AI Funding Gaps, Silicon Power and GSI Service Issues

Wait, why? Okay. Hey everyone, happy Tuesday.

Welcome to Techstrong Gang. I'm Alan Shimel, and I'm not in the Techstrong studio, so I'm a little out of sorts here. I have to apologize.

I feel like all the other guests on the show now. I don't have my production staff. I'm just a nobody.

But anyway, I'm here in Denver for Splunk Conf, where Mike Bizard and I are hearing all about the latest and greatest from our friends at Cisco and Splunk. And we're going to open up today with something that we heard at Splunk Conf, but not something you would associate necessarily with Splunk. But first, let me introduce you to our fantastic panel of gang members for today.

Joining us on Tuesday, which is not her normal day either, she's usually a Monday person, is our own Kate Scarsella. Kate, welcome. It's good to see you.

Nice to see you. Thank you. Also joining us, well, Tuesday I think is his normal day, is our friend Sid Nag.

Sid, good to see you back in the saddle. I know you were traveling. Yes.

And we can't get enough of him on this show, so we're lucky when he gives us the opportunity to join us, our friend David Nicholson from Futurum. Hey Dave, how are you? Doing well.

I always appreciate the opportunity to be referred to as a nobody. Well, we're all nobodies, so- I know, exactly. But- No, it's good.

I mean that genuinely. Absolutely. Every once in a while.

So guys, I wanted to kick off today, though, as I spoke about. They had an interesting keynote last night. They opened the keynote session, though today was really kind of the first day of the show.

And it featured Jeetu Patel of Cisco, who runs most of the different business units within Cisco, or they answer up to him anyway. He interviewed the Intel CEO, right, Lip-Bu Tan. And here's the good news.

Lip-Bu Tan says, "Don't worry. Don't listen to all this AI doomsday. Help is on the way.

It's our good friend silicon," which has solved so many problems in the world, is going to solve this for us, too. We are going to be able to build better GPUs, or next, or replace GPUs. We're going to have better energy usage, more efficiency, and best of all, security and guardrails built right into the silicon, so that these AI, this software, AI software will not spell doom for us, and we'll live in nirvana happily ever after.

Sid, I don't know if it was that rosy, but he did paint a picture. " But is there something to it? Yeah, I think this is Lip-Bu Tan's awakening moment for Intel silicon, right?

He's directionally right. The semiconductor layer is becoming more important to both AI economics and AI safety. But I don't think we should take better engineering, I shouldn't say confuse better engineering with complete concrete governance, which I think is really the requirement here, right?

So he's arguing that future semiconductor business that Intel's going to put in place could run some AI workloads, and that, I think there's some numbers I was looking at, roughly 1/10 to 1/15 the power of the GPU. That's pretty dramatic. This is the G1's paradox.

So that's the catch here, though, which is when you start to reduce the price of the resource, the consumption of the resource increases dramatically, right? So if inference is 10 times cheaper and more power efficient, enterprises probably won't consume one-tenth of electricity, right? They'll deploy dramatically more inference, but silicon efficiency, therefore, it changes the economics of AI, but it doesn't necessarily solve the electric grid problem.

We still are going to have to deal with that problem. And I think the other piece is hardware should become, this is our favorite topic on this show, I know Kate and Chris would appreciate this, is hardware should become part of the control plane. And that's where I think Mr.

Tan misses the boat on the conversation. So I think the future of AI security architecture should really be about silicon, to infrastructure, to model, to agent, to governance. That's the way I see the flow.

So those are my initial thoughts on this topic. Gang members, anything to add? Yeah.

Sure. Go ahead, Kate. So one of the things that I read in the article, and what we've also talked about on this show, is the idea that we don't know what really AI, we hear a lot about the doom, but we don't understand yet how AI can actually help in processing everything that it's processing.

We're building for tomorrow for something that I don't even think we'll need to actually build into. So with that being said, I think we have cybersecurity to look from the past. In the past, Intel has been a part of this.

Intel has had security on their chips. So I don't think it's a stretch for them to say it. I think it's sort of, for me, I think, well, we did put 10% in here, to help them.

And so I think that they're talking a lot. We gave them money, 10% in our funds. " We basically are helping them with this story.

I do think, at the end of the day, that there is a lot to put in hardware. It's good that we're looking at this because it's always been a part. Cybersecurity is not just one piece.

It's not just one part of cybersecurity. " It's all of our jobs. In every piece that AI touches on security, it becomes an important part of the story.

And so I think that Intel recognizing this, and putting thought and resources behind it, it's an important part of the story. Yeah. Do I think it's the answer?

I don't. Yeah. I think, however, just to give a little bit of a contrarian view there, I don't think silicon can determine whether an agent's intent or behavior is safe.

Hardware can enforce boundaries, silicon can enforce guardrails, but it cannot replace things like identity, policy, observability, governance, et cetera. So I think that's the piece that Intel is not talking about. And it's not talking about in a manner where it can be infused into the silicon.

You're hearing stories about small models. I was reading an article the other day, there's a company called Talys, I think, that's essentially burning a model into a silicon, in a chip. And think about that.

You can literally, now you're going to get models running on chips, like DCI cards you can throw away. Okay, that's fine. That's a software, firmware, whatever you want to call it.

But when it comes to the policy, governance, observability, those kinds of things, I'm not sure we're ready yet to burn that into a piece of silicon, and I think that's really my worry. Yeah, I can speak to the power consumption side of things. The technology already exists to dramatically decrease the amount of power required by inference.

And a couple of the companies that I've met with that are still stealth, these are all former Intel engineers. And the functional equivalent is screwing an LED light bulb into a socket that was originally designed for an incandescent light bulb. If you've ever done that, and you've ever been aware of the dramatic decrease in power consumption, you might go from 50 or 60 or 70 watts down to five In terms of power consumption.

The reason why that's important is because these devices open up this sort of reuse of existing data centers. Existing data centers can't handle, from a variety of vectors, the requirements of modern AI. When you decrease the amount of power required by 10 or 15 or 20X, and these devices already do, big changes.

So I think we're going to be saved from the energy consumption side of the AI apocalypse by this stuff. It's just going to take a while to work through the system because the barrier to entry is so high and the incentives aren't yet there in the marketplace because the true cost of AI inference is not yet being passed on to any of us very directly. But it is coming.

It's not pie in the sky to say absolutely silicon advances are going to dramatically reduce power consumption. So, I had a chance to listen in to the conversation on stage there last night around Mike's article, and I should mention I have an article and a full-on Techstrong special report on this issue up on Techstrong Sunday today as well. Went up this morning.

One of the nuances here that didn't get picked up was, and it was a shot at Nvidia, which was you don't need GPUs to necessarily do training as well. Right? The fact is, Anthropic, I think, recently announced that they used, was it 2 million Google...

Was it AWS Trainium or the Google chips? Two million to help train Claude. So the idea here is that the GPUs, no one's arguing that the inference chips, Broadcom's inference chips, are much more energy efficient, let's say, than an Nvidia GPU rack.

" But we could use chips. Yeah. Don't focus on the G.

Essentially what is required for training are a unique set of skills, and specifically- Got it ... memory that offers continuity, a certain amount of bandwidth, the ability to leverage parallelism. So call that an XPU, call that an AI accelerator, doesn't matter if it's a GPU.

GPUs weren't designed for AI. In fact, the Google folks will remind you of that every day. Mm-hmm.

So essentially, you can't do training efficiently on traditional x86 architecture that you're going to run in a general purpose compute server. So, make sure to clear that. I think the GPU architecture, as I'm sure you know and you pointed that out, was really not meant to run AI, right?

If somebody accidentally discovered that the GPU has the ability to multiply large matrices, so why not use it for AI, which is necessary to do all the functionality that's the fundamentals behind AI, which is a transformer model. So we get that. But the reality when you think about the price performance, a GPU is not a cheap piece of silicon.

It's an expensive piece of silicon, right? An H100 or A100 costs, I don't know, price of a Honda Civic. So that price of the GPU becomes more expensive than the edge location that it's residing on to do inferencing.

That is a real issue here, right? So I think from that perspective, Intel is looking at the problem correctly, that if you can reduce the price of that silicon that is doing inferencing, that is commensurate with where the silicon ends up residing geographically, that's a winner. Right?

So that's good. But my problem with this strategy is that he's extending the value proposition of that solution beyond its functionality of inferencing to things like observability, governance, policy, and that's a dangerous path to go down. Right?

But as Kate points... I'm sorry, Kate, you were going to point it out yourself, it sounds like. No, go ahead.

You go, and then maybe I can- No, go ahead. If you point out putting building security onto the chip is not new to Intel. And that's exactly what I was going to point out because I thought to myself that is one of the best things, because I worked at McAfee for a little bit, and they had a really amazing, Intel had this virtualization capability that offered remarkable security.

I really think that they're onto something personally. I really like that they're going down this path because they had something called Deep Safe, which was they created this isolated execution layer. There's something here, and I'm glad that they're exploiting it.

They're the right ones to look at this because they've done it before, and so I'm actually proud of them. So. They got it.

Because I, yeah. Whether you are a fan of the government owning 10% of Intel or not- No ... I think a lot of us are pulling for Intel.

Yeah. Just because we need balance in the market, and- Yes ... and maybe a little nostalgia for some of us, right?

We grew up in an Intel universe for the most part. Yeah. Let them die.

Let them die, Kate says, and eat cake. Yeah. They really had quite a chip, and malware had a harder time concealing itself because of the Intel chip.

I'm pulling for them, too. So maybe they'll- It looks like a fan of yours on here is giving you a big hello. Yeah.

Don't know if you saw that pop, but it's the nice part about having a live audience. But I agree. But we will have to see.

Yeah. We're about out of time for this segment, so let us hop on that to our next block, which is, I call it robbing Peter to pay Paul. All of a sudden...

Well, we've been hearing this story for months. I forgot what company it was. They used up their year's worth of AI tokens in March or something, and what are you going to tell them, not to use AI for the rest of the year?

So we've been literally robbing Peter to pay Paul to make up for that funding or budgeting shortfall, let's call it. And CIOs are struggling with this, right? We only got a few months left in the year, but how do you stretch it?

Dave, I know you talk to a lot of CIOs, CXOs about this subject. And by the way, this is another article. It's on Techstrong AI, and it's a survey, I believe by, it was a Big Panda, but no, it wasn't Big Panda.

That's the next one. But this was a Futurum Group survey of all things, right? Over 1,600 tech decision-makers.

David, does that hunt with what you're hearing? Yeah, absolutely. And just to be clear, a lot of people say they talk to lots of C fill-in-the-blanks.

I actually do. Yeah. This morning, in fact, I was on with 50 of them, and because I teach in executive education programs where these leaders gather, and no, it maps directly.

And this goes back to what we talk about is the CTO's dilemma or the CTO mindset. It's this idea that you've got to manage keeping the lights on and innovating. You've got infinite demands, but finite budget, finite amount of time.

Within that keeping the lights on part of it and innovating and deploying with AI, there's been this shift in token allocation in the direction of remediation against security holes that are identified by AI. So if you look at, to kind of double-click on this, if you look at where people thought they were going to be spending tokens, in other words, developing cool new things, it turns out a lot more of those tokens are being consumed just keeping the windows and the doors locked in the organization. So it's not only that more resources are being consumed than expected, but the allocation of those resources away from innovation is another extremely frustrating pivot point.

And the final thing to add here is something I wrote about recently. It's this question of How do you accurately and adequately allocate the proper model to the proper task? " But who among us has not been guilty of taking an empty Airbus A380 down to the corner Starbucks by ourselves in terms of resource allocation?

The last thing you did with a generative model, did you need to use that model, or could you have gotten away with one at one-tenth the cost? There's not a really good way to track that yet. So a lot of the frustration that these folks are expressing is they know they need to contain costs.

They see where the leakage is, but there's a huge delta between where the money is going, and they're not clear how much is effectively being utilized and how much is waste. I'll tell you why they're not clear. They're not clear because the budget overrun isn't necessarily the model.

It's everything surrounding the model, right? I think enterprise AI costs are routinely underestimated because companies focus on models, APIs, the cost of GPUs, while totally understating the costs of data preparation, integration, inference, security, observability, governance, all of that. So I think that's where the delta is, and that's what they are failing to see.

And I also have an opinion on why they're failing to see it, because the AI funding has to move beyond the CIO budget. If AI is supposed to transform the entire business, whether it's sales, finance, customer retention, supply chain, engineering, all of that, why is the AI budget sitting in the hands of the IT department? Why is the CIO controlling it?

We ought to think about a very different model of budget allocation, budget oversight, beyond the role of the CIO. And I think that's where the rubber hits the road. None of the CIOs that I work with control the budget, but it is in their hands to implement, and it is very much cross-functional.

This is a CEO-led down operation. Oh, I can see that, yeah. And by the way, I totally agree with what you're saying in terms of all of the things around the cost per million tokens, but the cost per million tokens is the largest line item.

And there's a 600X- And that's what I think people have focused. Yeah. There's a 600X gap between the cost for the frontierist of frontier model tokens and the cheapest of downloadable on-prem tokens.

Now, it doesn't mean that they can all do the same thing. Right. But there's such a huge gap that has to be read.

And frankly, OpenAI and Anthropic are trying to protect that gap right now with demand generation. I think that's going to end very soon. That's going to end very soon when the duopoly ends, right?

Yes. And I think people move from closed weight models to open weight models, and from proprietary software to open source models, that's going to be a whole different conversation. So yeah.

I do think, David, what you said there, and Sid, for most companies, when they talk about AI budgets, they're talking about their token budgets. They haven't even considered the add-on sort of pieces that go around it into this number. But let me dig into the Futurum numbers a little bit here.

47%, just under half, are over budget. Only 32%, about a third, are on plan. 6%, a minuscule 6%.

Well, they qualify for federal funding over 5%, right? But 6% are actually below plan, and then 10% have no budget at all. It's just the Wild West still, and they're token maxing probably.

Or some percentage of them are not using AI at all. So that's where those numbers fall out. But the next set of numbers, I think, is where it gets a little more interesting.

What do you do when you're over budget? 48%, almost half, they ask for more budget. " So almost, again, a good chunk of people are saying, "Hey, it's okay.

2, less than 20%, one out of five and a half or something like that, are actually pausing their AI use or reducing their AI initiatives, which I think talks about how important, or at least how important these C-level CIOs and so forth are being told AI is. That even though they're over budget, it's still damn the torpedoes, full speed ahead. Kate, that's pretty damning if you ask me.

It is, and it makes me go back and think about why is it, we're talking about CIOs and the budget, and yes, they don't control the budget. They usually receive the budget. But why is this falling just into the CIO area?

Why is it not going across the entire board? Because it is impacting the entire organization on whether you're going to be successful or not. So that's my question.

We're doing a really poor job at, first, the marriage between the CIO and the rest of the team, and yet we're expecting so much without really giving these guidelines. And I understand that we don't know these guidelines yet, like, what does this look like? But we're putting a lot in the lap of a CIO and not giving them the actual tools of what cybersecurity...

Because I'm coming at this from a cybersecurity point of view, and this is coming from the CIO budget. Is that it? No.

Yeah, we're fetching the rock here a little bit, Kate, with this survey. The survey went out and asked CIOs, and they answered. But what's surrounding them, I can assure you, it is all hands on deck.

And you're exactly right. And what Sid mentioned about getting your data house in order, the number one thing, and by the way, the AI program, there's two programs that I'm a part of, the CTO program, the AI program. CTO program is more CIOs, CTOs.

AI program is everybody in the C-suite. In both programs, the number one concern is reorganizing the structure of the company. And then followed by lots of discussion of per million tokens and then all of the things around it.

So this is just the four of us here are sort of being led down a path based on this survey. Yeah. But I think in the real world, it's a lot more like what you and Sid are talking about.

And it's important because we have to have these discussions. And I think in the past, we have actually not been married. We're just sort of like these comfortable bedfellows.

But at the end of the day, AI might drive businesses to get a lot better relationship, because we need this better visibility at the end of the day. So one final thought on this that gives me hope, and that is the folks that I work with, and hundreds of them. If you were to aggregate the essence, it would be we learned a lot of lessons from the move to cloud.

Yeah. We learned that it's not about the technology. We learned that it is about reorganizing the way work gets done in a lot of ways that will determine the success or failure.

So, we focus a lot on the technology, the cost of the technology, what the alternatives are. But just like the move to cloud, it was about, wait a minute, okay, there's this cloud service here. Yeah, but we have a network person, we have a storage person, and now we need a horizontal cloud services person.

And so most of these leaders have been through this before, thankfully. Yeah. And so I think that's- I think the cloud analogy is fantastic.

I do too. This is exactly what we are experiencing today. In terms of AI adoption, right?

If you look at the history of cloud, it's all about data center evacuation and consolidation and moving, closing down all the server farms and moving infrastructure to the cloud, right? It was an IaaS story. Then everyone said, "No, no, no, no.

This is not about IaaS. " So people, that's when the conversation moved to SaaS, right? And that's when people started to see real value of the cloud as opposed to counting how many units of compute they saved into the cost.

We go through the same exact analogy today with counting the number of tokens we're saving as a consequence of moving to the AI world, right? So I think it's going to evolve to more of a higher layer narrative where how do you utilize AI to really do business transformation? I know that sounds like a fluffy word.

No. That's where it is going to land, right? Yeah.

And I think Dave, and Sid, I'm sorry, real quick, what I love is that you are really being these little rays of sunshine into- ... this doomsday message that we had. So I would say to anybody who's listening, exactly what Dave and Sid are saying, there are so many of us who have been down this route.

It's not doomsday. If anything, if this episode about anything, look at Intel, look at the hardware chip, and look at what they're saying because we have been down this road. So yes.

As a follow-yeah. Let me ask you a question. What we're really kind of nipping around the edges on here, and I'd like a direct-- Dave, you might be the best to answer this.

Is AI considered an IT item still? It's interesting. Before you answer, let me just say something.

I was on LinkedIn today involved in a discussion around somebody's post, and in talking about AI it was a friend of mine from the DevOps world, and he couldn't look AI past what... The only thing AI he can think of that it does is help with coding. He couldn't look at anything AI, even consider anything beyond AI helping with code.

Is AI perceived at the C levels that you're talking to, Dave, as an IT thing? I'd like to think that some people come in with a conception, and then I change their perception. Fair.

But what I will share with you is sometimes it's good to look at the acronyms and actually define them. And what does IT stand for? Information technology.

What does AI stand for? Artificial intelligence. Artificial intelligence is information technology.

So there's no difference between is AI, IT, is it the tech first? It's like no, no, no. Pretty much all business leverages information technology, has for a long time, will for the foreseeable future.

AI is part of it. So- Yeah ... again, it's back to the same thing that we've been through before, which is you need to get line of business alignment with the people who are going to be leveraging technology.

The CTO mindset, which is not a job title, is this idea that someone in the organization needs to be responsible for leveraging the best technology that's available in service of the mission of the organization. And so the answer is AI is IT, IT is a tool of business. There is no separating any of it.

And to the extent that people don't recognize that, they fail. Fair. Fair.

Guys, I don't want to fail, but we've got to jump to our next segment because we're over time. So we're going to leave that one at that. And then Kate, I want to come to you on our next block here.

I don't know if this is news to anyone, but I mentioned this was a Big Panda survey. I confused the Futurum one in the segment before. But Big Panda recently did a survey of, again, US-based IT leaders at global enterprises.

And what surfaced was some real dissatisfaction with the usual suspects, right? The global service integrator crowd. Kate, you spent many years at IBM.

Many. And- Global services, yeah ... yes.

And I think all of us have, in some way or another, been on both sides of this equation. Yeah. I don't know if I've ever seen anyone bringing in cheerleader pom-poms and raving on their global service integrators.

But have we reached a new low? Is this just par for the course? What do you think, Kate?

So I'm just going to read some of these numbers. Because they're wow. 94% experience SLA breaches tied to outsourced IT ops.

Like level one- That's a good one ... Yeah. Level one- I've done this.

Yep. Yeah. Level one incidents took an average of 34 minutes merely to route.

34% were misrouted after that. I mean, come on. 24% were resolved incorrectly or required rework.

I mean, these numbers. Organizations average 19 monitoring and observability tools. 74% expect incident volume to increase during the next year.

Oh, gosh. Well, and they're so expensive. Like you want to talk about a large line item.

They're really expensive. So I'll tell you, as a person who was a part of the IBM services brand, and when we started to bring in, and IBM started the IBM security brand, and we had to bring in services and product, what a mess. What a mess.

Because services always won. And I don't know whether, I think that was a Gerstner thing, right? Where he redid IBM.

But I will tell you that when I was standing in front of a client, and it was a thing between QRadar, bringing in the product QRadar, and bringing in services, and they had HP. Let's say this customer had HP ArcSight. I'll never forget this.

That was the services people love HP ArcSight because it's so much configuration and blah, blah, blah. And QRadar, when it first came in, it could do something in just a week. And the services were like, "Are you kidding me?

We don't want QRadar to be a part of this deal. " And I knew right then, like, ah, we're not doing what's best for the customer. Yeah.

Right there. We're not doing what's best for the customer. And I still-- So these numbers, do they surprise me?

No. But oh man, does it bum me out? It does.

It does because it goes along what I've seen in the field and it's just not right. At the end of the day, we need to do what's right, and these numbers show that we're not doing what's right. Fire them all, people.

Just get rid of them. Fire them all. Dave, Sid, both of you have spent your careers in enterprises, right?

Dealing with these big... It used to be if you outsourced to a, and I don't want to name names, but an IBM or an Accenture or one of the big guys, it may be expensive. They're not cheap.

Deloitte, not cheap. But there was a quality that came- Yeah ... with dealing with these companies, right?

No one gets fired for buying IBM. " Dave, Sid I mean, have things gotten, has it always been this way, or have things just gotten worse? I have a bias.

I mainly come from the product side of things. Okay. So I look at Kate and I think of Kate taking food off of my children's table back when we were trying to sell EMC storage arrays into a client that's an IBM global services, that just doesn't want to manage the EMC storage.

But I did have a stint in the actual world of services, which actually had me come away even more cynical. I would say that we surveyed the wrong people here. What we should've done is surveyed the people who enjoyed trips to Monaco to see an F1 race.

Right. Or trips to the Masters. You can rent really beautiful homes in Augusta and host senior executive clients that you're seeking to sell your services to, and they will all report that they love Tata, Bain, McKinsey, Lloyd's, Schur, whatever, throw them all in there.

So I think it's worse now because, to be fair, these entities are stretched thin because they've had to get way out ahead of the tips of their skis to claim that they can do AI when no one's done it before. " It's like, really? It didn't exist until three weeks ago.

So how is this possible? So it is the blind leading the blind, and something's got to give. And I imagine that what they're doing is redeploying the best people that do the boring stuff, IT operations.

Nobody cares about that. Let's do the new AI stuff. So you redeploy the good people onto the thing that's the hot new ticket.

So I can't blame them for their troubles, but I am cynical about the way that these entities get positioned as plausible deniability. It's a way to politically not make choices and not take responsibility for choices. When you can say, "Hey, I hired this blue-chip firm.

They hire all Ivy League grad school people. " I just don't like that whole game. Yeah.

I think the problem is much deeper than that, right? I think the traditional GSIs are completely outdated with the approach. I don't care what their names are, whether they're Accenture or McKinsey or Blue Chip Incorporated, doesn't matter to me.

The reality is that deploying AI is difficult, and it has been already demonstrated why it's difficult. Why? Because OpenAI went and acquired a company called Tomorrow, T-O-M-O-R-R-O.

To do what? To deploy their technology, because they are noticing that deploying AI technology is damn hard, and they can't do it themselves. They acquired Tomorrow.

Anthropic did a deal with Blackstone to formulate a similar company called Ode, O-D-E. Okay? Why are these guys doing it?

They're doing it because this stuff is hard, right? So, the lesson from this is the existing GSIs, the way they're operating, their model is starting to show age. They don't know what the heck they're doing.

I'll be very blunt, right? And I think Kate's already pointed Dave's point, level one IT operations may be one of the clearest enterprise use cases for agentic AI, right? So hey, for decades, the formula was more incidents, more tickets, more people, larger outsourcing contracts.

36 annual million spend, right? But it really needs to move to more incidents, more automation, fewer human interventions, and lower operating margins, right? That's where it needs to move to, and the GSIs of yesterday just haven't figured out the formula.

And I want to say, this article talks about AI-assisted triage, and it's funny because, gosh, 2011, 2012, we were talking about Watson, and everything was Watson. And so Watson became part of the QRadar. And I'm telling you guys that when we first put it on site to customers and they looked at it, and I remember them saying, "So what?

What are you guys doing? What are you guys giving us? Our junior analyst can do a much better job than what Watson is pointing out that there's malware.

Give me a break. " And we're just not there yet, so. I have to constantly poll the-- "Well, my CIOs and CTOs are telling me," but they are, what they're not getting from the traditional GSIs that they work with and love, to varying degrees, what they're not getting is the support they need doing that reorganization thing we were talking about.

Yeah. And so they're turning to boutiques where it's like, hey, sending a team of three people armed with laptops, with access to AI, and have them help us figure out what we actually do here and how we can make it better with AI. They're not seeing that expertise from the places, the traditional Oracles of Delphi, that people consider to be at some of these bigger places.

So to me, it looks like opportunity. The four of us should just hang up the call right now and create a boutique to help people reorganize their businesses, and we'll clean up. I love it.

But to Sid's point, this is an inflection point. It's a disruptive moment for that entire industry. And shame on you, Kate.

EMC storage arrays are awesome. And shame on you for trying to keep them out of the SFA. I feel horrible.

I really do. Don't lose sleep over it. But yeah, I'll just add one thing, and then we got to wrap up.

The performance of these service integrators is also leading to a lot of tool sprawl, right? Especially ITSM type tools here. Where when you add up, they're all, or most large organizations, enterprises are using three, four different ITSM platforms, which probably contributes to the whole Tower of Babel thing.

But speaking of Tower of Babel, we're going to have to end this one today. We're about out of time. I do have two quick announcements, though, I want to make.

Before I do, though, Kate, Dave, Sid, thank you. Thank you for joining us today. What a great conversation.

Thank you for watching out there. I do have two announcements, as I mentioned. The first one is this right here.

I got this in the mail. Ah. Ooh.

This book. October 6th. Well, it's on pre-sale, but don't buy it on pre-sale.

The price on Kindle's coming down. So October 6th, I'm trying to do this right, but on October 6th, you can buy this book. It has a forward by our friend Daniel Newman.

It's all about what's happening in the AI build-out today, and then comparing that to previous grid, and I've spoken about it on the show before. But comparing it to previous grid builds out like the railroads and electricity and phones and, of course, fiber optics, which I think most of us lived through and worked through. So that'll be available on Amazon or wherever you like to buy books on October 6th.

And hey, if you buy it, let me know. I'm happy to sign them for you. Or if we- I was going to say, if you want the money, you can send me an autographed copy.

Exactly. Yes, I promise you one, Sid. But so that, the quick cheap plug for Alan's book.

Next, September 24th is, I think it's our ninth or tenth annual event, DevOps Experience. And the DevOps Experience we're putting on this year is very different than the one we did nine or ten years ago. DevOps is very different.

And we have, again, another outstanding lineup of speakers, including the man who actually gave DevOps its name, my friend Patrick Dubois, is going to be doing an Ask Me Anything live session. So anything-- And Patrick, if you don't know, is now very involved in the AI Native Dev movement. So Patrick will be there.

We have Futurum analyst Mitch Ashley presenting. Daniel Newman and I are doing a little bit of a fireside chat. I'll be presenting on a new report I just finished on DevOps.

We've got a terrific lineup, including, fingers crossed, someone from Anthropic talking about their playbook for using Claude in the software development life cycle. com. Again, it's September 24th.

About nine days away, 10 days away from today. That's all I have on public service announcements. tv, Textron TV YouTube channel, or our OTT channel on just about any screen you like to watch videos on.

But until tomorrow, this is Alan Schmel for Textron and Futurum. Have a great day, everyone. Thanks again.

We're out. The agentic AI race is on. Which approach will move your team forward?

Join DevOps Experience 2026 for practical insights into AI and DevOps. September 24th, free and virtual. com.