Tim: 0:15

Hello and welcome to another episode of the Cables to Clouds podcast monthly news update. And uh as always, I'm Tim and I'm here with Chris, my excellent co-host, and we are just gonna dive right into the news. Um, of course, we have the news available in our show notes. Uh you can get a list of all past, uh present, but not future yet uh uh articles for uh your viewing.

Chris: 0:41

We don't have that technology yet.

Tim: 0:43

Yeah, we're not there yet. Uh we will be though, because AI's gonna bring it to us somehow. All right. Uh let's just jump into it. Our first article is about uh Huawei releasing a new open source technique that is supposed to shrink LLMs to make them run on less powerful and less expensive hardware, which, you know, if it works, God, it's about time. Somebody went the other way versus uh making it run on bigger, more expensive, and more uh energy hungry hardware. So uh according to this article, uh Huawei's computing systems lab has introduced a new open source quantization method for LLMs uh aimed at reducing memory demands without sacrificing output quality. So they call this thing sync S-I-N-Q, and uh the basic gist of it is that you know it's from a network engineering perspective, it reads kind of like it's like cut-through switching or something. So so it's it's basically shrinking the uh the heavy math involved with uh with you know AI LLM training and LLM uh expression to just just lower the amount of memory that's required to run some of these mathematical algorithms um without too bad impacting the uh the output quality. So it mentions that it is available um you know on GitHub and Hugging Face. Uh it's uh using the Apache 2.0 license, so it's completely open source. Uh they've tested it on a few uh a few models like Deep Seek, for example, and it just goes the article goes into the details far, far, far more than I could as someone who is not a uh an AI, you know, geek here on the the learning uh perspective, but it goes into the algorithms and how it actually cuts down the the math. And uh yeah, so this is a really interesting article. I would love to see this validated. Um said God God knows we could use some some cheaper hardware uh for wants on this. Uh you you read this article recently. What do you what do you think?

Chris: 2:41

Yeah, it's uh I think kind of to what you said, it kind of reads like you know the idea that we all commonly see in network engineering about doing a little bit more with just kind of the first uh few details of uh of the you know kind of of the packet rather than looking at the entire thing end to end, right? And and only looking at the stuff that's relevant. So that's kind of how it reads. Um but to your point, there's there's a lot of math involved here that goes very in-depth about floating point numbers and things like that. So if that's if that's your shtick, then I highly recommend checking out the article and reading into it because it is it is quite interesting, um, but probably a little above my head on a lot of this. Um it does make me wonder two things. And and the first thing is where where does this uh plan to get us? Like I understand running on cheaper hardware is easier, but like is this kind of moving towards it you know, running these models locally on um uh on things like handheld devices, like actual you know, smartphones or even OT type devices, um, scanners, things like that, or if does this put it running on pieces of infrastructure in um you know, in um on site in inside of a rack somewhere? Um and the other thing is uh unfortunately we have to call this out, but this is this is brought to us by Huawei, which Huawei is one of the largest um, you know, consumed uh vendors for IT infrastructure in certain parts of the world, um, but they also have uh kind of a uh resigning reputation about them in the rest of the world. You know, there are certain organizations that will not touch Huawei just based on uh that reputation alone. So I wonder if that will um kind of impede um potential adoption of this, which is, you know, this is a completely open source thing. It's free to use um based on um uh Apache 2.0 licensing, so it's it's ready to be used um for free. Um, but I wonder how much you know it'll it'll actually take off just because of that reputation.

Tim: 4:40

Yeah, it's a great point. Maybe that's maybe that's why they went the way they did, actually, with the the open source uh licensing and just kind of throwing it all out there, uh kind of being aware of of that, or or maybe just who knows, right? But that's a that's a very good call out.

Chris: 4:55

All right. Uh next up, this this will be a brief one, but it's uh just funny to even see this uh written down. But um, we have an article here from CIO Dive saying that global AI spending uh is attended to approach 1.5 trillion spending this year, according to Gartner. Um, talking about the obviously large investments from many different organizations. Um actually doesn't seem like there's a lot of customers investing in this, it's all other vendors investing in this, um, et cetera, other service providers, et cetera. Um, but just crazy that we're thinking in terms of trillions now, um, 1.5 trillion in spending. Um so, you know, that's that's the numbers that we're talking about. If you think enough, you know, if you think tons has been have been spent already, um, it sounds like it's only going to increase um in a pretty uh exponentially uh larger pace. Um this is um and I think I think ChatGPT just launched uh like in I forget what it's called where you can actually let Chat GPT do transactions for you. Like you can give it your credit card info and it can do purchases for you, like in-app purchases. So um, you know, this is this is the this is the future we're headed towards.

Tim: 6:12

How I mean, how could uh how could that go wrong? Oh man. Now the article points out as well, like capacity constraints continue to be a problem, and simply throwing more money at investment in AI is not really solving that problem. Like, I mean, part uh uh there's plenty of of money being thrown at development of new data centers and and grid, but like the grid is what it is, you know? Like we've got uh companies now buying nuclear reactors to to fuel their AI workloads. Um, but like you can't build it but so fast, right? This this whole thing, this whole bubble could have popped long before the first data centers that were built as a result of all this money changing hands even goes online. Um, so I'm very curious to see where all this money is really going. Because the money's getting spent very, very quickly, but all of the horizons on the result of the thing we're spending money on are like years away, which as we've seen AI changing this quickly, it's gotta stall out at some point because they're just gonna run out, you know, of runway essentially, no matter how much money they throw at it. So I'm this that's why I think the the Huawei thing actually is is somewhat interesting to me because I think we are getting to that point where we're gonna have to start stop worrying about you know getting to the next level of technology and more and more start worrying about like how do we make this thing scalable, you know, just not not just balls the wall against uh you know the infinite sky.

Chris: 7:44

So scalable and maybe just a little profitable, maybe making some of this. Profitable would be good. Oh no, it was plenty of profit.

Tim: 7:52

I mean Larry Ellison made tons of money, right? Like, yeah, I mean not granted it was the same money changing hands between the same companies, but hey, the look at that motil the the they call that uh velocity of money. They call that in economics the velocity of money, right? Um and speaking of actually, so I have a uh we have an article uh from uh from finance, Yahoo Finance, talking about an analyst who sees a 38% downside for Oracle, um, mentioning that it's a quote unquote a risky blue sky scenario. What they what they mean by that, uh, or what the analyst means by that basically is that essentially he's saying it's a bubble, right? Like the Oracle, the the five-year deal or whatever that was uh signed, you know, for all of this money, uh, is saying that you know the the the price of the stock is so far above the value of what is being delivered. Uh, you know, in a five-year window with AI is just absolutely phenomenally long. And you know, basically saying, you know, the the the stock is not worth that. And this um these deals that are these five-year deals that are that are being penned probably won't actually uh go the distance, right? That's basically what the uh the analyst is saying. So very very interesting uh to see that finally where somebody's calling out the the bubble for what it for what it is, kind of.

Chris: 9:14

Yeah, it's exactly what I was gonna say. It was, you know, just previously you were just speaking of a bubble. Um this is kind of the manifestation of the bubble, right? I mean, you know, obviously NVIDIA is in this in this camp too, um kind of adding to this, but I mean Jensen Wong is is not a stupid person, but also he's he's not gonna go against the grain on this one, right? If it if it if the the path forward, you know, by the market seems to be open and a open AI and Oracle um doing this deal, he's you know, i it there's there's money to be made from him no matter what, right? So it's uh that's uh uh not a bad triangle to be a part of for now. But yeah, it's just like it seems like there was a fallout with the relationship with Microsoft, so now we've switched over to Oracle, but like the the facts are still the same at the end of the day. It doesn't matter which uh you know which big uh you know S-corp is is funding this, like where uh open AI is not even profited or not even uh speculated to be profitable for another what five years, something like that. So maybe. Um yeah, maybe that's that's projected, right? Um but I'm sure we can run this fucker into the ground well before then. Uh oh yeah, no doubt.

Tim: 10:25

But it's funny because the you know they're they're not they're burning this money, and how they're burning this money is promising it to like Oracle for and then Oracle doesn't have the capacity that OpenAI is buying. They have to build it themselves. And to build it, they have to go to NVIDIA and give them a hundred billion dollars to pop to buy the infrastructure to build the data center for open AI. It's just ridiculous, this velocity of money thing.

Chris: 10:52

Maybe this will happen again with uh with Google or something after after this relationship goes sour. We'll see.

Tim: 10:59

It's it's truly ridiculous. I mean, if you think about it, like there's so much money being created from you know, it's like Wolf of Wall Street style. Like it's a Fagazzi, it's fucking fairy dust, you know?

Chris: 11:09

All right. Uh next up we have a an article here from Cyberscoop. Um, this is pretty big news, probably reported on them by multiple um outlets here, but this is the article that we pulled up today. But um, so Cisco has uncovered um just a few weeks ago a new SNMP vulnerability um that allows attacks on iOS XE-based devices. Um it looks like this flaw, which is I don't know if it calls out exactly what the grade of the CVE is, but um, there's several of them. Yeah, right. Um but the flaw allows authenticated um remote attacks with low privileges to force targeted systems to reload, causing denial of service. Higher privilege attackers could execute arbitrary code with root level permissions on affected Cisco iOS XE devices, effectively gaining complete control. Um so obviously, like uh I hate kind of coming in here and just reporting on CVEs and things like that, because this happens to all of us, you know. But Tim and I both work in Vendorland. We're we're both going to be coming across CVEs on a pretty regular basis. That's just the nature of the game. Cisco has taken, you know, the approach, kind of pushed out an immediate patch and is urging people to patch quickly and uh effectively. Um the problem is patching takes time, right? So um and not everyone does it on time. And uh, you know, hopefully, I mean, to be honest, let's let's be completely honest here. We shouldn't be exposing SNMP to the internet anyway. Uh so if that's uh if that's the uh route that you've taken, then I would patch immediately because you're um you're already to be honest, you're already doing bad practices. Um so at least patch and then maybe move that to a um more modern uh exchange of uh modering uh capability. But um yeah, uh it's uh just wanted to call this one out because it's pretty important. So if you are um potentially affected by this, um definitely patch and please, please stop exposing SNMP to the internet.

Tim: 13:15

Yeah, this one there's there's actually like three or four of these um that were released at the same time, or just close to the same time. But yeah, and they're all really around C SNMP, and most of them are in the high skill. But all of them are like, why would you yeah, why would you like find yourself in this situation where somebody could explain this? You know, if you've done any basic cybersecurity, you really shouldn't have found yourself in this uh in this hole. Not not to make excuses, right? I mean a C V E is a C VE needs to be fixed, but uh all right. Um our next one has to do with Equinix. Equinix just did uh what they're calling an inaugural AI summit. And so they've released kind of a press release after after that summit talking about kind of the the new feature functionality products that have come out of this summit uh that Equinix is offering. So let's go through it. Um it's a they're calling it distributed AI infrastructure to help businesses accelerate. Um they mentioned fabric intelligence, so Equinix Fabric, of course, is the um kind of their uh automation platform where you can uh create virtual machines and connect virtual machines to to stuff that's in the Equinix rack or to third parties or to the internet or to you know cloud providers or whatever. Um this they're they're adding what they're calling fabric intelligence, which is AI insights based on telemetry tools that are available uh to be connected to the Equinix fabric. So they can do things like, you know, not only find the insights, you like make you like you can chat with your uh your network essentially, type of thing, where you can say, hey, what was my you know, where where am I having problems or something like that. Um but they can also, it says here specifically that you know it can take corrective action uh around the making the network responsive to demands and stuff. So pretty interesting if if if true and and if uh you know let's see how it gets how it's implemented. Um it also mentions AI Solutions Lab at quote unquote Equinix Solution Validation Center facilities. So I think what that means is that they'll there's like certain Equinix facilities where they'll have this connectivity to global AI uh partners. So this this reminds me very much of the uh the Megaport AI exchange thing that Megaport just came up with, with the idea of just kind of inference or AI as a service where they'll connect you to third party providers, you know, like maybe with one of these many data centers that are being built to house AI workloads. Um so interesting. I don't know how much of that is out there yet, still, that like this these third-party providers that will give you AI as a service, but they're building the capability out. And again, Megaport already has it with the uh also, you know, already did something similar with the uh AI exchange, so it makes perfect sense for for Equinix to offer this. Yeah, it's kind of the that's kind of long and short of that one. Anything to to add there?

Chris: 16:04

No, yeah, I think I was um kind of with you. I was gonna draw some correlations to the uh Megaport AI exchange that they launched. I think it was earlier this year or even close to late last year.

Tim: 16:17

I think it was this yeah, I think it was this year. Yeah, earlier this year, I think.

Chris: 16:21

Yeah. Um so uh this sounds like you know, kind of Equinix is also heading in that direction. It sounds like they've um not necessarily one-upped it, but just obviously they have a few more capabilities that they've announced in this one um one summit as well. Um sounds like the fabric intelligence piece is available Q1 2026, so that's coming up next year. And then the AI Solutions Lab, which we've kind of, you know, kind of made um adjacent call-outs to the Megaport AI exchange is available now. Um yeah, it's it's like the thing is these things sound like big announcements, but at the end of the day, it's just kind of connectivity between um right, services. Yeah, services and um a service provider and a customer, which you know that's kind of Equinix and Megaport um together. That's kind of their bread and butter, right? Is is building connectivity. Um the thing is like it there's even at this point, um, there's only a certain caliber of customer that is actually building their own AI type applications and things like that, right? So um probably the kind of top end of town is who's gonna use services like this, which uh obviously they're gonna pay the most money as well. So that's good for sure. Um but I just wonder I wonder how much this stuff is even getting consumed versus, you know, some of it is vaporware at this point, you know, it's announcements that things haven't actually been baked into the platform yet. But um I wonder how much this is if this is actually getting consumed, because I would I would see adoption probably being rather low at this point. Um, but that could that could hockey stick at any point, but that's for us now.

Tim: 18:01

Well, and I think it's easy for Megaport and Equinix to just build that kind of essentially build the road, right? It doesn't even really cost them much to do that, right? And then they can offer it as a service for interested customers. So for them, it makes probably perfect sense, right?

Chris: 18:15

Probably probably makes much more sense for Equinix because they probably own the goddamn data center anyway. So they're selling the data center space to the provider and then just be like, yeah, we'll uh we'll we'll plug some cables in in the meet room for you and bring you stuff. Yeah, it's probably relatively easy for them. All right. Sorry, I was getting my windows situated. Okay. All right, next up we have an announcement from uh Alkira. So if you're familiar with Alkira, um they are a uh multi-cloud networking vendor um adjacent to um the likes of Aviatrix or Prosimo, etc. Um, they have basically launched a uh two products called Alkira MCP and Alkira NIA. Um Alcura MCP, you can kind of already determine what that is if you've um been somewhat plugged into the world of AI at all. So they have a um uh MCP server that's basically allowing you to interact with their AI agents and talk to your network built on Alkira. And then they have this other tool as well called the NIA, which is the network infrastructure assistant, um, which it sounds like those things probably go together. I would imagine the MCP is how you would talk to the NIA ultimately. Um but you know Alkira is kind of building that multi-cloud backbone. You know, they have their own cloud exchange points or CXPs where they um interconnect with each public cloud provider and kind of build that backbone for you to build between uh cloud regions or even cloud providers, um, and even into on-prem uh with the likes of their um hybrid connectivity offerings. If it points out, uh I'll quote this directly, but there there's a section in here from Alkira that says why this is different. Uh and it says point tools give snippets, not the story. Alkira pulls together everything on-prem, multi-cloud, between regions into one reliable picture and lets your platform, your sorry, lets your in-platform copilot or trusted AI assistants use it to help gain a faster path from observation to action to verification. Now, you know, obviously if you're using your own uh in-platform uh AI tool like Copilot or any of your trusted AI agents, obviously this does give them a mechanism to talk to the network that you've built on top of Alkira, which makes perfect sense. Um you probably can gain a lot of insight from that interconnectivity. Um, but I I would I would question how much of that can be between on-prem, um, because the I mean Tim and I worked for one of these vendors for a long time, so we have a lot of context around this. Is you you own most of the cloud network, but you may not own any or much of the on-prem networks. And then there's even things like with the introduction of um providers like Zscaler or any SSE or SASE type providers that also they may integrate into this, but that is that doesn't necessarily mean that those type of things are going to be exposed via the MCP or via this network uh network assistant that they're offering. So I would question really what uh extent you can go end to end. Um I'm not saying there's not value in seeing uh what what Alkira owns um end from their um from their platform, but I do question whether or not it can really go as far as as they're claiming. But we'll we shall see. Um any thoughts from you, Tim?

Tim: 21:44

Nah, I think I think you nailed it. I mean, that's that's the question. The question is how much is can the NIA see to provide the insights that you're gonna chat to it about? And then the MCP, I mean, it's an MCP server, right? So it's going to be exactly as useful as what you can write to do with it, right? I mean, in theory, it should, it should basically it'll be it worked just like an API type of thing where you know you can make changes or connect to or whatever you need to do with Alkira using their MCP server uh instead of their API or something like that. I just think of it like an AI API at the end of the day. So if you build an agent to make connectivity or use Alkira in some fashion, you do now they've you've got an MCP to do it. You don't have to write the uh the API stuff yourself. Uh the last one is honestly feels a little science fiction-y, uh, and then it's a little over my head, but I'm gonna we're gonna cover it anyway because it is really cool. Uh so Cisco is is expanding its quantum networking portfolio with new software prototypes. So uh what that means is, and this is where things really start to fall apart. So I've read this article a few times. So there's a they have a package of prototype software that they're really that will facilitate quote unquote distributed quantum computing networks and support real-time applications. So the quantum labs designed a software stack that includes three layers, and I'm just reading this straight from the article because I don't want to say it wrong. An application layer with a network-aware distributed quantum computing compiler that supports quantum algorithm execution in a networked quantum data center. That means very little to me. I mean uh a control layer with quantum networking protocols and algorithms that support the applications and manage the devices that make up the quantum network uh through APIs, and then a third layer for device support that's like an SDK and API to the physical devices. So basically, think of it, all I can get from this is that we're talking about a big quantum software development kit that, like, you can, you know, goes from the application layer down to the hardware layer, and Cisco's making this available. The what's really cool about this is so OutShift, which is you know essentially Cisco's kind of spun-out think tank, if you will, uh, you know, for startups, announced a quantum entanglement chip. So they're able to actually do quantum entanglement with photon with photons uh via this networking chip. And what this allows you to do, because quantum entanglement means essentially instantaneous state transfer between two entangled photons, is to distribute your computing in a way that it essentially becomes there would be no latency in any kind of like network operation, you know, any operations between these uh these networked computing quantum computers, which is crazy. You don't have to worry about fire, we don't have to worry about fiber or anything, apparently. Like can't get faster than entangled photons. So 200 million entangled pairs per second. The chip operates at room temperature, uses minimal power, and it and functions using existing telecom frequencies. The thing sounds like freaking magic. Like, I don't I don't know else to say. We're we're just talking about building data centers that are gonna suck up the ocean and burn down the rainforest, and these guys are talking about quantum computers that operate at room temperature, like crazy stuff, right? So, I mean, again, read the article. I can't do it justice. The the the level of of just craziness involved in in this uh this new software stack is just really, really interesting. And what it might allow. It's funny to see technology not diverge, but just almost diverge, like just take two completely separate paths. So here we got AI doing what it's doing and just eating every resource on earth to become smarter, faster, better. And then we have stuff like this that's kind of taken the other track where quantum computing is actually so efficient that you're almost, you know, you're not quite getting energy back, but like it's almost it's almost not using any. So I don't know, what do you think?

Chris: 25:38

No, yeah, I mean, uh like you, Tim, I am but a simple man. So many much of this article does go over my head. Um, but the one thing that I do like they called out was the the importance of the error correction, um and to make sure that this is obviously accurate for uh transmission of you know photons and things like that. Because that's the thing that was, I think, the craziest thing is that it just sounded like all the existing infrastructure doesn't need to change. Um maybe there's there's some chips involved here, but like that you can use the existing fiber, etc., like you said, runs at room temperature. Whereas when we're hearing about these quantum computers today, it's like, oh, it needs to run at this temperature, it needs to be in this contained environment because the um the errors that can be introduced by such a small variance in that type of stuff um could be critical and potentially resource intensive and wasteful if it you know has to re rerun some of these calculations. So uh, like you said, the proof is going to be in the pudding. Uh I think, you know, obviously there's a lot of detail in here, and Cisco's probably skating towards the puck in a lot of these um type of scenarios. Um but um I'm interested to see where this goes. The thing that's the thing that's crazy about this is like this this is all related to like computation, right? It's not necessarily related to the network. Um, so it's kind of hard to grasp um what this means for us as like network operators, right? It's like it just sounds like shit's gonna be be moving really fast, like all the time. There's a lot of data moving really fast. Um, so it's hard to kind of grasp what this means for where things are gonna go, but um things are changing. That that's kind of the the only constant is change, right? And I think that's uh that's definitely prevalent here.

Tim: 27:30

Yeah, I forgot to mention one thing that I thought was also pretty cool. Um another research proto this is a prototype, this doesn't exist as such yet, but this idea of quantum alert, which is uh like a security feature that can tell if two endpoints in a quantum link if there's an uh an intruder in the system, basically. Because it's entanglement-based, apparently they can rapidly detect if somebody shouldn't be there, essentially if there's a thing in this that shouldn't be there. So they've already talked about this, and they're talking about like um, you know, but basically by an intruder even entering, you know, there's entanglement that has been screwed with and therefore is detectable. So think about cybersecurity, the future with quantum entanglement. I don't even know what to say about this, right? Like anyway. So this is really cool. I highly recommend everybody read this article. Don't don't skip out on it. It's really, really interesting stuff. Um, and with that, I think that does uh finish off our news uh episode. Uh thank you for sticking with us, assuming that you've made it this far, which you if you're hearing this or say you know, you you must have. So thank you.

Chris: 28:35

You skip to the end because you just wanted it to end that much faster.

Tim: 28:39

Yeah, there it is. Or you or you skip to the last the last story because this was the interest this was probably the most interesting one. So anyway. All right, everybody. Well, thanks for sticking with us, and uh, we will see you next month with another news update.