an AWS coming up with a control plane to control all of your stuff that you control, that's different than being an IT organization that's bringing in third party tools that have to somehow be integrated together.

And so, um, if I'm advising a company, I tell them to not try to find some magical universal control plane, but instead double down on your relationship, pick a horse and ride it double down on the relationship. There is no absolute right answer for the set of tools you're going to use. And the difference between an administrator and an engineer frankly becomes kind of gray when you get into the driving of the race car versus the building of the race car.

Um, so yes, those clouds have got engineers, but what are they engineering? They're not engineering something that's the same as IT infrastructure for, for an individual company. So I'm not, I'm not that impressed by the idea that, uh, you can take what AWS does internally for their own operations and model that within Ford Motor Company or you know, whatever, whatever company.

What do you think, Mitch? We hear about platform engineering and DevOps all the time. Are we moving down this path?

And maybe, you know, it may take a while, but it seems like folks are attempting it. Uh, I, I, I recall a point in time where I thought it was early in the OpenShift days, and I had one of my administrators and I thought it was a really good idea to try to set this up in our environment, to have kind of our own internal compute platform and storage platform that we could share across all of our projects. And that turned into a gigantic failure.

Um, not, not anything against OpenShift. It was just so complex. It was beyond what we could, we could do.

Um, so I'll get to your question in, in a second, but there's another name for a single pane of glass, which is single glass of pane, which is trying to try to really leverage that and do it at a correctly there, there's no one silver bullet to this. I think that the fundamental thing that part of this conversation is that what you're doing, the business you're in, and the business AWS and GCP and Azure and everybody else are in, are two drastically d different businesses. I'm running a hospital supply business, I'm running insurance company business, or I'm doing, you know, the technology for that.

I have no business trying to, to take OpenShift and creating my own internal platform for hosting applications when I, when I'm gonna go to somebody who spends all of their dollars on doing that as their business, and they will spend infinitely more money than I ever could. And I may do some things to help us, you know, large enterprises do invest in some of those things, but even even them, it doesn't make sense. So platform engineering to get to your question is, is to me, is more about taking the erector set of options and saying, here's what we're gonna use.

Let's use these five configurations, and if you use those, we'll support 'em and we'll update 'em and we'll make those updates easier. And you can concentrate on the fun stuff that you like to do in testing or software development or whatever it might be. Um, and that's one of the ways, and by, by doing that, we'll take advantage of some things that cloud providers do really well that we don't have to do for ourselves anymore.

Um, or handle some of the challenges around upgrading and interdependencies that you might have or work on performance, security, things like that. So I think platform engineering is trying to take, you know, everybody has to build the fundamental pieces, uh, out of the rec erector sector to eventually build what they're gonna create. And it's about getting those starting pieces in place for you so you can be more productive.

And David, what do we do about this fundamental problem of IT management? Because today, every time I add a platform, I add more people. The biggest cost center of it remains the people.

So, um, you know, is AI gonna solve all this problem for us and maybe we can get to the land of hybrid cloud, NIR Nevada through ai, or are we just stuck here forever? AI solves all problems, all problems. There are no, IIII, if I had a whiteboard, I would draw the formula out that shows that there are no problems in the presence of ai.

So let's be, let's be clear about that. Um, no AI doesn't solve it. Um, the COPA line that Mitch just delivered that I'm gonna remember for the rest of my life is here's what we're going to use that getting to that, that here's what we're going to use is such an overly complicated resource intensive process.

And I, and I, I firmly believe that if people spent 20% of the time figuring out here's what we're going to use and just said, here's what we're going to use, throw a dart at a dart board, make the decision and roll with it, and then make sure you have the expertise in the team to address the inevitable mismatches and things and build your system together and ride that horse and have really strong partnerships with your vendors that you're buying from. Um, that's, that's the way to go. So, no, I don't think AI is go, I mean, there are, there are AI based tools that will help people be more efficient as we augment human intelligence inevitably, but the problem is gonna go into the future.

It's gonna get, it's gonna be worse because it, because things get more complicated. But the, but the number one thing is just make a decision for the love of Pete. You know, make a decision and, and, and, and get on with it.

You know, AI is not EMC squared, right? It is not the answer for all things. Well, I worked for EMC, I worked for EMC and the Squared was for, for C Corporation.

But, uh, yes, that's true, but Different kind of Different kind of EMC. All right, folks, we're gonna see how all of this plays out, but I think the most important thing to remember is, um, if you do make a decision about who you're going to the dance with, just be c be, be aware of the fact that you're gonna dance with them all night long. 'cause there are no new partners after that.

So, gentlemen, Amanda, thank you for being on the show. Pleasure. Great ski, everybody.

You bet. All Right, lot of fun. And thank you all for watching the latest episode of the Text Drawing Gang.

We have an awesome lineup of content coming up right after this on Textron tv. Stay tuned. I'm Bonnie Schneider, sustainability contributor to the Textron Group.

I'm excited to introduce you to a groundbreaking new initiative from Techron Research, the sustainability pulse meter. The pulse meter offers valuable insights into how environmental responsibility factors into tech purchasing decisions for key players in the industry. Position your company as a leader in the industry and differentiate from your competitors with the sustainability Pulse meter offered exclusively from Techstrong research.