 From theCUBE Studios in Palo Alto in Boston, bringing you data-driven insights from theCUBE and ETR. This is Breaking Analysis with Dave Vellante. Slowing cloud growth has led some to the conclusion that repatriation of workloads is a factor in suppressing cloud spending. While we've been skeptical about repatriation as a broad-based movement, anecdotal evidence suggests that it is happening in certain situations. While we still don't see repatriation broadly in the numbers, certain examples have caught our attention. In addition, the impact of AI raises some interesting questions about where infrastructure should be and will be physically located and causes us to revisit our premise that repatriation is an isolated and negligible trend. Hello and welcome to this week's Wikibon Cube Insights, powered by ETR. In this Breaking Analysis, we look at a number of sources including the experiences of 37 signals, which has documented its departure from public clouds. We'll also examine the relationship between repatriation and SRE ops capabilities. And as always, we'll look at survey data from our partners at ETR and some recent FinOps data published by Vega Cloud. Several credible sources have published on this repatriation topic. On the bottom left here, IDC published a report in 2018 that said over the next two years that the respondents expected 50% of their public cloud applications to repatriate. That is an astoundingly large figure. Now this study was done by Michelle Bailey and Matt Eastwood, two highly respected, incredible analysts. I know them both personally and well. And from what I can tell, this was not a sponsored study. But this was a very surprising figure. Now of course, the most discussed report was and continues to be the post from Sarah Wong and Martin Casado of Andreessen, which posited that cloud expenses would become an increasingly large component of cost of goods sold for cloud native SaaS companies at scale and would become so onerous as to chew up its profits and either force repatriation or a large discount concession from cloud providers. The authors used a Dropbox case study as an example of the potential savings from repatriation where the company saved $75 million by moving infrastructure back on-prem. Now, you'll find some evidence of this large cogs conundrum in Snowflakes 10K. In the most recent March 2023 filing, you'll find a note that says, quote, in January 2023, we amended one of our third party cloud infrastructure agreements, presumably AWS, effective February 1st, 2023. Under the amended agreement, we have committed to spend an aggregate of at least 2.5 billion from fiscal 24 to fiscal 28 on cloud infrastructure services. And they're obligated, it said, to pay any difference between the commitment and what they use during that term. So that sounds pretty onerous, right? However, as you read on, it says, quote, the remaining non-cancelable purchase commitments under the agreement prior to the January 2023 amendment, the aggregate amount of which was 732 million as of January 31st, 2023 is not reflected in the table above as the company is no longer required to fulfill such commitments. So two points that we've made before and are going to reiterate here. The second statement that Snowflake made and that I just shared with you is an indication that Snowflake has renegotiated its contract with AWS, again, presumably AWS and retired $732 million from its previously committed spend ostensibly to get a larger discount. So so far, the quid pro quo between Snowflake and AWS has been and it's same with other customers. You buy more and we'll give you a break. So it doesn't kill your cogs in the case of a software company and customer expense from a customer standpoint. Not that Snowflake's not a customer, but a conventional customer. And this is likely going to continue indefinitely. Amazon's not likely to lose a customer to repatriation on balance. Now, the second point relates to Dropbox. Dropbox is kind of a bogus example because you cannot build a third party storage service on top of S3 and make money. All you got to do is talk to David Friend. He's the CEO of Wasabi or Gleb Budman. He's the founder and CEO of Backblaze, the two storage companies and they will confirm this. S3 or storage service, third party on top of S3 can't make money. So well, it might make some sense to do some things on-prem such as what ZScale or CrowdStrike have done as was reported by Wong and Casado. So far, repatriation from SaaS companies like Datadog and Snowflake has not been a factor. But let's not call it game over for cloud just yet. Lori McVitty put out an interesting post late last year describing the relationship between site reliability engineering skills, SRE, and cloud repatriation. We're showing this here, the curious connection between cloud repatriation and SRE ops. Her point was not that SRE causes repatriation, rather that SRE skills make it easier for companies to operate on-prem infrastructure using a cloud operating model. She cited an F5 report that was really interesting and showed a significant jump from 2021 to 2022 in customers repatriating apps. And she asked the right question, i.e. it's not whether or not repatriation happens. We know it does. The question is what percentage of workloads are actually moving back? Now it doesn't appear to be 50%, as the IDC study suggested, based on the data we have. But her point is that many more customers today have SRE ops capabilities and are in a much better position to repatriate if there is a business case, which brings us to the 37 signals case study. David Hennemire Hansen is the co-founder of 37 Signals. He's the developer of Rails. And it's a company that I've long admired in his products like Basecamp that I've loved. He's published a couple of posts on their departure from the cloud and he used this term sovereign cloud. You can see we put here sovereign cloud in the upper left, why we're leaving the cloud. And he's put this term out as an alternative to private cloud or on-prem. I think VMware has used this term for years. I think going back maybe even before 2019. And our colleague, David Nicholson, he loves this term as well. He basically says, look, it's all IT and he's applying that the off-prem and the on-prem worlds are converging in some type of equilibrium, which is really interesting point. And that's essentially what 37 Signals is saying. A main point that he made is he being David, a main point he made is renting infrastructure for a mid-sized firm like 37 Signals is more expensive now than when they were starting out with new products like hay in the cloud. The other key point is what Lori McVitty was saying. The skills overlap between what it takes to run software in the cloud versus on-prem is like 80 to 90% correlated with some differences, of course. First, you don't need thin ops and forensic accounting to understand your cloud bills. And second, you have to replace failed hardware like disks. And he said, these differences are minor and easier than learning Kubernetes, ha-ha. Here's a picture of the 37 Signals software in cloud and some nice shrink wrap boxes on a pallet and you can see the infrastructure they bought on the right. Now look, these guys are there geeking out over their Dell R7625s They got two AMD Epic 9454 CPUs running at two and three quarter gigahertz and 48 cores and 96 threads. So it's going to give them 4,000 virtual CPUs on-prem. So, and they got seven terabytes of RAM. They got 384 terabytes of Gen4 NVMe storage. And then Gen5 is coming. That's going to go to 13 gigabytes a second soon. So yeah, cool, sovereign cloud. So why isn't everyone doing this? Well, our friend over at Platformonomics, Charles Fitzgerald, created what he calls the cloud repatriation index. In this chart, he plots the growth rates of the big three cloud players against those of Equinix and Digital Realty, the big colos. And in the upper right, we show Fitz's repatriation index. And the way he calculates that, he takes the revenue of the two leading colos and divides by AWS's cloud revenue. And you can see the steady downward trajectory of the cloud guys. And you'll notice the chart below, sorry, the steady downward trajectory of the cloud repatriation index in the upper right. And you can see it's kind of flattening there. We put a little question mark, come back to that. But you'll notice in the chart below the growth rates based on 2023 projections, they're kind of converging between AWS in particular and both Equinix and Digital Realty, which are growth rates are actually popping up into the low teens. So that flattening repatriation curve in the upper right is probably going to uptick. So we're going to keep an eye on that. Now let's bring in some ETR survey data and take a look at whether there's any evidence there that repatriation is happening to a large degree. In this recent ETR drill down, ETR asked 400 respondents that said they were significantly cutting costs, how they were doing it. And you can see here, they're delaying projects, they're cutting staff, they're reducing consulting spend, they're holding off on hardware purchases, they're consolidating redundant vendors, by the way, consolidating redundant vendors. That was number one last quarter and had been in previous quarters and then reducing excess cloud resources. So reducing excess cloud resources might include some repatriation. We asked ETR to look into this and dig into any of the open-ended comments and repatriation wasn't specifically mentioned. That doesn't mean it's not happening in pockets, it just doesn't appear to be thing one, as Stu Miniman likes to say. Now, we recently received an inbound from Vega Cloud, a firm specializing in cutting cloud costs and they just ran a survey on FinOps. And this chart shows for customers concerned about costs, they drilled down and said, well, what are you doing about it? And like the ETR data, we see staff cuts, we see project delays, we see vendor alternatives, kind of like consolidating vendors maybe, but no specific evidence of repatriation as a call out. And just as an aside, what was really interesting was Vega asked respondents which line items on the cloud bill were giving them the most heartburn from overages relative to expectations. I would have thought compute and bandwidth would be the big culprits because, but that's where the spending is, right? But look at database, it's number one. I wonder how many snowflake customers were in this sample. We asked, because snowflake is very popular and charges customers for consumption as a bundle. So they don't necessarily know they're spending on, whether it's on compute or storage. And at least that's the way I understand it. So they didn't have that level of granularity. So the snowflake customer doesn't see that granularity of, like I say, storage or compute. They just see a consumption bill. We couldn't get that level of, we didn't, they didn't know, Vega didn't know whether or not there were snowflake customers in there or not. So we'll leave that for now. But the last bit of ETR data that we want to look at is from a survey on cloud adoption relative to on-prem and the preponderance of hybrid cloud. Now we've shared this before but we'll take a little different look at the data right now. It's from another ETR drill down from late last year which breaks down the percent usage between public and private cloud. The key points are one, 61% of respondents cite the majority of their usage is on-prem today. So that means 40% have most of their work being done in the cloud, which makes you question that claim that 90% of all workloads are on-prem today. That 61% figure that drops to 42% in two years. But the real call-out stat to us that we've highlighted here in red is that only 14% of customers today say they're all in on the public cloud and that is expected to remain flat in this sample over the next two years. So first, if there's repatriation happening at a large scale, it's not on these numbers. And two, the picture is very clearly hybrid as the dominant model for this foreseeable future. So the other question we pondered is how will AI impact this picture? Because that's all we talk about as AI these days. Well, first of all, foundation models like GPD as a tide that will lift all compute boats. It was interesting to see the stock market's tepid reaction to AMD's quarter. They beat estimates despite the challenging environment and soft PC market and the stock dropped. But when it was announced that Microsoft is helping finance AMD's AI chips, the stock popped. And while a lot of that is hype around chat GPT, it's also recognition that we're entering a new era where AI is going to drive a massive demand for data center hardware, which begs the question, where's that hardware going to live in the cloud or on-prem? So we asked, where are open AI's servers? Well, we know they're in Azure with the deal they have with Microsoft, but they're evidently also in Ohio on a supercomputer where the training is done. I actually asked chat GPT and got a vague answer. But when you look at AI, there's clearly an affinity between it and high performance computing. And many of those HPC workloads will land on-prem. For example, if you're a university and you have a supercomputing center, you're going to want to show that off to people and donors versus having hidden resources in a remote cloud. And then in a renting infrastructure is economically viable for bursty use cases and unpredictable swings in demand. But like the cloud versus on-prem argument, similar arguments apply here, especially with large scale implementations. I mean, I'm just sort of picking a low watermark, but think about like 50,000 cores are up. Some supercomputers have millions of cores. So economically, it's going to be more viable to keep them in your own data center. There are also legitimate concerns about leaking IP. We recently saw Samsung electronics ban chat GPT after an employee mistakenly put, or I guess unknowingly put proprietary code into the software. Didn't on purpose, but he didn't know that he was given his IP to or her IP to chat GPT. This makes AWS's announcement of bedrock quite interesting. AWS is not building a free service like Google search or Bing or open AI, rather it's offering large language model services and tools so you can build your own, presumably making IP leakage less likely. But then of course, there's always Alexa. So we'll see what happens with that. And in fairness to open AI, CEO Sam Altman would tell you that they don't want IP that shouldn't be in the model. Rather they only want folks IP who want to be in chat GPT as an example. At one time, I was working, asking chat, I was, what I was doing was ego GPTing, I guess I'll call it, just to see what it said about me. And chat GPT said I worked for the Wall Street Journal at one point, which obviously want to correct it. It was going to a nice compliment, but I wanted that corrected. So oftentimes you want your data in GPT models. And the last point is, as Matt Baker said in our LinkedIn post last week, this is batting practice for AI. The game will be played at the edge and in the world around us. Adaptation and inferencing do not favor centralized models, e.g. public cloud. Now, while Matt is biased, we agree. I think he's right on this front. AI inferencing is going to explode at the edge. It's not going to happen in today's public cloud. Now, how the public cloud players approach this is going to be very interesting. They got autonomous vehicles, they got robots, they have satellites, and they have other edge infrastructure. So yes, we're smacking GPT balls out of the park before the first inning has even started. So let the games begin and we'll see what happens. Okay, that's it for now. Many thanks to Alex Morrison, who was on production with Ken Schiffman, Alex also manages our podcast. Kristen Martin and Cheryl Knight help get the word out on social media and in our newsletters and Rob Hofe is our editor-in-chief over at siliconangle.com. He does some great work for us. Thank you. Remember, all these episodes are available as podcasts wherever you listen, just search, breaking analysis podcast published each week on wikibon.com and siliconangle.com. I think we're up to episode 177 now. So check that out. And you can also check out thecube.net for all the video action on theCUBE. You can email me directly if you want to get in touch, dava.balante at siliconangle.com or DM me at dvalante. You can comment on our LinkedIn post. Pitch us if you got some good ideas. You know, we'll listen. If not, don't take offense. We get a lot of them, but keep trying. And please do check out etr.ai. They've got the best survey data in the enterprise tech business. This is Dave Vellante for theCUBE Insights, powered by ETR. Thanks for watching everybody and we'll see you next time on Breaking Analysis.