All Media.
Hello on, Welcome to Better Offline. I am, of course your host ed zitron. Now we're in hate a season. Hate a season is when I bring people on allowed to get rude, because you know, this podcast is usually so reserved and nice in the way we refer to people, especially tech executives, which we've never referred to as funck wits or morons or dumbasses or shittheads or shit heels or ass wipes or fuck nuts or nob ends. We've never done any of that ever. Nevertheless, joining me today
is chief cloud economist that Doug Bell Corey. Quinn Corey, how are you doing?
I have delighted to be here because normally in all the other places I find myself, I have to be so reserved.
You have to pull your punches. Cory, today's Hatter season. Tell me about your feelings on Amazon Web services.
Oh, dear lord, it's it's like it's weird. I don't actually hate Amazon. People think I do because of the things I say and the things that I do. But if I hate it, then this directions, but ten years in I write a newsletter about it. If I actually hate the company and want nothing but ill things to happen to them. First off, that's a pathology and I
need a restraining order. And secondly, it always bugs me when people think that I have an ax to grind against the company, because ten years in, are you seriously suggesting this is the best I would be able to do if I actually wanted to hit them where it hurts?
Come on?
But yeah, they've been annoying the piss out of me lately.
So AWS is weird because maybe you can explain this to me. Over my entire tech career, I've heard this constant story that your AWS bill just it just goes up. It just arbitrarily increases. It's not really clear whether they're doing price increases, though I know those exist too.
Rarely open but occasionally increasingly these days.
Yes, but why what is it that makes the bill go up? Why? And why does this happen so often?
I used to think it was a conspiracy, and then I realized Amazon is in no way, shape or form organizationally competent enough to pull something like that off. It is the nature of charging per usage and honestly, assholes where Okay, we're going to launch a service and we're going to charge you for every gigabyte you put in it. That's all we're gonna do. Great, straightforward humans can rationalize
around that, great eazy psy. They did that when they first launched S three and beta back in two thousand and three. Is storage right, it's a big object store picture it as an infinite sized disc and you're pretty close. But they would that Originally they only charged for the data you stored there. Great, so people stored a bunch of objects that were zero byte in size, so they just used the names of those objects, and the ability to query with that object existed or not in what
its name was. As a free database, this at the time, also, according to legend, didn't just it wasn't just about taking free services. It was also just crippling the performance of S three at the time. So what they did is they started also charging every request you make with jarj a small fee. I think it's something like a penny per thousand requests of memory served. Don't book me on that. I look it up when it matters, and that solves
that problem. But then you have things like that that are patched on and tied to things, and tied to things and tied to more things, and in any complex organization. What happens is is now suddenly you have this morass of one hundred million dollars a year and spent great, what's in there that's costing all the money?
Well?
Everything? But if you look at anything individual and you don't think it's being used, Okay, am I turn I'm going to turn it off. If I'm right, it's going to save me a bit of money. If I'm wrong, it's going to take down production. And now I have a serious problem trying to serve my customers. So the safe bias is don't turn things off. Also, you're an engineer. Hypothetically, I realize that can be an insult to your part of the world. Roll with it for a moment. Exactly,
you're an engineer. You want something to spin up, and no one is letting you spin it up. You will take up residents, annoying the hell out of them every ten minutes until you get the access that you need. You do your job. Things are done. Two things now work against you. One, you had to ask and bag and plead them to spin that up. You want to hang on to it in case you need it again because you don't want to go through that. But this is an organization exactly exactly. This is not Amazon's fault.
This is human nature's fault. They haven't patched that yet, though they're I think they're trying their best iteration of a Laxa plus it. Why not, I'm kidding that thing just sells ads. We're good boy way.
So I'm an engineer and I've advocated for my thing.
Yes, and now you're done using it. No one is going, there's no there's no call to action to get you off your ass to go, and no one off or no one will come and badger you about it. It just it tends to exists. It's why people cynically say you're not charged for the things you use in the clouds so much as you're charged for the things you forget to turn off.
And Amazon loves this. I'm guessing they love the fact that there's just this cloud anchor system.
I would say that, on some level, contrary to popular belief, they don't love it as much as you'd think, because all right, I'm trying to run on my blog and I've spun up eight attempts at this blog, so it now costs eight times what it should. Great and the narrative becomes if they're not careful with all that wasted usage that's doing nothing is huh, the cloud is pants shittingly expensive. We should go back to data centers. There's really no one on the other side of this issue.
So why doesn't Amazon help people turn tune these things down and turn it off? They made some a board of attempts at part of the organization. But this is going to surprise you. Amazon is not the most organizationally competent company you're ever going to meet.
But what does that mean in practice? Is it just that they don't know they're ass from the air hole. They don't really know how things work. All of the above.
Silos, it's all silos. They're over two hundred services. And we long ago across the point where I can speak incredibly convincingly about AWS service is that do not exist two Amazon employees and not get called out on it because who in the world knows everything that Amazon does? AND's just th guy, I sound as plausible as AI these days.
Yeah, So it's just a labyrinth, the Leviathan of different SKUs and product categories that.
Make million SKUs. If you want to be precise, Yes, Jesus.
And those are all just different kinds of virtual machines and the like.
Okay, we'll start there. Sure, virtual machines. You want to spin up a virtual machine in Virginia, US East one? Okay, which one are you going to pick? Because there are approximately seven hundred different types you can spin up. Are you going to pick the right one? Of course you're fucking not.
Oh so how do you pick the right one? This is the thing because it seems that the people making these decisions of which virtual machines to get or whatever spin up on aws or engineers, right, Usually sulfware engineers software engineers aren't necessarily infrastructure experts, are they.
You've noticed very often companies will standardize on one particular size of instance across the board. Why well, because in development it's the one that the original founding engineer picked, and we'll come back and revisit this later ten years ago. No one is revisited.
M take that.
They do have a tool that I like, it's even free, which I'm sure someone loses sleepover called compute optimizer that looks at you're running workloads and says, hey, this one should be bigger, this one should be smaller. I was extraordinarily skeptical when it came out. It is better at figuring that out than I am with a tooling that I built internally. So once again I'm thrilled to turn it off. It's one of the Amazon at its best things.
I can only assume that the people responsible for that are being eyed for layoffs or.
Pips, because Andy Jesse and several NBA snipers are outside at the location right now.
It's frust stay, but I like AWS truly. Something changed. He got a job that no one in their right mind could possibly want but also could not refuse. And I feel like I sould a soul. Yes, absolutely, but he just he was the CEO of a double us. And he was he when he gave keynote talks, when he gave presentations, when I was fortunate enough to basically sneak my way into them, you could see the irrepressible humanity leaking out around the edges. He was a person.
Now he is a figurehead. He doesn't want to customer, just to be clear to Congress.
This shift was when he became CEO of Amazon or CEO of a WS.
CEO of Amazon. AWS was his own Fifdom, and he could do whatever he wanted to my understanding within that company, it was the only people that mattered, Brother Garrett who had an opinion on this were Andy Jassey and God. And God was sort of absence, so great, We're just gonna leave it to Andy. Andy went deep and he understood every aspect of this stuff. Factly, a year ago he went to Reinvent in Las Vegas and gave a talk on stage.
I think because he missed it.
He missed being able to get up there and talk about computers with people because now he has to ship underpants all over the world instead.
Right, But is he technical because he's an MBA? Is he not?
This is one of the dangerous things that I have found people underestimate Amazon leadership on is it is never the smart bet to assume that they don't understand the technical nuances of the things that they are working with.
They go so sarcastically deep in so many different ways that it is never a smart move to say that they don't understand it is unlike virtually any other tech company I've ever spoken with, other than small scale ones with technical founders, he very clearly understands the technology and used it to build things himself.
Right, But here's the reason I question that. Please, he's talking about general IVAI. I've just pulled up his notice.
Oh no, one understands how that works under the hood. Let's be very clear on that. And yeah, he lost the fucking plot when it came to gen AI.
That's the thing, because he's like, we're also using GENERATIVEAI broadly across our internal operations in our fulfillment network. We're using AI to improve in ventur replacement. So not generate IVEAI. Demand forecasting definitely not generative AI. And the efficiency of our robots not generative AI. I don't know, maybe he does know what he's talking about. He's just a fucking carnival barker now he's.
Doing to say that it all seems to be. And the trick to disprove everything you just said about that, rather everything that he said that you relate to us is very simple. Go out with an Amazon employee for several drinks and then ask them about their experience of using AI within the company. Oh, I have two different companies. There is no there is no congruity between some of those two perspectives.
So Corey, maybe you can help me with this because you, certainly, without saying exactly what we're talking about, you knew about a contract that I knew about way before I knew, and I knew about it months early. Yes, with a large company we discussed, so you're very well connected. It seems up until jeneralf Ai, like AWS was at least a somewhat sensible operation, that it was something that we had, if not focus, at least a focus on scale and
useful cloud computing. Yes, what the fuck is it about GPUs that's having everyone insane?
Because I want to point something out that I think is being lost in the streetwide because I've talked to folks at AWS about this, and folks who have left AWS, who have I used to dabble working in data centers, which means I know just enough to know what I don't know, and I've talked to these folks. When Amazon commits to being able to bring certain amounts of capacity online,
it is real capacity. It is each data center they build out, each region they build out, is a multi billion dollar investment, and it takes years to go from sign contract to its serving customer traffic. They are real when it comes to this stuff. There are a number of other companies building well, yes, and they agree that running it. They have operationalized how to run these things better than you or I would ever be able to do. That is their superpower. Other companies and they're spinning up
a data center, do not do this. They are, well, we've we've rented a warehouse out, we round up, throwing some generators in the back and were check. Call Comcast and see what they can do to get it hooked up to the internet netney to go. Yeah, you say other companies, You don't mean Google and Microsoft here, No, I'm taught, well, I am talking Oracle, but yes, Microsoft kind of Microsoft. Microsoft is a shitty at building data centers.
Azure has been a disappointment across the board. Capacity shortballs plagued them for a long time during COVID. They generally tend to view two racks in a data center somewhere as a region, which is weird and messed up. Google is real, but Google has made other interesting technical trade offs that I can see why they did it. But I would say when it comes to infrastructure, in my experience, AWS is number one.
Google is a close second. I have heard some crazy shit about Google that they're doing these things called stalks where they just like like create these pop up data centers.
For a long time. Every company does that for a point centering stuff.
Yes, anyway, but back to the major question, which is AWS seems like a sensible business that does real things. I'm not disputing that.
Yes, it prints money. Let's be clear on despite all, remember, I help companies negotiating money cloud contracts and understand what they're spending and where the money is going, which means that companies can say a lot. People say a lot of things. But one of my prime rules about this stuff is that customers lie mostly to themselves. They're not intentionally trying to mislead me when they tell me what's
going on. But I find the bills of the ultimate source of truth, because if you're not paying for it doesn't really exist. And the AI spend in most of these companies hovers, in my experience with a few outliers between five and seven percent of their total infrastructure spend. Right, it is not three times. Companies that are doing a three hundred million dollars a year contract with AWS are not turning around saying better make it four hundred because of all the GENAI that is not happening.
That's kind of my point though. It's like Amazon Web Services doesn't seem like a business that chases fats, or if it does chase them in a more sustainable way. They now they's credit.
They docked they Dutch blockchain almost entirely, and I was worried they were going to fall down that rabbit hole.
But I mean, they doing two hundred billion dollars in capex this year. What is it they see in AI? Because to your point, just now, what like seven to eight percent of a three hundred million contract, that's dog shit like that's that's not going to pipe out that Capex Sandy.
I believe this is my belief here. I can't base it on anything other than what I've seen Amazon do, uh I, and what I've seen from customers. The Amazon is selling the capacity that they are spinning up. They are spinning up as fast as they can spin it. The question that I have is twofold one. Okay, you invest in all these AI data centers in the AI bubble pops to flates whatever, how repurposable are those facilities, and that capital expenditure if it's data centers and networking, Yes,
very repurposable. Yeah, GPUs, well, we're all going to get real in online gaming in a few years. I don't except they.
Don't have video out, they don't have DisplayPort.
Yes, yeah, we still call them graphics, which I love.
I think I like the people are I've also heard people say general purpose units, which is the opposite of what a GPU is like. It is very much not general purpose. It was.
It became public through a press release that they have signed a thirty eight billion dollar contract with open AI with insane. There is zero doubt in my mind that, because I did some digging on this, is there line of sight for AWS to bring all on line thirty eight billion dollars of capacity to provide to open AI. Yes, yes, there is where my doubts come in. Is open AI good for the money?
No?
I have some questions here, mostly from reading your work.
And that's the thing. It seems like something about GPUs and large language models have broken everyone. That's kind of the point I'm getting too, because like everything I've read about AWS, and I've read because I'm a dickhead, I went and I read all of the previous coverage of the lead up to AWS becoming profitable. Kevin Rust, by the way, literally a month before they did so, wrote an article saying it would be a while. What fucking three points in that ass wipe? Anyway, everything I read
was everyone's saying, oh, Amazon's being really spendy. They're spending sixty seven billion in kapex a quarter. Oh no, and then the total capex of sixty nine billion. But everything from Amazon was like, we're building something sustainable, we see the revenue potential behind it. It was all very boring. Now with AWS, everything they talk about with large language models, it sounds like they've all been huffing ayawasco or something. They sound insane they.
Do, which tells me one of two things is true or honestly, this is the real world. Unlike on Twitter, two complex competing things can in fact be true at the same interest time. One of them is they're still doing a lot of that boring stuff, but the publicity all accrues to the things that are far flung. This is also two hundred billion dollars across the entirety of Amazon that includes distribution centers, that includes Low Earth Orbits
and basically Amazon Basic Starlink. I think they're calling it Leo now. It includes yes, their AI nonsense, their factory boondoggles, their robotic stuff, office buildouts, et cetera, et cetera, et cetera. They don't give clear breakdowns on how much of that is AI services, and frankly, I wouldn't trust them if they did not, since I saw a job ad a few years back on Amazon dot Jobs saying experience with other AI services like S three storage service, that's not AI.
You can use it to feed AI. But my hairanthropic anthropics one of the largest S three customers I've heard.
I've heard that. I've heard from seven different companies so far that they are the largest customer really three. Yeah, that is funny how that tends to work out.
God, I miss I miss working in cloud software so much, where everyone is the everyone's the exclusive partner, everyone's the number one partner, and everyone's the larger.
It's so but the qualifiers get They're like, we're the largest S three customers in Australia, great in the Northeast, in the Northeast region between the hours of six am and twelve am.
Okay, So here's the thing with the capex. I'm sure a lot of it is for is for AI, just because if you look at their previously as of Capex, it's like four A fifty sixty Yeah, four A sixty one sixty three fifty two eighty three one hundred and thirty one two hundred billion.
Gerald Over at Platform and Nomics has been tracking capec span to the hyperscalers for a long time now, and he was one of the first people to call out the cloud pretenders IBM and Oracle specifically, because they're making all the cloud noises, but they're not spending anything on Capex and you kind of have to build the data centers for this to work. Not to worry, Oracle's taking
care of that. What I want to know, my question is, when you start, okay, we're going to spend even assume it's all AI whatever, fine, two hundred billion dollars on AI spent, how much of that is just in the form of a check written to Nvidio.
And that's the question because from my understanding, my understanding the I should bring up the chart. I'm just gonna do some live on air. I just bringing up an image and of course can't find it. Nevertheless, there is an image I have the Amazon doesn't. Well, Amazon builds its own the trainium GPUs. And do they still use inferentia or.
They That sounds like that that my grandpa got diagnosed with.
Yeah, but do they liked inferentia? I thought was there?
Oh, they talk about it a fair bit. They talk about trainium. And the thing is is what you don't see is people using it in the wild. I looked at a bunch of stuff with OLAMA. That's the run your own stuff there. I searched about a year ago for the term inferentia. It showed up once in a pull request that had been closed after sixty days because no one responded to it. No one is using this
in the real world. A year ago year in the month now they had with tranium two I think it was, they had speakers from Anthropic on keynote stage talking about it, which, yeah, okay, they invested how many tens of billions into Anthropic? Yeah, I'm sure there's a contract requirement around to sending it executive talk on stage. And the other one was Apple. Apple doesn't admit what they're doing to their own employees,
to each other, let alone publicly. So yeah, that looked about as natural as an oral bowel movement.
Nice, but those are.
The only two companies I've seen doing anything with it.
So based on this chart I've got, which involves various stars and company names, it looks like Amazon gets their GPUs through fox Con also known as On High Precision Corporation Limited. I love Chutneyese and Taiwanese names. By theme is almost certainly in there. No, TSMC isn't because builds the chips.
These are the training ships themselves, because they have they bought and do the design, but they don't do the fab.
So they ship the I assume that Anna Perna, which is the internal place that Amazon makes its chips, ships them. It looks like the fox Con, Quantum Computing and Jabil, and then they put them in the service. So really it's them cutting the checks to them. I'm going to be watching the monthly earnings of those companies in Taiwan quite viciously, because here's the thing. If it's all two hundred billion dollars of capex, if they really think they're
going to get paid that much, but what happens. Has Amazon ever faced the situation where they did and overbuild Has that ever happened to them? What did they do.
Yes, they made a bunch of noises about it. During the pandemic. They overbuilt their distribution centers.
Ah, but that's people buy stuff exactly.
That was my position on it. It feels like, Okay, they slowed it down and they didn't shut them down, but they waited for people to grow into it. The thing is is a warehouse that you use to ship out underpants. Great, that has the same utility more or less five years from now as it does. Yes, exactly. A lot of these chips are depreciating assets. M like, oh man, I can't wait to get that seven year old computer in here so I can do some real work, says no one ever.
Well, also training them especially they've got well they are on generation two three.
Now they're now shipping three, right, it's time to pre announce four. And then a desperate attempt to Osbourne Computer themselves.
Yeah, explain that reference.
Oh.
Osbourne Computer was a company back in the eighties.
There.
They launched the Osbourne one and their CEO went out and did a whole press joke and he's like, yeah, this is great, but Osbourne two is gonna blow the crap out of it. Oh, luck in video, and everyone's like, great, we're not gonna buy the Osbourne one. Then the company went under and never shipped the Osbourne two.
Well okay, not like in video. Then this is the thing surely that sets Amazon up for a big problem with those tradium chips success. Like even if in some wid well the Trainium four or five was amazing, I don't think it will be. But just saying doesn't that mean they're gonna be sitting on a bunch of obsolete Trainium It just feels very questionable. You know.
You can say a lot about open Ai, and you do. But that dress release where they announced that thirty eight billion dollar contract, they didn't mention Trainium.
They didn't, did they?
Which tells me that if it were half as good at this at the model training as they say it is, open Ai would be all about it. Yeah, I am not seeing that to be true.
Well, Clammy, Sam Moltman, he's he signed a thing with Sarah Brusber, Sarah Brush.
Is there any one that Sam Mortman has not signed deals with?
Ah, let's think Grock. I haven't signed one with Grok yet?
Okay, good?
Good? I mean actually sorry, just to be clear, hasn't signed with either, well, I guess it would. He wouldn't sign one with GROC as technically Grok is a competitor, I mean g r o Q. The annoying fast inference At this point, my hated dom and people are going to say this episode was two technical and the answer
is calm down, this is my podcast. I'm just I'm wondering at what point they they say uncle, because okay, two hundred billion dollars and let's say one hundred and twenty five billion of that is AI, right, AI chips takes four or five years probably to install all of that to what's end? What happens in forfit? Are they gonna make? Well?
I mean it would be That's what no one is able to explain to me. What use cases today do are we looking at? Where? Wow, if we had ten times as much inference as we do today, then X would be possible solve for X. I'm not hearing it.
No, I'm not either. I'm not even being my usual hate you self, even though this is hate a season. It's I genuinely can't get the answer of anyone, even the boosters. It's like there's this insatiable demand for compute.
We're gonna summon God through Jason is step one, and step two is we're gonna ask God what to do.
I mean, that's the thing I do think that they I think that have you seen the revenue projections for these companies? Have you ever seen like Oracle's cash flow? So if you see this that they.
Put out, also known as there's a reason that every time Larry Ellison, co founder of Oracle, goes and gives a keynote or talk somewhere, they preface it with a disclosure that is the lawyer version of everything you are about to hear is a fanciful imagining of reality.
Don't listen to him. He's he's at a long day. He hasn't taken his nap. Fucking if hr Guy Gooz,
Jerry Stiller. He The thing is, you look at these cash flow positive these cash flow diagrams for Oranthropic, open AI, and Oracle, they all follow the same thing, which is terrible cash flow negative cash flow for years and years and years, and then in twenty twenty nine everything changes, number go up, and I at this point genuinely think that they think they're going to invent Agi in twenty twenty nine, that that's the only plan they have.
Because they're not doing the things that they're not if they've genuinely believed that they would not be doing these things. We think that we are three years away from launching AGI that will change everything. So what are we gonna do. We're gonna find a way to put ads into chat jibbity. But that's not the play when you are when you think you're on the verge of the singularity.
I mean, nobody was demes habit. I don't know. I don't like him to say should extent I refused learn his name. The deep mind bloke. I try not to.
I try not to anthropomorphize either AI or its creators.
Yeah, deepest mean beepus. He's the one that every few weeks does an interview and he's like, and then God is going to come out. I'm really scared of what the computer will do. He doesn't sound like that he wishes And it's just you make it sound exciting, Yeah, exactly. It sounds fun and interesting and comical when these people
are just boring, boring people, boring motherfuckers. Bom. It's just these massive overpromises and these weird fucking like these weird things where I like, look in let's look at this. There are six hundred and fifty billion dollars in CAPEX being spent this year just by Alphabet, Amazon, Meta and Microsoft. Where is it going? Genuinely? Actually simple question. I actually don't know where it's going anymore. It's into a giant hole in their balance sheet. At some point.
You'll notice that all of the hyperscalers, all of them that have succeeded, even the pretenders, have other businesses that financed this. There has never been a successful hyperscaler that launched as a hyperscaler. The closest story you've got to that, and I question whether you call them a hyperscaler is cloud flair because they start as a CDN and then just added the rest of the cloud stuff.
And let's see, I picked this stock. Is I got fifty nine billion dollar market cap? Well, they're a real company. Yeah, you wouldn't w I won't want to invest in a company that wasn't making egregious promises. But it's just Jesus, it's just so strange because even if they think AI is going to be big, surely all the money they're spending today is not going to do that, Like all of these chips aren't going to do it. The models they train today aren't going to do it.
It's like they're trying to bootstrap from thing to thing. To think, the story that concerns me is I, okay. You take a look at these companies and their subscription plans. Top end of it is two hundred bucks. Great. I pay it myself just so I can dispatch a winged monkey to build me a shit post website, which is awesome, and sometimes it works and sometimes I have to prompt it to do different things. Great, it's better at shitty front end than I am at shitty front end. Terrific.
I'm not going to spend five grand on that. Yeah, but the model of for zooms not. And we're not just talking me here, we are talking effectively everyone. It's like the unspoken message of a lot of these folks is that your boss is going to fire you and then take your salary and split it with the AI company.
Right, and also you will be able to code everything with AI versus what people love that I quote the same thing Carl Brown's saying, makes the easy things easier and the hard things horder.
Like we are building software at duck Billet, We are hiring engineers. Is this because we're stupid?
I don't believe that to be true.
Yeah, And that's kind of the thing. It's like, I don't know, people still seem to be hiring software engineers. The engineers are getting fired and replaced. AI story doesn't seem to work. And also, every time I go and read someone who claims that clawed code changed their entire life, it mostly comes down to, yeah, it took something from a few hours to an hour, and it's like, cool, great, okay, but in.
My life, what did it do? It taught me to clearly explain what I wanted.
Yeah, but also I still had to fix things. It's like if all of this money just went into creating slightly more efficient software engineers, and even then the data suggests they're less efficient. I don't know was this worth it? And Amazon certainly isn't. Amazon certainly isn't. I guess they're getting the chunk out of anthropic from this, like they can claim anthropic success. They can have them pay for trainium in project.
Platform company, where when people build interesting things on top of you, how do you how do you take credit for that without being obnoxious.
I mean, has that ever been a problem for Amazon?
Well, not that that's nothing their word about being obnoxiou, it's just they're bad at messaging.
Well, I mean they'll mention Emlierate panos Panee m is panas Penee came over from Microsoft from the Surface team to lead Amazon's Alexa Plus. I mean, now there's something obnoxious. Now there's some real generative AI bullshit. They've lost like billions on Alexa now as well? God, what does strange company Amazon's become?
They've always been peculiar. The problem is is at some point of scale, like we're a little weird, no longer carries water. It's great, you're now one of the lynch pins of the global economy. You have to explain what you're doing a bit more effectively.
Yeah, I I what do you think You've known Amazon for one hundred years? Now, what do you think happens if the AI bubble busts? What do you actually think they do? Like, what do you think are they the types to actually mothball this stuff? Would they just pretend like this never happened aspects of it?
But there are again, they are making money hand over fist on the actual nuts and bolts of cloud computing. That's not going away. They will their Stockell tumbled because they're not posting massive growth numbers, but they have a banging business of providing computers to the world. The question is, is what happens when there's a giant hole in their balance sheet because well, it turns out that hundreds of billions of dollars in contracts everyone's die and they suddenly
they don't got the money. What do we do? That's going to be a systemic problem. It's not going to just hit Amazon. It's going to hit a lot of folks. And I think that the answer everyone's sort of hoping for in that space is government bailout. No but of Amazon, though I would be so irritated.
Well, also what they bailing out? Amazon's not gonna go bank or even if Amazon had to take a massive impairment on elder Tradeum chips and it was like thirty billion dollars the shareholders had the shareholders, they'll be mad, don't get me wrong, But it's not gonna destroy Amazon, like it's not. I mean, it might destroy Anti Jesse's asshole, like it might be might send Andy Jesse and a Nye and bail into the sun. But it doesn't feel
like it will kill them. I just don't know what a bailout would be.
I am so annoyed that I had great hopes when Andy took over, but it seems like the company has just continued down the in sertification curve. They're not doing surprising things. The reason I started the Last Week in Aws newsletter in twenty seventeen was that every week they were fixing massive customer facing problems and it was hard to keep track because they were just as bad that as they are now twenty seventeen. Right now, they don't
do that anymore. It's basically inertia. I would not be able to start the newsletter and build in our readership today the way that I did back then, just because it people do not care nearly so much about the platform as they once did.
So when did you notice the shift happen?
It was first graduate than all at once, I think is probably the way to frame it. When I realized that I went from every week having the problem of there's so much stuff here, which ones make the cut, to there's so much filler here, where's anything actually worth writing about.
And when you say philo, what do you mean they used to say?
They used to talk about things like as lambda, a transformative shift in how a compute could be run. They put that out with the same enthusiasm corporate voice as there's now a third cloud front edge location in Dallas, which even people in Dallas do not give.
It to us about, right, what does LAMB that do? In effectively? You give it its code.
It runs the code when certain trigger events happen, be a web request, be it a time being hit, be it's something that it's some other event hitting. And it only charges you for while it runs. Huh and massively scales up and you don't have to worry about any of the care and feeding of the infrastructure around it, which that's really cool. It is it was cool. There's
beta in twenty fifteen. But now for the last truly great things Amazon put out that change the way I thought about how this could work.
Except now Amazon doesn't really put out fun new updates.
It's just incremental updates. We have a new thing. We've launched, our own large language model called Nova. It's not as good as the ones you currently use, but it's less money. Yeah, what's start poc stage. They don't care.
What's great is if you look when Nova is Wow, it's in the UH top twenty five below Kat's code of pro Kimmy two point five four point one memo me and the Max Quinn thinking Kimmy K two basically every other model.
Right, like exactly, it's like they're racing neck and neck with the taxidermy in Frontier AI lab that it really is.
That though, it's like it's something called yeah Ernie from Baidu, like every Chinese model appears to cry, that's so funny. What a what a horrible situation we found ourselves in.
Well, Amazon prides itself on being frugal, and it turns out that when I have to assume the reason the Titan models that were so bad they never really saw broad release in light a day was that it's okay, it's gonna cost some billion dollars to train this model. What can you do for twenty million? Like that? That is the Amazon ethos. Fail is the answer.
Unless it comes to Capex though, Like that's the thing. It feels like what you say there goes directly against how they're dealing with AI. It feels like AI is just poisoning their brains.
Oh my god. Yes, I wound up doing a so I back when these stuff all started coming out, I had fun with it. I ran a bunch of custom questions against these things, like rank the US presidents by
absorbency and then see how long long is it? We would bully them into doing it or give me some criticisms of core Quinn and one of the early chat GPT things made me question like aspects of my life and go home and drink heavily, Whereas I asked one of the early Amazon models that question, and it would have been like Exhibit A and the defamation case, except I'd have to prove that anyone took anything Amazon said even slightly seriously.
Right. It's it's just weird. There's something that I think the I don't know if AI side coast this is it, but I mentioned this with David Gerard. It's like people getting one shot at by this. It must one show executives. I think, oh, it absolutely does.
I use it myself in the curation of my newsletter for example something because this is this is a terrific use case, and I think it encapsulates my philosophy on this uh where Amazon. I consume all the RSS feeds every week and there are roughly one hundred and fifty items that come out. So what I do is I have a scoring thing that pops up two columns. Here's here's the good ones, here's the crap ones. And I have a whole rubric that I have put into this.
It saves me a tremendous amount of time. I go through and I put some from, I move some back and forth as I read down the list and increasing it learns from my decisions. It's getting it mostly right. It speeds up my workflow. It is convenient. Is this worth changing the entire world?
For? No?
No, it is not. But I'll make Hey, will the sun shines? Besides, if I'm going to make fun of something, I should be using it so I can understand it to make fun of it more effectively.
Right, But would you pay the real rates for it? Because that's what I think is going to break this as well. It's when these companies start having to charge the real costs, when they have to start getting paying the margins.
Right now, this is costing me seven cents a month. I am glad to pay that I would probably pay as high for that as I don't know, twenty bucks a month, which, again that's a significant increase there, sure, but too much beyond that.
No.
I spent seven years going and doing this by hand. I don't feel the need to stop. I did use claud code for several weeks to rebuild the entire interface because I was I had this janky, horrible newsletter publication thing that I kept meaning to. I looked into what it would take to have an engineer do it between twenty to fifty thousand dollars. This was just three weeks over the holiday break of prompting the thing, and it
finally got it dialed in and correct. Now, in practice, if neither one of those two things is a viable option, there's a bunch of stuff off the shelf these days. It didn't exist back then that I could have worked with and had a much better base. But this was fun. This was I could use some software in this small place. The cost of generating this software now from my time perspective, is minimal, and it fills in a gap. And these things are good at solving a building tools to solve
one very specific problem. Where these things fall down. Software has always fallen down on this is as soon as the requirements change, I said on a link list every Monday. I built software myself to handle this. When I started sending out blog posts with it as well, I had to go back to the drawing board and redo an awful lot because oh, it's a different model. The entire flow changes when I introduced the requirement downstream. When software
becomes free, is longer the bottleneck more or less. You can build custom tools for any given situation far more effectively. I started writing a Bash script to give someone access to just run run this one command in your term, It'll give you access to this thing. And I started writing it myself, threw it into one of these things. It came out with a much better script as a one off that I don't need again. I could I have done it myself. Yeah, about four hours, But would I have.
No It's soldings. It's also those.
Small problems or point solutions. But I'm not firing people for this. I'm not replacing segments of the economy with this I have. My brother is one of the few people I have heard about being definitely impacted by the rise of AI. Until last year he was a freelance translator. I believe that that feels like it's something that is easily prone to disruption. I buy less stock photography when I can have shit post images of giraffes on fire and data centers.
Now that they're not They're not gonna love that. The list is not gonna love that. Oh no, I don't even I think Ai is just genuinely ugly, you know, even that it would suck the life out of the out of the ship post.
And I paid for artists to do things correctly.
Okay, good by have stickers.
And I gave out a reinvent for ship posting dot Ai and it was bad Ai art in a nineteen fifty style, like a dog with an extra leg, people with extra hands and whatnot. That we hired an illustrator as a human to draw in the style of bad Ai. I thought that was a lot of fun for something like this. Yeah, I am paying artists and I'm making it work. I'm building a ship post meme to throw on Twitter and never think about it against the thing.
Though posting is an art, posting is an art. It is bringing this. There is no bushido in in generating images, but let's move on from this subject because I have one final question, Please hit me with it. So you've seen this thing about how everyone's freaking out about, and I know you're going to love this. Everyone's like, oh, hey, clawed code is replacing software everybody. How funny is it?
How ridiculous is it to you? I'm leading question. I realized the idea that instead of buying on a percy basis, people are going to just replace SaaS with their own internal share. I think it's one of the funniest and stupidest things I've ever heard.
Oh.
I think it's a terrific thing that is totally happening because obviously, again I am building software for enterprises. Well, they can just write this code overnight to do it. Yes, they can easily generate a website that purports to do something. But what if they need that thing to be correct? What if they're, oh, I don't know, signing a billion dollar contract on the result of what that software spits out exactly? Maybe Yolo slamming it isn't the right answer.
We're going to just replace all the software with custom bespoke stuff.
Great.
A terrific example I saw is that for payroll anthropic. The company uses ADP, Well why would they do that? They can clearly build their own eternally because you're not buying the software, you're buying the compliance, the keeping up with all the different jurisdictions in which you operate. You're someone who is going to certify with their reputation that
this is going to solve a problem. There are a bunch of things out there, like there's some software that will be disrupted by this picture, a bunch of the obnoxious stuff like I just want to convert this from one from one video type to a different video type, and you'll find some like random thing for ten bucks. Yeah, well, now cloud code or whatnot is going to fix that because I don't know how FFmpeg command line flags are supposed to work. It does great better, it's easier than
looking it up. Am I going to go and replace actual business? Line of business? So line is a software I wish?
But no, Well that's the thing as well. People think the people pay for like Salesforce what have you, because they're like, oh, because I have to pay for it, and also just because it's a simple piece of software. No, they're these Intrica API ratking things that roll around in their own filth, and then you they get they maintain them on the back end. Even if even the shittiest cloud software you use is has some weird back end stuff.
They have to constantly maintain because things break, and the bigger the company, the more things that break.
Yeah, and it's just the gen one of all of Salesforce is just an Excel spreadsheet. It is who am I contacting? What is the last thing I did in the rest? And it gets complex quickly, and where the leap happens is Okay, Now it's not just you. Now it's an entire sales team. Now you're going to like you have some turnover in that sales team over the course of doing business. How do you wind up having a provable chain of issue on this? Someone pops up for app for three years from now they said they
talked to someone here. We have no record of that because Claude rebuilt it five times this week. Where does that going to live? How is that going to exist?
And even then it's like audibility, auditability, even Jesus Christ, but this is.
Not an AI phenomenon. When Dropbox launched, the number one comment on hacker News was this just is just our sync with some extra bells and whistles.
This isn't a product. Yeah, yeah, well that's about yeah. I mean even then drop Boxes like, it's not just storage, is the support services around it.
I don't know believe a dropbox doesn't want to do what Dropbox does anymore every time I use it. A folder that sinks everywhere is what I loved about drops Now it's trying to basically compete with zoom in Google Docs and all the other stuff.
Stiles, well, gory. Let's wrap it up there. We've done some good hating we can find. I'll have some good links. View will link to duc Bill, link to last week and Amazon was it.
Last week at aws dot com with an obnoxious, sarcastic platypus. That's one my use of AI before we go that I think you'll appreciate. If people email me Corey dot Quinn at DUCKBILLHQ dot com and I don't like what you're doing and you're trying to sell me something. I now have an AI powered executive assistant in the form of Billy the platypus, who will tell you in corporate appropriate ways to go fuck yourself, which is just chef's kiss, because I can't say that to people. I have a reputation.
The obnoxious platypus responding to so salespeople does not have this constraint. This is the one liability shift I have found that actually works. And you can blame AI for things you personally approved.
Well, I don't endorse the AI platypus. You should get a real one. You should get a real one.
Danger that customs officials had many questions I can not wish to answer.
Train the platypus Cory. All right, you've been listening to Better Offline. I'm at Zidra and you have a monologue this week. I guess. Thank you so much for listening.
Everyone, thank you for listening to Better Offline. The editor and composer of the Better Offline theme song is Matasowski. You can check out more of his music and audio projects at Matasowski dot com, M A T T O S O W s ki dot com. You can email me at easy at Better offline dot com, or visit Better Offline dot com to find more podcast links and of course, my newsletter.
I also really recommend you go to chat dot Where's youreaed dot to visit the discord and go to our slash Better Offline to check out I'll Reddit. Thank you so much for listening.
Better Offline is a production of cool Zone Media.
For more from cool Zone Media, visit our website coolzonmedia dot com, or check us out on the iHeartRadio app, Apple Podcasts, or wherever you get your podcasts.