Welcome And Monthly News Format

Tim 0:06

Hey everybody and welcome back to another episode of the Cables2Clouds podcast Fortnightly News. That's quite a quite a mouthful. And uh so we're covering stories that we thought were interesting and came up in the last month. So this is October, so these are gonna be mostly stories from uh September. And uh that's that's how we do it. And with me is my co-host, Katherine McNamara. Say hi, Katherine.

Katherine 0:37

Hi.

Tim 0:39

And uh actually she's gonna launch us off today. So uh yeah, she's got some uh really interesting story about California where she lives.

Katherine 0:47

Well,

California AI Laws For Workers

Katherine 0:48

there's a couple stories, but uh basically they went on a flurry of uh bill signing for laws. Uh some of it's interesting, and some of it is just uh I'll spare my opinion. But um so the fur the first uh there was a group of bills that basically are aiming at protecting uh workers' rights uh you know against basically the AI boom and stuff like that. I'll leave it up to the audience to decide whether this is enough or not. But you know, basically uh a few of the uh legislations that were uh laws that were passed and signed into law, like um a couple of them are really uh good, I think. Uh for example, one of the worries is HR, uh AI uh powered HR. So some of the laws that were passed were um things that would require uh human decisions and and review in any AI uh, you know, AI generated disciplinary action or termination decision for employees. Um when people, when companies decide to lay off people, they have to actually be transparent on whether it's not or not it's you know caused because of AI. In my opinion, um I feel like a lot of companies are pretty honest about this with Wall Street, but now they have to fill out fill it out in the form that they submit to the government prior to layoffs. Um one really cool one is protecting employee uh privacy, basically banning workplace surveillance tools by employers in bathrooms uh located in the workplace. You may ask, what do you mean? Like they're not setting up cameras, right? No, they're probably not setting up cameras, but think about it this way like uh Wi-Fi telemetry. A lot of companies monitor your cell phone locations and stuff like that. And um we've heard horror stories about companies like writing people up because their cell phone's in the toilet too long. Um public ed uh uh you know, higher education has to provide AI training for uh and procurement training for students, faculty, and staff now. Uh one of the things that have been hit hard is the legal system uh with AI slop, um, both from a um you know, from people trying to be pro se, but also lawyers just basically submitting briefs and uh uh and uh submitting uh documents basically that are AI generated, not reviewing it. So now uh they codified it in California, basically saying that if you're a lawyer or somebody submitting something to the court, you're responsible for anything you submit. If it was uh, you know, if you're providing legal judgments, briefs, whatever else, you're responsible. That kind of already was in play uh already. Like we've seen sanctions being handed out by judges for people submitting like, you know, briefs and stuff with fake references, but this just kind of codifies it further. Um as far, you know, in medical care, they've also some uh created some legislation in California regarding that. So um doctors and licensed providers can use their own professional judgments when AI or like other clinical tools are used in patient care. They don't have to use that, they don't have to use that uh, you know, decide uh basically trust the decision of AI. They can make their own decision if they're a licensed professional. Um and uh, you know, uh it also codifies that developers of clinical decision tools have to take reasonable steps to reduce their known or predicted bias uh predictable bias and how those systems are used. Some other things that combat deep fakes as well, uh uh basically prohibiting the uh deletion of digital watermarks. So, you know, basically allowing access to metadata and other things. I don't know how easily they're gonna be able to enforce that one, but uh, you know, if it's involved in a deep fake in a crime and deep fakes, that would be easy. But outside of that, I don't know how you're gonna, you know, arrest someone for using, you know, wiping exif data for some like silly picture you made. Um and uh, you know, they're um they're making they have to also make clear like local government meetings uh have to be from a uh uh must come from like uh real people, not AI bots. So people have to be clear in like public notes and stuff like that for uh public institutions that these are AI generated notes or not. Um and uh they also updated the laws to protect against false impersonation and deep fakes. Um that's that's a big thing, that's a big thing as well. Um I still think it was pretty uh uh, you know, like they they did have some uh uh laws in place and framework in place to uh prevent against like fraud and and identity theft, but this just strengthens it by adding language that helps against deep fakes. And then

Data Center Bills And Wildfire Risk

Katherine 5:17

uh there's some uh another uh, you know, some other news in California is that California basically signed seven data center bills that uh, you know, kind of clamping down on the data center building uh to some degree. So for those of you who don't live in California, but probably look at the news. Um, California has a lot of wildfires. We have a very and and you may ask why. I mean, yes, we have deserty portions of California that are naturally deserts and they can be very dry climate. But one of the bigger problems and why a lot of these wild wildfires start is because our power grids are very, very aged. So a lot of it is just the power grids like sparking in the summer and causing a fire, and it turns into a huge fire. So um, and unfortunately, some of the privatized entities like uh Southern California Edison and others don't want to, you know, it's it's really burdensome to pay pay to upgrade that. So we kind of am used to fires at this point, sadly. Um so adding a bunch of data centers onto those power grids would be pretty disastrous in a dry climate. So um some of the bills aim to kind of circumvent that. So if uh if if uh um if folks wanna like uh companies want to build data centers, they have to actually now, according to law, um, have to help uh pay for the upgrades to those grids uh and uh help pay, you know, offset those costs so we don't get more wildfires and other things. They also have to uh provide you know the you know, we also have updated in the in these laws that were signed, water and efficiency disclosure rules. So um they have to basically disclose like how much water, power, and other things they're you're they're using and how they're gonna make that a little bit more efficient. It's basically providing transparency on that angle and you know, some due diligence to pri uh to try to uh basically uh make sure it's efficient and not just draining all the water and power around them. But I do think that the grid upgrade part is uh uh is uh a really good part of that. Um you know, honestly, that's that's uh that requiring the data centers to help pay for those grid upgrades is something that will probably discourage some of them from opening in California. But to be fair, rather have them not open California and us have less wildfires than you know, or them to say, okay, if we're gonna open it there, we're gonna at least pay for some of the costs so we're not causing more wildfires.

Tim 7:48

Yeah. The it's it's honestly it's so far past time that uh you know these people who build stuff start paying the costs of actually you know being a burden to their locations. Uh I think. So, you know, as an example, you can't just buy, you know, uh up until now you just the companies would just buy a plot of land, you know, pay to build the pay to build the data center there, right? So so very temporary jobs, usually from out of state, would come in, and then they would just basically become a burden on the location on which they were built, you know, for water, for power. Um to say this is overdue is you know, burying the lead a little bit. Like that it's it's well overdue. Uh it's i I I will say I'm not surprised that California is probably the first one that stepped up to actually codify this in some way, uh, just because you know they they kind of have a history of this. And of course, like you said, the power grid is so ancient out there um anyway, and most of the wildfires are caused by by that, by exactly that. You know, I can't rem remember how many articles I've read where you know the wildfire started because of you know transformer blowing or something falling off of the one of the um poles or or whatever, something like that, right? So yeah, this is yeah.

Katherine 9:14

I'm not a hundred percent sure on what the percentage is, but a large number of our wildfires are like are are because of the aging power grid. Um and maybe this you know, I don't know what the power grid allocations and and climate is like in every other state. Maybe there are other states that don't really require that. Um maybe they have X, you know, uh more up-to-date infrastructure or more power uh yeah and capacity. But in California, this is absolutely needed, or else it would cause it it would cause an enormous amount of problems if if uh and real cost of human life and property if they could just build wherever they want and you know, F the infrastructure.

Tim 10:00

I mean it's a even yes, of course, there are definitely other areas of the country that have better infrastructure, but I mean, even in Northern Virginia, like where they've built, you know, that has like the most, I think it's Virginia, Texas, and Flor Florida? I forget what the third uh state is, but they have the most uh California actually is the third one, sorry. Uh California, it's Virginia, Flor it's Virginia, Texas, and uh California have the most data centers of like the whole country. And in all cases, I mean let's not talk about Texas's power infrastructure. You know, next time it snows, we'll have to, you know, read another story about that. Virginia, I guess, probably is the best in the best situation of the three, but it also I think has just an insane amount of infrastructure there already. And I've seen uh articles from Virginia. I have a you know family that lives there, so friends that live there that talk about it. Uh, you know, what is it, Governor Spanberger? I think just put out new new tax stuff uh associated with uh you know energy grid stuff, you know. So it's somebody's gotta pay for all that infrastructure. It's so why not? Why shouldn't it be the companies that are gonna be taking the lion's share of that of that power and um essentially you know subsidize using taxpayers to subsidize it? So yeah, I'm 100% on board with uh with this. And you know, this isn't the only thing, right? This is just one, this is just California doing it. I think we're we have seen and are gonna see more of this uh in other states as well. But generally, we're also seeing just a just a big pushback against data center building in general. Like some, some just at the community level. We've heard we've I think we've seen many stories by now. Some some of we we even you know have reported in the last few months about communities standing up to uh data center builds build out uh companies. I think I just read an article this morning about something, maybe Penn I think it was at somewhere in Pennsylvania or something. There's a data center company is trying to pay every uh person who lives in this town where they want to b build a data center, $10,000 to like let them build the data, you know, to let them let the city council like let them come build a data center. Personally, I don't think $10,000 is enough to have a data center in my backyard. I don't I don't know what the number would be, but it wouldn't be $10,000. Um yeah.

Katherine 12:19

Yeah, uh yeah, that could be, you know, it really depends on how close your house is to it and stuff like that. But w some of the things that we've heard is, you know, that there's white noise or noise that c uh that emanates from the city. Yeah, that uh that you know, i i depending on how much property value is there, it could significantly drop the property value with the noise, the the uh additional cost of electricity and water, other things like that. There's it's ten thousand dollars, you're right. Like it, you know, if you're if your pro you know, your real estate property goes down by you know one third half, it could be ten thousand dollars could be not not even touch that.

Tim 12:59

Right. Because I mean the think about it, even if the property value itself dropped, like who's gonna who's who's buying that house next to a data center? Like just forget about the property value itself. Like, let's assume you went, oh well, we went from a $300,000 house to a $250,000 house. Who's looking to pay $250k to live next to a data center? Like nobody, right? Like it's not, it's so it's the point is it goes beyond liter the literal loss of of of value of the property into into can I even sell this place if I wanted to, you know.

Katherine 13:31

I mean, there's a reason why they're not setting these up in the back of like in the backyard of billionaires.

Tim 13:37

Yeah, right. Exactly. So anyway, yeah.

Pushback Stalls Data Center Projects

Tim 13:40

So um, so we have a story, uh, let's see, uh US data center speculation is curbed, $170 billion worth of uh investment is stalled. So utilities and regulators are adding vetting requirements, large load of tariffs to filter speculative data center projects. So under over $170 billion of AI data center capacity has been blocked, withdrawn or stalled since 2020, so since like 2024. So but that's against like almost $600 billion worth of planned spend. So so I mean that's a significant amount of block, basically. So the um this R.

Katherine 14:19

That's like one sixth or one fifth.

Tim 14:22

Yeah, I mean it's a it's a significant large amount, right? So um the utility piece, the vetting, like just the extra moratorium stuff. Like North Carolina just put moratorium, like the just signed in uh moratorium on data centers, for example. Like there's a lot of of blocking that's coming up uh from communities and states uh on this. And it's not gonna be one I think that just writing a ten dollar check is gonna solve. Um so you know, is this the question though really is is is this is this ultimately good in a way that only the largest players are gonna be aboard afford are gonna be able to overcome these hurdles and then they're gonna be the ones that are gonna have the biggest legal uh weight to throw around? Or or you know, I guess is that good? Because it what the un I won't say unintended, but like the side effect is that we draw drop out other competitors that can't afford to essentially spend against this kind of uh these kind of blocks that are being put up. So I don't know I don't know if that ends up being a good thing or not, to be honest with you. I think just less data centers in general is probably the right answer, uh, regardless of who's paying for them.

Katherine 15:37

But um, I also feel like it gives some of the bigger like AI labs and stuff a doubt. Um, you know, like I I heard that like uh yeah, I I remember like we did a news story a while ago about how OpenAI was like pulling out of some of its like uh some of its uh committed like RAM purchases and stuff, uh like uh like memory purchases and stuff like that. And this kind of gives the like you know, like given that some of those larger AI labs, uh Frontier A labs are having a real problem with like uh still turning a profit and they're trying to focus on um going public, this might just give them the opportunity to to pull out of some of the things they already wanted to pull out too. So it may not just be like, oh, only the big AI labs can uh afford it. It might be actually an easy out for them to kind of pull back on some of their uh some of their predictions. Some of their commitments.

Tim 16:31

Yeah, some of their commitments and predictions. Commitments, sorry. Yeah. Yeah, yeah, yeah. No, I mean both, both are true. Um actually it's interesting because this next article goes right into that, right? So

The Six Trillion Revenue Reality

Tim 16:41

Bain Capital is estimating that basically in order to pay for what's what's committed and and planned from a build-out perspective, the whole, I guess the AI industry, whatever you would call it, like the whatever that looks like, whoever's using these data centers needs to make about six trillion dollars in annual revenue to pay off, to pay for these investments. And that's that's not one company, right? That's just the industry as a whole.

Katherine 17:06

The whole industry, yeah.

Tim 17:07

Has to make six not not the whole tech industry, right? This is the AI data center, whatever industry itself.

Katherine 17:14

Yeah, like the AI labs, the basic AI labs, like uh like anthropology.

Tim 17:18

And all the downs. Yep, exactly. They all have to make a coal mined total of six trillion dollars in revenue. Not that's revenue. Like that's that's crazy. Like, I just think that's nuts.

Katherine 17:30

Yeah, um, I I do think that we're like so. It was interesting. I was talking to someone about the profitability of AI a while ago. Um, and you know, like the actual AI labs are not yet profitable. Like, let's let's just again like focus on that. They're not yet profitable. There are companies that are utilizing their services that are probably that are getting a profit from this, but not currently the actual company. Yeah, the AI uh labs. At some point, they're gonna have to raise rates and and other things in order to uh in order to basically make that that six trillion number to basically justify all this spending. They're either or they're gonna have to cut some of their capex, like and and and and a lot. Basically, like uh yeah, by a lot. Um and I suspect it'll and I was not this is opinion, but I suspect it'll be a balance of those two things. They'll probably cut some of their capex, but they'll also probably will have to raise rates, especially if they want to make a lot of money when uh they go IPO.

Tim 18:32

Yeah, I mean, and this is this is Bain Capital uh that is kind of figuring this number out, right? So it's not these are these these are the you know Bain Capital loans money, right? This these are the the the the investors essentially saying like, hey, uh this is gonna have to take a shit ton of money to uh to be to for us to get our money back and for everybody to get their money, basically, for this whole thing to work. Uh so yeah, it's gonna have to be both, right? They're gonna have to I am waiting, I I have I keep hearing about more and more efficient, you know, AI inferencing and and and just generally how they're trying to make the models more um efficient without without sacrificing improvement in reasoning and and all of those things. I I still don't think we're there, and I'm curious how big the where the ceiling is on it, or they could break through and and all of a sudden this whole thing um you know, does like a a Moore's law kind of reversal thing where I don't know, where where things get exponentially cheaper because of some technological breakthrough. I don't know. The I know we're not there yet. So I'm curious to see um how this one break uh plays out.

Katherine 19:41

Yeah, I'm not a hundred percent sure, but I I do think that w you're you're right, we're not there yet. But each model like that comes out that like you know, up upgrade from the previous model is more efficient, um, is you know faster and and and getting better. So there but we are very far away from uh from being able to just ask. Yeah. Yeah. Um like like look at Astra, like they were able to get like hallucinations way down to like I think it was like 2% out of scope things down to like almost 0%. So there's they are improving models, but um is it enough to offset all of this? I don't know. And and eventually, no matter what, they're gonna have to raise rates because they can't afford it. Like even the most efficient system, they still can't afford the like OpEx and CapEx for what it's costing them.

Tim 20:32

Yeah, my $20 subscription is costing them, you know, five thousand dollars or something every month or something. It's ridiculous.

Katherine 20:39

Yeah, pretty much right now they're the drug dealer just giving you the taste to get you hooked.

Tim 20:44

Yep. They will they will raise the rates and we'll be and I'm actually very interested to see what happens at that point. Like I said, I my my prediction is that what we'll see is kind of like what would happen with cloud, where you'll see a a lot of repatriation, meaning like a lot of coming back, you know, taking the a what we gave to AI and doing some and doing it ourselves instead for cheaper. Um button.

Katherine 21:09

I'd really like to have stable code for cheaper at this point.

Tim 21:13

Stable code.

Katherine 21:15

Yeah. Um sadly, uh you know, it's a whole other discussion. But you know, like unfortunately, the industry's kind of gotten like uh with the rise of vibe coding, we've kind of accepted like patchy unsecured code, and it's just kind of like we have to be okay with uh uh with um more errors, and it's like, wait, I don't I I don't want more errors. I don't want to be okay with that. Um, but I I digress. Um I you know, we already have companies that are complaining about how vibe coding is costing more than the actual like developers. It just does it faster for them than developers do, maybe. But like the the whole like yearly salary of their whole like you know, of their their developer staff is like was cheaper than uh than vibe coding. And um yeah, it'll it'll be interesting because this is nowhere near their final form on what the rates will end up having to settle at for them to make a profit.

Tim 22:11

Yeah. Yeah.

Vibe Coding And Slop Grenades

Tim 22:12

Well, and I was I wish I was reading an article today uh about slop grenades, which is what the Shopify CEO calls uh coined after he he first started by saying like in order to not fire people, you have to prove me that the the job can't be done by AI. And then a year later basically he's like okay yeah we're wasting so much productivity because all these workers are creating slop and hand and throwing it over the fence to the other to their coworkers to consume which is causing their coworkers to like spend more time than they would have consuming the slop grenade and then you know like apparently this is like a whole thing. They're actually losing productivity because of AI because like the worker will just create something you know here's my AI made this go consume it. And have you ever had this happen to you where a coworker gives you something that they basically slopped with with AI and like hey I need your comment on this or I need you to edit this for me or I need you to do that.

Katherine 23:08

No, that's never happened at all to me at all.

Tim 23:11

No. Right. So I mean it sucks right like then you're spending way more time than you would have and then all they've done is taken off their plate and given it to you. So now you caught the grenade basically and you get to spend all the time unslopping it. So this is a thing that happens um so that's the other piece right does is the did the product activity actually get better? I think the jury is still out on that one. But

Chips Cannot Fill Planned Data Centers

Tim 23:36

anyway uh all right so let's let's let's move on uh the next thing is let's see oh yeah so on top of pushback against AI what we're actually finding you know kind of coming from the other side of this is that our AI data centers are hitting physical limitations.

Katherine 23:55

Right? You mean physics still apply?

Tim 23:58

Yep today today at least until AI figures out a way to vibe code itself a way for physics not to apply uh they currently do still apply. So you can build all the data centers you want but you still have to fill them with gear that doesn't exist. So um this article from the register is talking about basically that America's planning more AI data centers than chips can fill. Like there's no you can build them all day but like they'll be empty husks until you start filling them with racks of gear but those gear does not exist and will not exist for quite a while. So satellite imagery is putting ADI data center deployment at about 11 gigawatts uh that was in 2025 so we're really looking at more like 16, 18 now. Practical capacity capped in the low 20 watt 20s of gigawatts by 2027 and then forecast it on from there. Doesn't matter the big deal is that they can build these data centers and run them all day but like they cannot actually produce the silicon fast enough in Taiwan via the fabs to fill these data centers. So you're creating empty buildings basically you're kind of China did this but they did it with housing back in like I don't know 10 years ago or 15 years ago now where they just created like they figured the way to solve housing problems was just to build a lot of housing um but then a lot of these houses were built using government spending and then you know did sat empty essentially because people couldn't afford it or didn't want to move there or whatever it was. So you just had all these cities that were basically empty from all the but all the work was done. All that all that wonderful economic activity happened and uh you know these buildings are created and then nothing. So I guess we're good we're gonna do that here in the US but with data centers instead.

Katherine 25:42

I suspect that they're probably thinking that they'll figure something out or they'll find a way around it or they've got so many uh back orders that it'll Micron or whoever else will take care of it.

Tim 25:53

I mean but well I would think that way if it was one governing body but really you've it's it's all these companies racing to do to get their piece, right? So I think that's more likely what we're seeing is that you know however many data center companies are trying to get their contracts and trying to build their data centers.

Katherine 26:11

Aaron Ross Powell I'm sure the big AI labs though think that they since they're a big customer they probably have like first pick as it's the smaller companies though that will probably suffer more. Oh yeah no doubt they'll they'll build it and then they'll go out of business because there's no like they're not gonna get paid basically they're gonna get pulled out or yeah they're they're in line for resources behind some larger AI labs. Like like if you're a little company like try you know that built a data center for your own like uh your uh for your own uh project uh or for your own AI company and you're a small company and then you're competing with open AI and anthropic for gear I mean you're yeah good luck right it's the same as if you're not you're not the hyperscalers yeah so I mean and this this is actually interesting because uh we've been thinking that one of the ways we're gonna get around frontier you know frontier paying frontier costs is that we can repatriate or or bring our AI on-prem but I mean there's gonna be a shortage problem there as well if you don't already have it like you and the the big AI labs are trying to get open models um uh regulated and and banned so you have to go through them. That's basically you'd have to have a license for the government to run your own AI uh and it definitely can't be open uh open uh source AI.

Tim 27:32

Yep yeah so I mean there's that too right the frontier models want to own the whole pro the thing new soup to nuts they don't want competit essentially competition uh from free free models or or open source models rather um so there's that right so we'll just say AI is gonna kill us all uh so that they please regulate us we can't do it ourselves for some reason so yeah there's a that's a whole thing we covered last time as well um yeah all right let's uh this actually segues into the the memory shortage no less by the way like I was gonna say like you know not only are we dealing with silicon requirement you know Micron has been comp you know complaining that uh you know the shortages in RAM and memory are gonna continue on for uh you know and it get w even worse in 2027 to 2028.

RAM Shortages And The Cloud Push

Katherine 28:20

So if you have a computer or a laptop sitting around with a lot of RAM in it hold on to that bad boy because you're probably not gonna be uh getting a new beefy computer for a while um with that much RAM. Um just hold on to it I'm holding on to my nice little like system 76 I got before this all blew up 96 gigs of RAM. I'm rich.

Tim 28:39

I wish I'd I wish I'd gotten one at that point. Uh but who knew who knew at that point I thought I had time.

Katherine 28:44

So oh well I actually uh got my spouse a like uh like right before all this blew up I I got uh her like a brand new like laptop like MacBook for her work with like 128 gigs of RAM and stuff like that and I'm like you know I'm like wow that that would have been that would have been like 10 grand now.

Tim 29:07

Yeah absolutely insane now yeah and that's another one like how are we gonna deal with that because uh you know that's that's not just memory for you know data centers that's the same memory that everybody needs for their business for their personal like yeah it's it's everything.

Katherine 29:24

Well how how are we gonna deal with it? Unfortunately there's a lot of big corporations like Bezos and like even Google that are pushing for like let's do cloud computing like let's just have you set up with like a VDI or something which is really going to upset the gamers because a lot of the gamers are like you know they buy big beefy machines so they can play their the latest game and they don't want to deal and not deal with latency or like graphical issues. And uh yeah like you're competing with those big data centers or big uh companies with uh for like GPUs and stuff.

Tim 29:57

Yeah it's it's insane.

Katherine 29:58

I don't I don't know what to say about it but that's I'm I'm glad that I'm not a big gamer at this point in my life. Not that I don't love games but I just don't have enough time to do it.

Tim 30:09

Yeah I am but I I bought my computer right before this whole thing started. So I think 2024 is when I bought my computer so it's still it's still doing alright. I'm just gonna keep it for I think I have six I forget how much I have it 6496 gigs of RAM in this one. I I bought it right before things blew up and I got lucky on that one because I didn't it's not like a new right I just that happened to be when I needed a new computer.

Katherine 30:33

Um I got lucky my my nice 76 uh seven system 76 machine like with like that's like got ton of SSD space and like a a RAM was only like $2,500. Like that would have been god awful expensive right now. I went back and like looked at their site and like put the same like it was double the price.

Tim 30:53

It was just like I believe it it's absolutely insane. So yeah good luck to everybody may may the odds ever be in your favor. Alright so we got a couple off the wall uh stories and then we'll wrap it up. Um so there's one that uh so here here's one that's interesting. Uh it's got just gonna do it the not lightning round but not not spend as much time on these.

Google’s Orbital Data Center Satellite

Tim 31:19

So Google launched its first orbital data center satellite. So this is something like Elon and and then multiple people have been talking about this idea of essentially moving edge computing to the far, far edge of space. So it launched its first suncatcher prototype to test TPUs in orbit. So the idea is we pack a a satellite with essentially CPUs, hard drives, etc just for edge computing and launch it into orbit where we can uh do edge compute I mean there's not much to say here like if you're using satellite that's definitely edge. Yeah that's edge gets that's the serious edge computing um it does have to get cheaper though uh it says a 200 200 per kilogram threshold is what makes it roughly comparable to terrestrial uh deployment so I don't actually know what it is right now but it's it's higher than that so definitely the costs have to be solved come down a lot before that really becomes a thing.

Katherine 32:19

I would also think that like uh I don't see it replacing all terrestrial data centers just because of latency latency issues and stuff like that. Like could you imagine your VoIP going to space and having it come back down?

Tim 32:32

No no no not at all.

Katherine 32:34

It'd probably feel like charter cable or something if you were a customer you you'll you'll know what what I mean. No, I mean it's an interesting concept it's cool like uh I don't have a lot of personal thoughts about that except like you know I hope it could be cheaper but obviously it's gonna be for certain use cases. It it wouldn't be great for like like VoIP for example and and certain like latency sensitive sense yeah latency sensitive applications but it you know it's cool.

Tim 33:04

Well I like the idea of edge compute because I mean if in theory the the the idea would be that you know hey we have data going going to the satellites so we could do work on that data there at the satellite edge. Remember we talked with uh Tom Coonath and he we he was talking about he was on here talking about low earth the LEO constellations and stuff right so this is this is right in line with with that idea of of a constellation of like you know compute and networking so it's an interesting idea.

Katherine 33:34

Next one I think you should probably cover the Stanford okay yeah yeah so all right no worries.

Tim 33:40

So this is a really interesting idea.

A New Network Protocol For AI

Tim 33:42

There's a Stanford uh professor who basically is saying essentially that TCP doesn't work for AI. It's it's too slow. And I think that's probably to some degree accurate um because of the way that TCB works with the SINS and acts you know like the the the transactional nature uh which you know ends up being reliable um but reliable but chatty. Yeah reliable but yeah but chatty and also just you know because of recovery and all of this it it ends up being really slow comparatively right the that's why we don't do VoIP with uh TCP we use it with UDP because you just slit you just throw packets all day and very quickly and the idea is that it's the same with AI inferencing or just AI workloads in general that there's so much data moving at such speed that something like TCP just doesn't meet the need. This is what he's saying. I don't know if that's true or not. Again I don't I'm not an AI uh you know data center professional or anything but he's saying basically TCP is too slow. And he's promoting a new protocol called HOMA HOMA which is I don't know anything about it. But in this case basically he's saying that networks should it's a clean slate rethink of how a network should be managing the traffic congestion. Right? So he's working on this protocol uh and literally this he's he's saying this is a live mission his life's mission is creating this protocol. So um it's he says it's message based rather than stream based. I don't quite understand what that means. I mean it ultimately well okay so the article mentions it's like an RPC like a remote remote procedure call. So it's kind of an explicit messaging protocol. So he says that uh unlike TCP it designates the receiver to manage congestion control. Okay. I although doesn't the receiver even in TCP manage congestion control by yeah by managing windowing? So I'm kind of curious what the means by that the first packet the receiver gets has information on how much data is incoming and then it can explicitly schedule when packets should be sent. Oh okay okay I see that's and he's built an algorithm that basically says send the shorter messages first, send the longer messages later. So this is about uh from a congestion control perspective it means that get it get us more data faster and save the longer data you know save the longer messages you know for for later. So kind of like QoS I guess this is really interesting. I highly recommend I don't want to get too far into it it's it's it's it's really long actually it's not super long it's just that it gets really deep in the weeds but I wanted to more like just make everybody aware of this uh of course the links are always in the show notes. I highly recommend taking reading reading through this whole article um to talk about it talks about how TCP isn't enough but it also talks about how his protocol aims to be enough so it gets really kind of deep in the weeds on how how different it is.

Katherine 36:41

So really really interesting uh read highly recommended yeah and it'll be interesting it yeah you know some of the things that was mentioning uh I was mentioning is that it's not a general TCP replacement it's more of like uh for some workloads um so you know but it'll be interesting to see the next book TCP IP illustrated uh slash HOMA HOMA illustrated so for those of you who studied network engineering or networking in general there was this famous book called TCP IP illustrated it's actually a really good book um I have it and I think uh most of us who study for CCIEs and other things actually studied this book as well but uh uh it'll be interesting TCP IP slash IP uh slash HOMA um as the illustrated illustrated volume one yeah exactly yeah volume two or something like that um but you know TCP IP is uh older but reliable protocol uh you know in general but you know if we have to like go back like they've had several improvements over the years obviously like it's not like they just come out with a protocol that never gets changed but like uh you know it was originally developed like what 30, 40 years ago. Um so like things change new protocols are added all the time as as things happen. But the big thing to kind of focus on here is that even he admits that it's not a complete TCP IP replacement. So uh it's it would be useful for things like data like the data center uh center AI use case, but it wouldn't necessarily be like what you're what you're playing around with on the internet, like pulling a website or something.

Tim 38:17

Yeah I mean the use case makes the use case itself and what he's talking about r the way it working means that like a lot of applications would would would need complete rewrites to use this type of protocol for example. Like there's there's definitely it's I could definitely see where we don't want to use that as it's not a replacement for TCP except in very specific situations where you basically just have like ultra fast requirement but you still want some element of reliability. I kinda basically that's kind of a mix between TCP and UDP I guess.

Katherine 38:49

Yeah and I and when it comes to writing data center applications, we're updating those all the time anyway. So it like those are like moving forward with like you know such uh velocity that adding another protocol in there actually probably wouldn't be too hard. But doing it in every single like desktop app out there is be insane.

Tim 39:07

It's impossible. Yeah.

Katherine 39:08

I mean it wouldn't be impossible but it would just take forever. Like look how long like we WPA3 came out like what almost eight nine years ago and majority of companies and personal users are still on WPA2. So like just because no one's gonna go and upgrade their actual hardware and stuff then that's hardware but like um you know like even software stacks it just takes forever to adopt something new.

Microsoft Retires Exchange Web Services

Katherine 39:34

Yep absolutely and last story because I I know we have to go back to a uh back to real world um Microsoft uh yeah ends exchange web services in exchange online so basically being able to like not have the hardware client and being able to like log into the like OWA and like check your emails uh that is uh officially ended on uh October 1st but you can temporarily reenable if you're an admin of the exchange online until April 1st 2027 uh after that no exceptions I actually don't really like that because uh but you know like one of the things if if folks have been paying attention to the industry is a lot of people are moving away from Windows since Windows 11 has become w you know pretty terrible and has had a lot of hiccups along the way. So people have migrated to you know Mac OS which is fine there's a ex there's a uh Outlook client for that but a loss also a lot of people have migrated to Linux and there is no you know Outlook client for it. If you want to run Outlook uh in Linux you pretty much are having to run it in like a VM or like or Wine or some some sort of application that ports over like Microsoft uh like Windows uh applications. So this is more going like this will probably affect people who like checking it on the web and not having an installed app, which is probably a smaller portion, but definitely going to hit Linux desktop users harder.

Tim 41:06

Yeah. I mean it also hits any Windows user or and just any user that used the for you know had a legitimate case to use the exchange web client. Like um you know I'm thinking of a hundred different you know situations where you want to be able to check your work email but you know you're you can't connect to your work VPN for some reason or or something like that, right? Where you just need to be able to get to a SaaS version of your uh email without having to to go through jump through other hoops to be able to use the hard the um ethic client. So but yeah it's it's interesting. I mean basically they just basically said you know we can't support it. It's too you know too full of holes and we can't keep it running. So yeah I am I am interested to see what people have to say about it. Like what's the you know like I mean people are just gonna have to deal with it I guess until unless Microsoft changes its mind but I I I don't think it plans to change its mind. Um and what what will come out as a result? Like what what what will be the successor? Will it just be only the client or like what you know where are they going to draw the line?

Katherine 42:15

Yeah or more people will migrate off of Office uh off of exchange. Um you know like Microsoft kind of coming down heavy is why um why a lot of European governments are migrating off of Windows and a Azure and stuff because you know there's been some news stories in recent months where you know Microsoft is unhappy with the government's decision and cuts them off of Azure or something along those lines. And so um this is just another decision that s I could see some major companies but mostly governments being like nah we're we're good now.

Tim 42:51

Yeah any any big change like this I mean you you risk lose uh shaking loose customers that either were you know wholly dependent on that feature you're res you're sunsetting or otherwise you know that's just like the last snail in the coffin for them, right? Like they were already looking at other stuff and this is just enough to push them off. So I am curious to see what the backlash if any uh will be from with Microsoft. All right.

Links In Show Notes And Wrap

Tim 43:17

Uh and I think we are out of time for this month. So um all as usual like I said all the episode notes will include all the links to all the articles. Highly recommend people check out the ones that they find interesting. Uh I personally highly recommend checking out the one on that new protocol. It's very interesting to us at network engineers. And um yeah we'll see you next month uh we'll see you next month for another episode of the uh Cable Clouds podcast Fortnightly news