Cloud Repatriation, AI Challenges and the Future of Enterprise Tech – Infrastructure Matters EP77

This is Textron tv. All right, double sevens. This is the lucky episode of Infrastructure Matters.

I'm going joined as usual with my co-host, Diane and Kimberly. A lot of interesting debate, I think this week. Not a whole lot of news, but a interesting debates.

And we're gonna start out, I think, with the more spicy old one, the SAP Diana First. So first, though, first, so we have an international traveler going on here. I mean, we, we are jealous about Diane's life here.

Maybe we'll all do this. Yeah, We, we'll, I highly recommend it. Go see the world while it's still there.

'cause everything's getting turned into a, unfortunately, a a, a Walmart and a Circle K. So, yeah. Well, uh, uh, for the last, uh, about 12 years, I've been, uh, uh, alternating between the United States and overseas.

And I've been, uh, uh, living, uh, in Indonesia, uh, France, uh, Malta, uh, all sorts of places. So, yeah, I'm, and, And Bali too. Yeah.

Old. Yeah. You, You need to change your ultra handle to like the nomad analysts or something.

I don't know. Is the, the nomad. Well, but Aren't we already a, a, a nomads?

'cause we're constantly at these events, right? So I just, I just, now I vary the backend as well as the front end. There you go.

So, speaking of international SAP, which I think, I don't know, Diane, I don't know if most people realize SAP is not an American company. Like it is so much. They've done a really good job of having a US not, not a US company.

Yes, exactly. They're, And, uh, they've seemed to, uh, benefit from all of the geopolitical turmoil and, and the economic uncertainty that's hit the world in the past few months. At least that's what the numbers suggest, Correct?

Yeah. So, uh, the data says that, um, SAP sales are up 40%, uh, in Europe. Uh, and this actually is, uh, not just a trend in the last two months, I've seen for a, a couple of years, a real desire to, um, to buy local, uh, um, for, uh, for pricing if, uh, no other reason.

Uh, but also for performance, uh, to have, uh, even though a lot of the cloud providers have regions that are locally, uh, local, there's still a concern about a lot of the SAS applications not necessarily having enough, uh, local regions to have performance. And there has always been concern about, um, the ability for the US government to seize data. That's been one of the, the, the overarching concerns.

So the Patriot Act has a, a, has a backdoor in most cloud for most cloud services. They could technically, if the data is resident in the United States, sees the data. So for, for, uh, I've seen concerns about, I, we don't wanna use Slack.

We'd rather use Rocket Chat. Even though no one in the has heard of Rocket Chat, it's very popular overseas because it's an alternative to Slack that is not resident in the United States. But SAP, uh, is now seeing due to the tariffs and due to the geopolitical uncertainty.

And, and the concerns about, you know, if the US actually does make moves against Greenland, uh, or, you know, and, and the things we're saying about Canada is, is causing a lot of concerns and alarm bells, and people are trying to buy down it. People are not about risk. They wanna buy down their risk and wanna say, let's, let's go with something that we're absolutely sure we're not gonna have problems with.

Um, and SAP has, has been a stalwart IT provider for a couple decades now. And, uh, for a lot of overseas buyers, uh, they're, they're, you know, they're, they're the safe choice now. It's very interesting.

So this, the, the uptick is primarily a European uptick. It's not a migration of It is, uh, also in, in apac. Um, but it's most pronounced in Europe.

It's what the data says. It just, it just came out this week. So that's why I makes around news.

It makes, makes huge sense. Makes huge sense. So is this the, an overall expansion of a market?

Or is SAP taking, taking share, uh, share away from folks like Oracle as a result? It looks like it's taking share overseas away from, um, um, you know, Oracle provides a lot of what SAP provides in terms of, uh, you know, their ERP suite, um, you know, all their fusion applications. And so, yeah, this is a real loss for Oracle, whether we haven't seen that data yet, it'd be very interesting to watch our numbers.

Uh, these are gains for, in terms of, of net new wins for SAP For our data privacy, though Oracle, and, and I don't know the patriarch, so obviously you sound like you know a little bit more about it than I do. If Oracle has a, a site in Germany or France or whatever, that's housing their ERP system and the data resides in those countries, does the Patri Act have oversight on that? Technically, yes.

That's big. That's the big objection. Now, it's never been tested.

Uh, I mean, the, you, sorry. They've tried to test it. Um, but though unfortunately, the US companies have just given up.

I mean, they, they've not taken it to court. They've, they've given up data. Yeah, because I thought we had that and may be wrong done through the Microsoft in Ireland and emails that were, but that's foggy right now.

So, No, there's, there has been a case, yeah, that, that one was a, a good example, uh, of a company standing up for their, their customer's rights. But in the end, the Patriot Act allows, even technically, even if the data is overseas, it still applies. The company has to, has to, has to hand over the data act.

Hmm. And this is why, you know, uh, there's the debate in, not debate, but movement and the reality that at the engineering level, you need to maintain your own security key infrastructure encryption. You can't rely on the providers to protect your data, whether we're talking about from state actors, uh, regulatory issues, or even from bad actors themselves.

Uh, cloud providers in the case of AWS Azure and, uh, Google all make it really easy for them to manage your key management and your encryption. But if they're managing that, that's, uh, subject to, uh, re regulations, whether you're talking about eu, Asia or the us. So you do need, this is a great example, that you do need to protect your own data regardless of where it sits.

This is what this, uh, emphasizes why, you know, the trend right now is for holding your own keys, in which case it doesn't matter. Uh, if the data's always encrypted at rest and in motion, then you still own your data. Even if it's on a cloud provider, they, all they can do is hand over encrypted data.

Yeah. And unfortunately, encrypting data, you can't take advantage of the data reduction tech technologies and those kind of things because it doesn't reduce, and therefore your costs are more. But Yeah, I mean, it tough to be Encrypting everything anyway.

So there is no perfect solution, and you have to be aware of the trade-offs. Speaking of perfect solutions, and, uh, being aware of your trade-offs, 37 signals, if you're not familiar with 37 signal, uh, they've been pretty disrupt disruptive Kimberly in this whole market around, uh, cloud repatriation and this idea that we're, everyone is moving to the cloud, right? So not for the faint of heart, because, um, about three years ago, they repa started repatriating their core, um, technology capabilities and bringing on-prem be, and they have cited that they've been saving millions and millions and millions of dollars in, in doing this.

Um, if you don't, 37 signals are the people behind Basecamp and a couple of other technology areas. Um, and their CTO, David Hy High and Meyer is very, very vocal with this post post. But what they've been doing, you know, when I say it, it's not for the faint of heart, is because they're, they're running on open source.

And, um, so when they repatriated, um, the server piece of the business, it's all open source technology that they're doing. So they got their own people running this. Um, not every company is ready to go and do that, and the difficulties that are around that.

Well, but they, they used to, this is the funny thing. Everyone used to run their own data center. So, uh, you know, that, that also has gotten really underdeveloped.

Uh, but I drop, Dropbox has done exactly the same thing, saying, you know, uh, if we have always, if we know what our workload, uh, is, um, then, you know, why would we pay for the cost of sales marketing, um, the profit margin, um, and all, all, all the fixed overhead, uh, when we could benefit that ourselves. It's kind of like flipping the whole original cloud argument on its head. It's very interesting.

So the second piece to this has been their data and their data was on a four year contract. So they could not move it, um, out of AWS it's big S3, it's about 18 petabytes, um, that they are in the process of moving, um, I guess there is a get out of jail kind of no charge for egress going on with AWS right now. So they're trying to take advantage of that, um, and be, I think it's the end of May is when they have to be out or they start at $5,000 a day, that would be the charge for taking.

So they, the Rush is saying, can you bring all this, all this data back on prem? And if also, which is pretty cool, is they're using, um, they did a bid out with the different disc vendors or drive vendors, um, looking at combinations between hard drives and Solit eight drives in or deliver whatever they're going to do. And they ended up with Pure Systems.

So I don't know if they're running completely a pure system disaggregated kind of environment or what it is. I haven't gotten gotten the information yet from Pure, but what I do know is they're running with Pure as direct flash drives, which are 150 terabyte solid state drives. And the economics they say are better than the hard drives.

Um, which goes contrary to what we are seeing in some of the cloud providers, which are still using hard drives and saying the economics are there. So there's a bit of a battle going on, but this is kind of like, we're gonna see, you know, see if they can get all that data off within timeframe to avoid the $5,000 bill a day that's going on there. Yeah.

Yeah. So this has been a really interesting story 'cause I've followed it since the, uh, original, uh, volley using a MD servers and, uh, from Dale, and we'll get into kind of that discussion later on. But the, their CTO talked about the shock and awe of when he first started to explore moving his workloads from AWS on premises, and the sheer amount of cost around the fluff, whether you're talking about VMware, vpe, uh, support for open source, finding the engineers, because they did, Diane, they did lose that muscle of being able to run data centers.

He was surprised to see how difficult it was to, uh, find and retain talent around. Well, it's, and It's a completely different proposition. Everything's all totally virtualized and, and, you know, hypervisor and all that, that's a very different world now.

Yeah. I, I, I run into very few net new engineers who want to learn virtualization like the, the, the, the, you know, it's like finding a net new application developer that wants to learn, learn FORTRAN or cobol. I, I just a little bit, but it's a similar thing.

Why would I, why would I learn how to deploy a commodity when the value is in the higher end abstraction? But, uh, they're moving forward and they're safe. They've saved a ton of money and they're turning a familiar knot.

Kimberly though, of sweating assets. Yes. So the other item, as I was reading through, um, they, there was another blog that they put up is that they talking about how they are now into the year 10 of sweating the Dell servers they originally bought, when they started the company 15, you know, 10 years ago, whatever it was, I think, believe it was 10 years ago, for half a million years.

All that says is they're not doing ai. You know, I'll talk about, I I, I've, following this cloud rate repatriation, uh, at the CIO level for three years now is, it's interesting, But from an, but, and I agree with that. You know, classically what we do is we, we, we, you take servers and do something else with, but where that economics goes in is that when they did the economic analysis of coming out of the cloud, um, they looked at five years of sweaty new the servers.

So if you go back and say, okay, so I'm gonna do 10 years of servers, I'm gonna tear them down to doing low life kind of stuff, or low level, not low life, low level kind of activities. We Dunno what they're doing, but Yeah. Well, the, the, the, have you talked to your latest Linux man?

They are low life. I'm sorry, I'm sorry. Data protection people letting people, sorry about that.

Um, I'm moving it down to lower priority kind of high, you know, uh, availability kind of areas. But still, if you're in the cloud, you're paying, I mean, I'm sure that there's cheaper costs and everything else, if you analyze what you would do over 10 years of that with lower cost servers that are in the cloud. But that has an implication in terms of your, you know, your depreciation schedules, your cost, et cetera.

It does, uh, well, and, and you also have to look at, they must have a pretty, pretty steady state workload because they, they can't have all those. They just had a, a, let's say a North American audience. Then they would have their servers, you know, quiet, um, you know, 16 hours a day.

And that's not that, that can't be the case. 'cause then you're paying for all those facilities and all that overhead for 24 hours a day, only using eight hours a day for only in one, you know, only in the in narrow set of time zones. Um, but when I'm arguing now, when people say cloud, you can't assume public cloud anymore.

Uh, cloud is now very much a private abstraction as well. It is an equally valid destination. Uh, and certainly it looks at it this way.

Say, you know, when we look at workloads, we're now much more agnostic, is what I'm hearing. And my latest research on private cloud, uh, which is, uh, coming out, uh, should be this week, uh, tracks the, the, the, the very ra rapid growth in how willing, uh, over the last four years, CIOs are, are willing to reconsider where their workloads run. Is it public cloud or private cloud?

When we say cloud, now we don't know which one means it means the same set of, of cloud technologies, but it could be used on premise, it could be used in colo, it could be used out there, uh, in, in, in the hyperscaler world. It doesn't mean anything. And it could be edge as well.

We can, it could be, you know, running workloads very, very close to where it needs to run if it's, if it needs to be performant. So, um, I think cloud now just means a set of technologies, uh, doesn't necessarily mean public cloud anymore, is the argument that I'm making. And certainly 71% of CIOs agree with me.

They say they're willing to consider, uh, 71% willing to consider where they put their workloads, um, are reconsidering where they're gonna put their workloads this year. So it's very interesting. And, you know, I think it was, I don't know, six, seven years ago when we did survey on cloud and we asked the question about cloud, and then we'd ask about what they meant by their cloud environment.

And the majority of the time back then, that was a VMware virtualized environment was their cloud. Oh, Very interesting. I didn't know that.

Yeah. Yes, it was, it was, it was like this, this, you know, what did you, if it's cloud first, what do you mean by that? And, and you'd get down into what it was there.

There, Yeah. Okay. Yeah.

And I think, yeah, so that was checking the box to say, we're in the cloud back to the executives. And there was an awful lot of that. It actually just got a little, I hope he doesn't watch this.

So, 'cause he'll identify, he'll quickly identify that. Yeah. But I, I talked to a, uh, executive, I got a note from an, an executive who basically said, uh, circle that error Kimberly, that he has learned so much more about cloud since the last time I did a project with him.

And he's, you know, doing other more advanced cloud transformations. But back then, that was the check box. You could go to your C-E-O-C-F-O and say, we've migrated to the cloud by simply virtualizing your environment, was the standard for, uh, moving to the cloud, which gets us into kind of the enterprise thinking versus the provider thinking.

The enterprise thinking is, you know, ai, uh, it technology capability is a tool. It is a hammer. And we're going to sweat that asset until, uh, until there's no, until we can't, until it just needs to be replaced.

That's the majority. I think of this. Well, but you, you can do pay as you go with Amazon now on, on, on, on premise.

You can do pay as you go with HPE. So the, the pricing models have gotten so you can sweat other people's assets now. I mean, it's a all new world.

The options that enterprises have are unprecedented today, um, where you wanna put those workloads. So, And because of that, abstraction enterprises can stay on legacy architectures longer. So I don't need to re-platform my, uh, my storage system because Amazon has provided me S3 for 14 years, and I don't need to look at a different protocol for accessing data.

There is a argument amongst technologists that you should always modernize. You should always stay current. But we've abstracted the, at least the, the underlying technologies that we don't need to.

We can build a cloud app 12 years from now, use load balances, blah, blah, blah. If there's a new modern architecture, we, we don't worry about that because we've established the baseline for the, or the value. So yes, there's a new nail gun that will, uh, provide me faster, the ability to frame faster, but I don't need to frame faster.

I, the nail gun is working fine for me. Well, here. And, and that this is where the hyperscalers have shot themselves in the foot as well.

So if you, again, if you look at the management level, uh, the CFO is asking the CIO what is their, their budget projections for the next three years. That is very standard. The CFO wants that reliability, especially for public companies.

They wanna, they wanna report this out. Uh, government is also the same way. They have to have these budgets.

And when, when you're using an external provider, uh, and you don't have that a three year contract, you can guarantee they're gonna jack up the rates and the, the cost of cloud has gone up faster and inflation for years. And, uh, SaaS as well, same thing. Cost increases have gone up faster and inflation, the CIO can't tell a CFO accurately what their future it cost is gonna be.

Now, if you have your own assets, you know exactly what your, what costs are gonna be. And that has really changed, uh, the equation, the, the recent price increases for SaaS and cloud have dramatically changed the, the acquisition, um, uh, you know, template for many organizations, they're now look much more amenable to saying, we want predictability. We want financial predictability in in cloud.

It's killing us the last few years. So that's also another factor. And that gets us back to paid licenses as well.

Mm-hmm. Yes. And which gets us to an amazing real life use case that we get to watch unfold in real time.

The IPO of Core Weave. Hmm. So for those who don't know, core Weave is an NVIDIA backed startup that basically provides AI cloud services.

They're laser focused on providing services, high level abstractions, cloud abstractions for people running, wanting to run ai. But Diane and Kimberly, I, I can't help but realize after coming out of, uh, GTC the week before we're recording this, that plant obsolescence is a real thing. When you're a, uh, hyperscale provider like Core Wheat and now the economics of the, you know, millions of dollars, uh, hundreds of millions of dollars that you spent on H one hundreds hyper chips, I can now get much better efficiencies, operating efficiencies.

There's like a five to six month ROI from, from my OPEX budget to move to the next generation of GPU. How do you, I I guess the, the beyond the news, we'll get into like the TikTok of the stock price, but this, we're infrastructure folks, how does Core Weave operate at that level? Like how, how can they d cloud providers are sweating assets?

Amazon just announced that they, uh, they're extending some of their depreciation schedules to what, like from five to seven years, so that increases their profit margins. I can't imagine that core weed can do this. Not with AI chips.

You mean sweat their Yeah, you can't, you, you can't sweat a H 100 not from a provider perspective. You're lucky to Get a year. I mean, they, and they want two years outta AI chip these days, but you're lucky to get a year.

Maybe that changes who the target markets are. One, one of the things we're finding, um, you know, is that we know that CPUs can address some of the work that has to be going on with Inre and for the enterprise and for the other companies. So the, the brace right now seems to be after these AI centric companies that are building a business that may be around ai.

Um, and I'm not privy to the data that we've got coming out of from probably Bradham and, and, and, and Nick, Nick patients on the use cases. But what I hear from the teams that are testing things, you still have the ability to use, um, chips or that are probably sold to be able to get things done. Now, the flip side of that is the cost is the energy consumption because we've, we seeing, we're seeing some efficiencies and such coming out of the new technologies.

So the economics will be interesting as they look at that, both from a use case, where can the older ones being used, is there a different market for them? Um, as you upgrade to the other stuff, it's classic what we're doing in the data center, right? We, we, we, those old, old servers that's 37 signals are doing is not doing core applications or running transactional operations.

I'm sure it's going down and doing some low level, not low life, low level file serving. Yeah, well, those productivity applications and not transaction applications. Yeah.

And not, Yeah. And then the enterprise, just the culture and the math and enterprise is different. When you're not providing your technology as a service, the clock of your technology does not directly impact the in output.

So for example, if I bought a H 100 today based on a justification around some capability over the next five years, it really doesn't matter if a B 300 comes out or a b or, or I'm sorry, a BG 300. The naming, their naming is crazy. And you can get 40 x the performance and have an ROI over.

Uh, Well, if you need to have competitive competitive pricing, it might matter, right? I mean, that's the thing. So the, so that, so, but if I bought, so if I have, uh, if I didn't outsource that risk to the provider, if I bought the H 100, the just the, I've been in the enterprise on the operations side much too long.

I'm way too bitter around this. So correct me if I'm wrong, but I don't pay my power bill. If I run infrastructure, I don't pay the power bill.

So if operationally I can, there's a ROI that's five, six months that does not impact my budget. 5 million for these H 100 things. When I have to answer to the CFO, he wants to, he or she wants to know, am I on course to continue to u use this, uh, depreciable assets through its useful life, the operations, the, the, uh, if I was to buy a D 300, the new chip, and it saved power from a OPEX perspective as the c as the, as the operations guy from the IT side, I get no benefit from that.

I get no professional benefit from saving the facilities folks cost, even if, and, and the simple fact of the matter is, I didn't save them cost. They're gonna use the same amount of power, et cetera, et cetera. So that culture completely changes with providers.

The, uh, the, it's Why you see, why you see them buying nuclear power plants now, it, it is all been focused down at that, at the facilities level. Now, It would be interesting to pull their SEC filing and see what depreciation schedule they're using on their AI GPUs. And that's, I think that's a lot of the debate change policy.

'cause if I was, if I was a financial investor looking at them and saying, what's, you know, what's their balance sheet look like? Um, what's their capitalization look like? That would be one of the areas that you'd have to dig into.

And our CEO who's, uh, pretty astute around some of these things, uh, Dan Newman posted to that effect. The depreciation is a really big question around Core Weave. And their IPO, uh, results, I think hint towards the, he's probably correct.

They were targeting IPOing at around 47 to $55 for initial offering. Uh, uh, but they ended up much softer at about $40. They didn't, uh, put as many shares out for the offer.

5 billion CLO as opposed to the $2 billion they one raised. Well, they haven't, I they haven't gone out yet, have they? I think they, they, I think their IPOs today, today, uh, was, uh, yeah, they opened at 40 at $40.

So the, it is a interesting, it is a interesting, uh, I just looked it up. It looks like, uh, so the hyperscalers, uh, have gone from about a three year GPU, uh, depreciation schedule recently till five years, which is very surprising to me. I would've figured that, that they'd be going the opposite direction.

So that's Google Alphabet. Amazon and Microsoft have all gone from about three years to five to six years now. Okay.

Maybe on the, If all of this is a really big proxy on not just Nvidia and whether or not they'll be able to continue their billions of dollars of growth every year, but whether or not AI is actually being a adopted at the pace and speed as advertised, and it would be, I think, hard to believe, or at least hard to reconcile, I really need, would want to dig into the math of why that depreciation schedule is expanding versus, to your point, Diane, shrinking. Well, I, I, I think it must be being used on other workloads as well. Um, but what's, um, yeah, what's really interesting is what companies are gonna do going forward in terms of, uh, how they're, how they're gonna source source compute.

I think I see we're very much going towards the spot market where, um, if you, you know, for a lot of, uh, the, the cloud primitives, you know, cloud storage networking, um, you're gonna see much more of a spot market, uh, evolving. You, we are seeing it now. So, All right, so the, again, this has been a engaging conversation.

We're starting going the meaty part of the spring from, from conference season, uh, GCN Google Cloud. Next is just a couple of weeks away. Albe at that, uh, early summer, late spring, uh, Dell technology world is coming up.

We're in, uh, the more traditional part of the infrastructure shows. Diane, Kimberly, anything that you're keeping your eye open for? So Q con is next week, Uhhuh and, and the big and the, the first day of coupon, which is not the official day, but they have all these sessions that are going on is a full day and it's the highest attended full day break session is on platform engineering.

Ah, yes. Put that in your bon It is hot. Happy for sure.

We need to have an extended conversation on platform engineering. It's one of my favorite topics, and it's kind of like that cloud conversation we just had. How companies are checking the box when they have VMware and say that they have a cloud.

Well, you don't necessarily have one and you probably don't have a platform engineering team if you think you Do. Well, we had a lively debate, um, on Textron gang with, um, Mitch Ashley and a couple of others about platform engineering, DevOps, secure ops, all that kind of thing about, you know, whether or not platform engineering was really IT operations, where that fits how that's operating. So let's see what happens.

Kind of, I don't think s people can do platform engineering, but anyway, that's, that's, I think that's a, I think that that is a fascinating insight, Diane, and I think it will stick a pin in it for post-up coupon. I think tune, make sure the tone in next week, not just to find out about this whole platform engineering conversation, but where in the world is Diane Hinchcliffe that Will, will we, we, the, the, I didn't even get myself, so he doesn't even know the world wants to know. Talk to you soon.

Thanks a lot for, uh, uh, staying with us.