Beyond The Rumors and Into the Future of Cloud Computing [CL 94]

00:00:00
Ned: What are we doing here?

00:00:01
Chris: What’s going on?

00:00:03
Ned: I don’t know. We just asked the 4 Non Blondes. I don’t think they knew, either.

00:00:08
Chris: No, no. They were asking that question a lot.

00:00:11
Ned: Yeah, I guess that was the central conceit. And I guess they never figured it out, and just chose to disband. And then one of them went on to write songs for Pink, so I guess maybe she did figure out what’s going on, and what’s going on is making a lot of money in residuals.

00:00:27
Chris: And having a lot more free time, I imagine.

00:00:40
Ned: [laugh] Yes. Hello, alleged human, and welcome to the Chaos Lever podcast. My name is Ned, and I’m definitely not a robot. I’m a real human person who loves a good rosé on a bad day, driving with the top down to hear my—feel my hair in the wind. Or the wind in my hair [laugh].

00:00:56
Chris: Wow.

00:00:56
Ned: It’s just—I don’t know.

00:00:56
Chris: This is going great.

00:00:58
Ned: I’m going to start all the way over.

00:00:59
Chris: No, you’re not. You’re grinding forward.

00:01:02
Ned: Okay. And systematically eradicating all life from this planet. I mean, puppies. I love puppies. With me as Chris, who’s [laugh] also here.

00:01:13
Chris: You’re systematically eradicating puppies. That’s worse.

00:01:16
Ned: All of them. Just gone.

00:01:17
Chris: That’s so much worse.

00:01:18
Ned: No fucking way, man.

00:01:20
Chris: If you just rolled up, and you were like, “You know what? It’s time for humanity to go,” and you just had, like, a big red button. Everybody’d be like, “Yeah, I get it.”

00:01:27
Ned: [laugh] Like, look around and be like, “Yeah… w-, we did this.”

00:01:32
Chris: But you start going after puppies, I mean, dude.

00:01:35
Ned: It’s true. Like, whenever they really want to express that someone is a villain in a movie, they injure a dog.

00:01:42
Chris: Yeah, kick the puppy.

00:01:43
Ned: That is, like, universal language for this person is despicable.

00:01:48
Chris: Usually while eating an apple.

00:01:50
Ned: It’s why it was like wildly surprising to me that they tried to make Cruella de Vil—they tried to make, like, a movie just about her where she was the sympathetic character. Because she killed dogs, man. Like, that is her central thing is to kill dogs and wear them.

00:02:08
Chris: She didn’t yet.

00:02:10
Ned: I guess. But like, you know where it’s going, right, and—

00:02:13
Chris: Yeah, well—what was the name of that movie?

00:02:15
Ned: Terrible.

00:02:16
Chris: It was called Cruella?

00:02:18
Ned: I think it was called Cruella, yeah. And I… honestly, I haven’t actually seen it because it just seems stupid.

00:02:25
Chris: It’s fine.

00:02:26
Ned: Is it?

00:02:27
Chris: It’s fine. But yeah, you’re right. They don’t—there’s no dog killing in the Cruella movie. That’s for, you know, part two. Cruella Part Two: Shit Gets Dark.

00:02:37
Ned: [laugh] Real dark. I watched the original 101 Dalmatians recently and A, it’s very long, and B, it’s very jazz-forward. It’s basically a jazz movie.

00:02:50
Chris: That explains why it’s so long.

00:02:51
Ned: Yes.

00:02:53
Chris: We didn’t need a 17-minute flute solo, Cruella.

00:02:56
Ned: Are you sure about that?

00:02:58
Chris: No.

00:02:58
Ned: It’s my favorite kind of solo. Not to go completely off the rails—but hey, we’re already here—my son wants to do a solo in his next jazz band concert, and he wants to use one of those, do you know the thumb… pia—I don’t want to call them pianos, but like, it’s like a hollow wooden block that has the metal tongs on it, and you use your thumbs to ring them.

00:03:27
Chris: I know what you’re talking about. I don’t know what it’s called.

00:03:29
Ned: It’s called a kalimba, but if I just said kalimba, 99% of the people would be like, “The what?” So, it’s sort of like a little hand harp, I guess. Yeah, he wants to do a solo with that at the next jazz band concert, and I was like, “Dude, you are my son.” [laugh].

00:03:50
Chris: [laugh].

00:03:50
Ned: Let’s talk about some tech garbage, huh?

00:03:52
Chris: Do it.

00:03:53
Ned: Cloud is back, baby… as I predicted at the beginning of this year. That’s right, there’s a rumor going around—not sure if you heard it, Chris—that companies are abandoning the public cloud. They’re all moving it to rack space or whatever Equinix data center is closest to them. We’ve already done at least one show on cloud repatriation and why it’s actually bullshit, and yet, like cockroaches and R. Daneel Olivaw, the messaging somehow persists. You appreciate the deep cut, I know.

00:04:29
Chris: I nodded. I nodded knowingly.

00:04:32
Ned: Thank you. It’s a perfect thing to do on an audio medium. My stated opinion is fundamentally this: are some workloads being moved back from the public cloud? Yeah. Is it a majority? No. Do the cloud providers need to worry? No.

00:04:52
Chris: I would actually modify that last answer. Do the cloud providers need to worry? Absolutely not.

00:04:59
Ned: [laugh] Just—so I would qualify that a little bit, at least in the short term, they have no worries. And by ‘short term,’ I’m talking, like, the next ten years.

00:05:08
Chris: So, the next ten years in infinity—

00:05:10
Ned: Yeah.

00:05:11
Chris: —in internet time.

00:05:13
Ned: Essentially, yeah, because beyond that, assuming we haven’t fully succumbed to any number of doomsday possibilities—and there are many—a decade is simply too long to predict anything in technology. The artificial general intelligence could rise to prominence and build colonies on Jupiter using the remnants of Amazon’s logistics network after the great drone war of 2025. If you don’t think that’s plausible, I don’t know if you’ve been paying attention [laugh].

00:05:42
Chris: I don’t think we’re going to have to wait that long.

00:05:44
Ned: [laugh] I was reading through the AWS earnings report—well, the Amazon earnings report—and one of the things that they mentioned is that they’re continuing their drone delivery program in the UK. And I was like, “There’s a drone delivery program in the UK? Do they know that?” It seems awful. But back to the matter at hand.

00:06:04
While the cloud providers—and for cloud providers, I’m talking about AWS, Azure, and to a lesser extent, Google Cloud because no one else is even in the same ballpark—they have continually grown year-over-year since their inception. And their rate of growth, it has slowed in the last couple of years. There was this tremendous spike that happened during the pandemic, as we all remember. Companies raced to transform digitally, and that pretty much just meant putting up an e-commerce site, but that wave has crashed. And 2021 through the first half of 2023 looks pretty anemic if you looked at the cloud provider growth.

00:06:48
Now, part of this is just the law of large numbers, which is a phrase that you hear often enough that you might think you know what it means, but I think illustrating it with some hard numbers might actually help put it into context, if you will. So, I looked it up, and in 2017, AWS had revenues of $12.2 billion, which was an increase of $4.3 billion over 2015. On a percentage basis, that’s a 54% increase year over year, which is preposterous. We should all be so lucky.

00:07:26
Chris: Bonkers, some might say.

00:07:28
Ned: Bonkers, indeed. Now, if you fast-forward to 2023, AWS had revenues of $90.7 billion. Yes, that was an increase of $10.7 billion over 2022. Now, despite the amount of growth being two-and-a-half times bigger than the growth in 2016, the percentage growth is 13%. And that’s what we mean when we say the law of large numbers.

00:08:01
Chris: And also, this is what I should have done if I wanted to be independently wealthy. I should have just started AWS.

00:08:08
Ned: I guess. Even though AWS grew their revenue two-and-a-half times more than in 2015, the percentage growth was one quarter of the size, and because that’s the number that we—and by ‘we,’ I mean analysts—fixate on, it seems like growth is slowing, and maybe the public cloud is in trouble. I assure you, it is not.

00:08:31
Chris: Yeah, let’s focus on that on ‘billion dollars worth of’ number.

00:08:35
Ned: [laugh] It is $10.7 billion of not, if you will [laugh]. Let that sink in. Let’s do another fun example because I like fun things. So, do you, Chris. Microsoft’s intelligent computing category—that includes Azure and Office 365—had revenue of $32.2 billion in 2018, and that was up $4.7 billion, or 17% over the previous year. In 2023, they had growth of $13 billion over the previous year, which also comes out to exactly 17%. You could say, looking at that, that the growth rate is flat. And you would be right. Or you can say that growth had increased 4x and also be right. Math is fun.

00:09:28
Chris: I have a headache, and we’re only five minutes in.

00:09:30
Ned: [laugh] I’m so sorry. But like, that’s why when people just say off the cuff, you know, ‘law of large numbers’ and we move on, it’s really important to think about that a little bit and realize how ridiculous our expectations have gotten around perpetual and constant accelerating growth. It’s just not realistic. Except for with Tribbles. Very, very common.

00:09:56
Chris: That is. This is the trouble.

00:09:58
Ned: I know. “The Trouble with Tribbles,” even. And we’re sued [laugh]. So, with all that in mind, growth did in fact slow down in 2022 and 2023. So, is this the beginning of a great decline? Not if AI has anything to say about it. I read through the earnings for all three of them—you’re welcome—and the Microsoft quote from their most recent earnings report really says it all. I’m just going to quote the block. “We’ve moved from talking about AI to applying AI at scale. By infusing AI across every layer of our tech stack, we’re winning new customers and helping drive new benefits and productivity gains across every sector.” End quote. That was Satya Nadella. He’s not wrong. They really have infused it everywhere. Yay.

00:10:51
Chris: And it appears that it’s not horrendous, either.

00:10:55
Ned: I think it’s very uneven [laugh]. In some cases, it’s tremendously helpful; in other cases, it’s a dumpster fire, but hey, they did pour it across every product in their entire portfolio. Not all of them are going to get it right.

00:11:11
Chris: Fair point.

00:11:12
Ned: So, let’s start with high-level numbers on the big three. AWS had a $97 billion run rate with 13% year-over-year growth, Azure was at $74 billion with 28% growth, and Google Cloud is a distant third with a still impressive $37 billion and 26% growth. Some of these are estimates because Microsoft doesn’t have the same financial year as Google and AWS because they hate me. Specifically me. They called me. Satya was like, “Ned, we’re changing our financial year, retroactively, back ten years.”

00:11:49
Chris: Yeah, when I’m emperor, that’s the first thing that’s changing.

00:11:51
Ned: [laugh] The financial year?

00:11:53
Chris: Do you know when the year starts, Ned? January 1st.

00:11:56
Ned: [laugh].

00:11:58
Chris: End of conversation.

00:12:00
Ned: I agree. It’s so confusing because Microsoft just closed financial year ’24 Q2. No, [laugh] that’s not how it works. You’re not even close. Anyway… I digress. If you break out the AI spending from Azure, the growth rate actually looks closer to AWS. So, if you just separate all that AI spending in its own little realm, the rest of traditional compute, that was about the same growth rate as AWS, which means that traditional workloads are growing at a normal pace. The real growth in Azure is being driven by hosting OpenAI’s products, and Azure AI’s own offerings. And you can see that borne out if we look at another metric, one that is more indicative of the future outlook than the growth numbers. There’s an old saying, Chris, I’m sure you’ve heard this, that cloud is just someone else’s computer.

00:12:55
Chris: It’s not just a saying, my friend. It’s a t-shirt.

00:12:58
Ned: It sure is. And while it is a gross oversimplification, it’s also really—it’s broadly correct. The cloud providers have a metric shit-ton of physical computers that they time-slice amongst all their different tenants, but even with the best possible bin packing of workloads across their existing fleet, eventually they run out of space, and they got to buy more. Like, a lot more. In fact, the cloud companies need to project growth out over the next few years to figure out things like how many new data centers are they going to need to open because that takes a while to do. You can’t be, like, snap your fingers; one pops up tomorrow.

00:13:40
Chris: What about, like, a long afternoon? You can’t put up one data center? Jeez.

00:13:43
Ned: Jeez [laugh].

00:13:44
Chris: I thought you were a professional.

00:13:46
Ned: [laugh] That was your first mistake. Which existing data centers need to expand because, again, that takes a certain amount of effort? And also just how many servers do they need to buy in the next six to twelve months? Data centers, power plants, and racks of thousands of servers don’t just happen. If Microsoft needs 100,000 1u servers next quarter—and that’s probably kind of close to what they order—they can’t just put an order on TigerDirect and cross their fingers. I mean, you know, maybe if they do the TigerDirect professional level.

00:14:23
Chris: That’d be really funny if they did, though.

00:14:24
Ned: [laugh] TigerDirect would be, like, “We’re rich. We’re out of business.” This stuff all takes time and lots of money, so hence, capital expenditures are a good way to track the amount of growth each cloud thinks that they’ll have in the next six to twelve months or even beyond that. Charles Fitzgerald over at Platformonomics—which is a hell of a thing to try to say out loud—he’s a CapEx-head—self-described—and he’s been tracking capital expenditures of the cloud giants for several years now. He released his report on it, back in the beginning of February, so let’s take a quick peek at those results. We’re going to get in some big numbers again because that is just the theme of today.

00:15:09
Collectively, the big three spent $127 billion on CapEx in just 2023, which is actually the same amount they spent in 2022. Spending was flat overall.

00:15:22
Chris: Growth is done. Cloud is over.

00:15:24
Ned: Ugh—

00:15:24
Chris: You lied to us at the top.

00:15:26
Ned: Stick a fork in it, it’s done. Like fondue. Mm-mmm. Fondue. I kind of want fondue now. Anyway, [laugh] Amazon in particular severely curtailed their CapEx spending in 2021, with the drop of 9% in spending in 2022, and then 20%, in 2023. Now, that’s all of Amazon, not just AWS, but AWS, in particular, spent less on CapEx in 2023 than 2022, and that’s the first time that’s ever happened since publicly available numbers for AWS were published.

00:16:05
Chris: Wow.

00:16:06
Ned: So, if you think for a second that AWS is really serious about AI, that metric might give you a little pause.

00:16:15
Chris: Maybe the thing is, they’re not serious about AI; they’re serious about talking about AI.

00:16:20
Ned: Yeah, and they have done a lot of that. I’ve heard some other folks say that if you listen to the thing that AWS is talking about the loudest, it’s the thing they’re doing the least.

00:16:31
Chris: [laugh].

00:16:33
Ned: Ouch. It’s like telling everybody how cool you are. [sigh]. Not that I’ve experienced. Shut up. In comparison, Microsoft’s CapEx spending grew 45% over the same time period to $41.2 billion, and the main driver behind that spending is… OpenAI and they’re never-ending thirst for more GPUs. They even said that they spent less than they wanted to because they simply couldn’t get enough chips from their suppliers. They’ve had to go so far as to rent capacity from other providers as needed because they don’t have enough GPUs. And they’re also spinning up their own AI chip silicon project, and diversifying away from Nvidia. Because they just can’t get enough of these GPUs. They’re delicious. They go really well with guacamole, you know? Yeah?

00:17:34
Chris: I’ll reluctantly allow it.

00:17:36
Ned: Oh, fine. Side note, since Nvidia has a stranglehold on the InfiniBand market, don’t be surprised if Ultra Ethernet development accelerates largely because of Microsoft and their need to diversify. Side, side note, at Cloud Field Day a few weeks ago, we watched a presentation from Broadcom that talked about their latest chips that can deliver 400 gigabits per second to the NIC with eight PCe—PCI 5 100G SerDes ports. So, it can actually do more than 400 gig on the PCI bus, they just don’t have a port that can do it on the NIC side, yet. If you pair that up with their latest Trident5 chip that goes into a switch, that can do 16 terabits of switch processing over 800 Gig ports. And even with the overhead of Ethernet, InfiniBand might actually be displaced.

00:18:35
Chris: So side, side, side point, this is actually a highlight of Chris Hayner’s Law of Large Numbers, and that is, once numbers get to a certain size, they have no meaning anymore.

00:18:49
Ned: [laugh]. I know. 16 terabytes flowing through that chip at line rate. Just… what?

00:18:56
Chris: Exactly.

00:18:58
Ned: Side—

00:18:58
Chris: Like, I really hope that’s all it says in Wikipedia, just, “Wut?”

00:19:03
Ned: With a U [laugh].

00:19:04
Chris: [laugh].

00:19:05
Ned: Side, side, side, side note—and counterpoint—Nvidia’s market cap is now $1.79 trillion, making it the fourth most valuable company in the world. So, maybe Nvidia ain’t all that worried. Look at you, your little lines. It’s cute. They do. It’s a cute. Side, side, side, side, side note slash lament, I had Nvidia stock in 2013, and I sold it. I hate past me a lot.

00:19:38
Chris: This is also Chris Hayner Law of Stock Investment, and that is, the most successful stock investor is the most forgetful stock investor.

00:19:48
Ned: [laugh] In this particular case, I was buying the house that I currently live in, so, like, I can’t be too mad about it, but, like, still I can.

00:19:55
Chris: Yeah, yeah.

00:19:57
Ned: Anyway, what I was saying [laugh], way back before we went down this rabbit hole is that Microsoft’s CapEx spending is ludicrous. And even Google managed to increase their CapEx by 2% to $32.3 billion. Amazon, Google, and Microsoft are all in the top ten biggest CapEx spenders on the planet. Who else is in there? Some other familiar faces: chipmakers. Samsung, TSMC, and Intel are all also in that list of the big ten. So is Meta with their incredible outlay for data centers since they don’t host on anybody else’s gear. And then, of course, planet-destroying companies, Exxon and Aramco, who love to, you know, pump petroleum, and they have to build things to do that.

00:20:52
Chris: That’s kind of sad. Well, I guess it’s kind of eye-opening is a better way to put it, that of the massive CapEx spenders, most of them are not in what we would call a traditional industry.

00:21:04
Ned: It was very shocking because you would think, if you look back 10, 15 years ago, none of these companies would have been that high on CapEx because they just sold software.

00:21:14
Chris: Right.

00:21:14
Ned: Software doesn’t require that. But once they became hosting companies, that changed the entire equation. Anywho. There’s also another phenomenon going on in the background that both Google and AWS have publicly acknowledged. It used to be that the servers that they bought would have an estimated three to five year lifespan, after which they would be retired and fully depreciated, but unlike my oven, microwave, dryer, and every other goddamn appliance in my house, it appears that server manufacturers are actually getting better at building them.

00:21:50
And at the same time, Moore’s Law has slowed down to a point where a five year old server is still a perfectly good asset. So Google, AWS, and I suspect, Microsoft, have all extended out the lifetime of their existing fleet by two to three years. Which means that while CapEx has to keep pace with growth, it skews the replacement rate by several years. They just don’t have to buy and replace those servers right now.

00:22:18
Chris: Right. And even if they do have to buy new, they can still use the old on different, intentionally lower-performing instances.

00:22:27
Ned: Right. They have plenty of their own services that may not need the latest and greatest that just does background tasks. Just keep using that hardware, especially if it’s still good. Now [sigh], I’m surprised that Google and AWS have not invested more heavily in AI from a CapEx perspective—yes, they constantly talk about it, but you know, seriously—but that might just be a simple matter of constrained supply at this point. Microsoft was the first mover here with OpenAI, and they hoovered up as many GPUs as Nvidia and AMD could possibly spare, and that really just leaves table scraps for Google and AWS, and neither of them appear to want to get into a bidding war with Microsoft over these chips.

00:23:11
So, I’d like to bring it all back here, bringing it all back to the core question of whether the cloud is in trouble. Based on all the available numbers, I am deeply skeptical of the premise. A slowdown in Moore’s law, longer lifetimes for servers, and the law of large numbers easily explains the smaller percentage growth in traditional cloud compute. AI has been a tremendous shot in the arm, much more so than any of the previously overhyped technologies like blockchain and crypto. Remember when crypto was going to rule the world?

00:23:47
Chris: I don’t want to talk about it.

00:23:48
Ned: [laugh].

00:23:49
Chris: Still too soon.

00:23:51
Ned: I watched, I think it was the Emmys or something that was at the… crypto stadium… or it was adjacent to it. You could see it in the background, and I was like, “Wow, they’re still calling it that, huh?” [laugh] That’ll change soon. Whether you like AI or not, it is very much here to stay. There’s no way to put that toothpaste back in the tube.

00:24:13
Microsoft has the first-mover advantage, but honestly, it’s still early days. I mean, ChatGPT came out a year ago. I know that sounds strange, but it’s true. So, AWS and Google can eventually get their supply lines in place and start bringing in all the GPUs they need. Even if some customers are moving their workloads back on-premises, there are still a large contingent of folks who simply do not want to manage infrastructure, and they’re happy for the clouds to continue to do it for them.

00:24:44
And as a last side note, because VMware is always on my mind these days, let’s not forget about the customers of VMware who look at the bill for next year and go, “Screw it. Let’s just move it all to the cloud.” I mean, honestly, half their stuff is probably already there. And maybe in the meantime, it was cheaper to keep some of the stuff on-premises. Broadcom is making sure it doesn’t stay that way. I’m kind of shocked, actually, that Microsoft, AWS and Google haven’t ramped up their marketing campaigns to migrate folks off of VMware. Scale Computing and Nutanix sure have.

00:25:22
Chris: I wonder if it’s more of an under-the-covers kind of campaign? Like maybe Microsoft, AWS, and Google just don’t want to make waves or don’t want to make headlines that’s going to piss people off at Broadcom.

00:25:36
Ned: Ah—like, I mean, they probably do care about their relationship with Broadcom. I started saying that out loud, and I was like, wait, no, that—some of the biggest switches that they buy are from Broadcom, or at least the chips that go on their biggest switches, so maybe you don’t want to piss off that dog. I don’t know. Or maybe, you know, they’re just quietly going to each customer that they have an existing corporate relationship with and going, “You know, if you need to get off of VMware, I can do that for you real quick.”

00:26:03
Chris: You said that so quietly, literally no one could hear you.

00:26:06
Ned: That’s okay. You could hear me [laugh].

00:26:09
Chris: Stop it [laugh].

00:26:10
Ned: [laugh] I’m trying out my NPR voice. Actually, that’s not even NPR. I wasn’t personal enough and soft in my delivery. That’s my NPR voice. It puts people to sleep immediately. I saw your eyes, like, drooping right away.

00:26:24
Chris: Yeah it was—how long was out?

00:26:27
Ned: Well, I got to say, I did just cover a whole bunch of financials about the cloud, so [laugh] we might put everybody to sleep [laugh]. I’m so sorry. Ah, no, I found it pretty interesting. Hey, thanks for listening or something. I guess you found it worthwhile enough if you made it all the way to the end, so congratulations to you, friend. You accomplished something today. Now, you can sit on the couch, ask ChatGPT if the cloud is collapsing, and get back a nonsensical answer. You’ve earned it. You can find more about this show by visiting our LinkedIn page, just search ‘Chaos Lever,’ or go to our website, chaoslever.com where you’ll find show notes, blog posts, and general tomfoolery. We’ll be back next week to see what fresh hell is upon us. Ta-ta for now.

00:27:15
Chris: I like how you observed my sleep-inducing episode about insurance last week, and your reply was, “Hold my fondue fork. I’ll knock the rest of them out.”

00:27:32
Ned: [laugh] Yep. Yeah. I really, uh, one-upped that, I guess. I don’t know. Words. Words are stupid.