Telemetry News Now.
Welcome to the latest episode of Telemetry News Now.
Lot of stuff going on in AI, if you haven't heard, and a lot of stuff going on with China in particular. So we're gonna be a little bit heavy on that conversation today.
So without further ado, let's get into today's headlines.
So first, DeepSeek beats GPT o1 on certain benchmarks. And this is out of Tech Crunch in particular, but you have probably seen this all over the news, read some stuff out of Reuters, out of AP News, a few YouTube, channels that do tech news as well. So from a variety of sources, the Chinese AI Lab DeepSeek, which I, again, I think everybody's familiar with at this point, has released DeepSeek r one, which is an open source reasoning model under an MIT license. This is a big deal. They're claiming that it rivals OpenAI's o1 in benchmarks like, AIM, that's AIME, Math five hundred, a couple of other benchmarks as well. And it so far, it looks like it does rival, you know, OpenAI's o1, making r one, highly, highly capable. But the thing is that it's much smaller, and then even distilled versions of it allow it to run on local hardware, much cheaper, kinda like off the shelf server hardware.
Unlike traditional AI, r one facts checks itself. You might have heard of reasoning. You know, the whole idea there is that it improves reliability, specifically science and math. But, usually, that's at the cost of much longer processing times and, therefore, higher inference costs.
Right? So r one is subject to Chinese regulatory constraints. That's a concern. There's filtering, you know, possibility, politically sensitive topics, that sort of thing.
And this whole thing comes amid the US China tension over AI technology in general with the Biden administration, if you remember, tightening export controls.
And specifically from the Financial Times, they put out an article, that explained that OpenAI has evidence that DeepSeek used the GBT model to train the DeepSeek model, leading to White House AI czar David Sacks, this is just from today, to raise the possibility of alleged intellectual property theft. There's so much going on here. The thing is that, in my opinion, there's so many amazing new opportunities with an open source model that can operate at this level that's much cheaper, that can run on much cheaper off the shelf hardware. So we're talking about revolutionary tech that can really change the world even more even more than AI already has been. Right? But the security, the privacy concerns, the geopolitical concerns, I mean, they're off the charts right now.
Oh, yeah. There's so much to unpack here. You know, I think it it's interesting to think about what is the difference between what r one is as far as an open source model and what is the you know, what do they keep back for themselves too? Right? From a security perspective, I think that's the part that's the the most scary to me is if they're willing to open source this, the odds that they have something even more powerful that they're holding back, I think, are probably pretty good. Right?
Yeah. Yeah. I mean, the thing is that when you say they are holding back Deepsea, this particular firm yeah. I mean, Deepsea, you know, being funded in completely by that hedge fund or whatever they are, some kind of financial company, and then being able to produce this product for something like five or six million dollars in investment, you know, that's the game changer.
But, you know, what are they holding back? Could they be holding back that it no. It costs much more than that. And, no, they did not use, like, the older cheaper NVIDIA chips, and they were actually or chips, not that the you know, the entire, platform.
I've, you know, I've been going through various news articles and YouTube videos where folks are are wondering if that's true, if indeed they are or were or did, use the more expensive NVIDIA products, at much greater cost to develop this after all.
Well, you have to remember that, you know, the Chinese nation, the media is all controlled by the by the government. Right? Sure. It's a very different setup than the US where the media, in theory, at least, has freedom, free reign to to publish what they want.
So you always have to take anything that comes out of the Chinese media with a little bit of a grain of salt. Right? Because it they're going to position it in such a way that it makes them look good. So to your point, you know, is that a hundred percent factual that they used older hardware, that it was cheaper to develop all that?
I mean, that's what they're saying, but we, you know, gotta be a little careful taking that as as gospel, I think. But, yeah, either way, you know, I think it's it's gonna be interesting to see, and, you know, it's it's interesting the effect it's had on the US markets and and the AI arms race, if you wanna call it that.
So Yeah.
Yeah. I mean, the thing is from a cost perspective, if it is true because we have it's an open source model or actually, you know, DeepSeek has we've been known we've known about DeepSeek for a while now.
Mhmm.
It's just this recent news with these, these recent models. You know, it being open source, we can sort of see how it, by nature, is going to be cheaper because they're able to accomplish these same benchmarks with far fewer parameters. And the parameters, you know, is gonna dictate the size of the model and dictate the cost of the model, usually usually also dictates how accurate and and useful it is. And DeepSeek is showing how that's not necessarily the case.
If that's true, that could imply that maybe, you know, the big AI companies, OpenAI OpenAI, backed largely by Microsoft, Meta, you know, with with Llama. You know, we have Claude. We have Anthropic. All these companies, are are they approaching it the right way, investing hundreds of billions of dollars in order to get these models up and running with hundreds of billions into the trillions of parameters.
Is that necessary to do?
And if not, you know, what what what does this mean for the future? Now another thing is being open source, if we can integrate this model, if companies can integrate this model being so powerful and accurate and cheaper to use into, like, everyday apps all over the place, it's kinda like TikTok, isn't it? All over the place exfiltrating information, unwillingly. Right? Yeah.
I mean, if I look at the silver lining here, though, I think, you know, the fact that it's out there, the fact that it's open source does give other companies the ability to dig in and see what have they done, what can we learn from them, what can we benefit from them, can we do this more cost effectively, energy effect effectively?
I mean, we've talked about in the podcast before. One of the big challenges that we foresee as a society with a lot of the AI boom that we're going through is how are we even gonna do the power, right, to build these data centers to or doing this compute. So, you know, if we can learn from something that the the DeepSeek and the Chinese have done here to to build this a little bit more efficiently, then, I guess that's the silver lining. Right?
Yeah. Yeah. For sure. And the fact that it's, you know, open source does suggest that there is this you know, how scientists share information Exactly. All over the world. So that element is there as well, at least maybe from my naive perspective.
On a very related note, also from TechCrunch, DeepSeek's release of its r one AI model, which we've been discussing, has also significantly impacted NVIDIA. Surprise, surprise. So following r one's launch, NVIDIA stock dropped nearly seventeen percent.
Different sources are gonna have that number pegged a little bit different, but that's right around seventeen percent, erasing almost five hundred to six hundred billion billion with a b dollars in market value. So as investors speculated that the high end AI chips may not be as essential for top tier AI performance, which is what we've been talking about. Right? And it comes amid this, you know, shifting US policies, with the outgoing president Biden and now the incoming president Trump on AI chip exports. I mean and what I'm referring to there is that Biden's restrictive order, right, it was recently reversed by Trump's five hundred dollar billion Stargate project, if you remember that in the news, which is basically an investment in AI infrastructure in the United States. So, you know, DeepSeek's success kinda suggests that the US, AI dominance may depend on more than just controlling chip supply.
I gotta think this is a little bit of an overreaction in the markets, to be honest, Phil. I mean, I get that DeepSeek makes it more efficient, but there's still a lot of of room to grow for NVIDIA here. Right? You just mentioned the investment that Trump wants to make in Stargate.
So the US still has a lot of way to go to building out a lot of these, environments, and that's just one country. Right? All the other developed nations on this planet in, you know, in theory are gonna be starting to do something similar, right, where they're gonna wanna build data centers, they're gonna wanna keep up with this digital arms race that's going on around AI. Right?
So I'm I'm personally not concerned that, you know, this is really gonna be a major impact on NVIDIA. I recently finished reading, the NVIDIA way by, Tae Kim, which is a a book about the founding and the growth of NVIDIA, a little bit about Jensen Wong and how he re runs that company.
For anyone who's listening and is interested in all of this kind of Philip race and all things custom ASICs and custom silicon, I highly recommend it. It was a really good read.
But, you know, they talk a little bit in that book about how, you know, Jensen has always been a little paranoid. In fact, I forgot exactly how he says it, but paraphrasing is, like, we're thirty days from going out of business. Even today when they're one of the most profitable companies on the market, he runs the culture internally that way so that they're always paranoid that a threat like a deep seek is gonna pop up, and they're gonna have to be able to deal with it. Right? So he doesn't want his company and his team to get complacent that, you know, we're, you know, the big player here, and we can relax and and just enjoy, you know, the spoils of our the fruits of our labor, if you will.
So I have a feeling that they've got a plan or they're developing one quickly.
Yeah. And Deepsea did train their models on NVIDIA technology. It's not like they were able to, you know, buy some stuff off of eBay and, you know, whatever. It's just to a far less extent and and presumably older technology is if we if we believe what we hear on the news.
So it is still true that NVIDIA is still very, very important, but you know how it is. An entire market gets swooped by by news like that. And, so yeah. Well, you know, it's gonna be interesting to see how this plays out in the in the coming days, especially when we start to, see how this balances with the geopolitical nature of this whole issue and not just investor money. You know? And and, of course, a lot of this investor money is crossing international boundaries. So who knows?
Yeah. I think we talked about that on the last podcast where there's investment funds from foreign nations that are investing in AI data centers in the US. Yes. There's a lot of lot of money going into AI and and the infrastructure underlying it for sure.
Yeah. Yeah.
Well, I warned you at the outset that this was gonna be a heavy I AI episode, and I was not lying. So from the Cisco newsroom, Cisco has announced AI Defense, which is a comprehensive security solution designed to protect enterprises as they develop and deploy AI apps.
So with, AI technology, right, with it's evolving rapidly.
I guess, more traditional cybersecurity measures, they're not enough to address, like, these emerging threats.
So, you know, data leakage that you might have, you know, using AI, model tampering, AI misuse, however you wanna look at it. So Cisco's AI defense, it leverages, Cisco's network visibility, so their security, you know, expertise to provide kind of an automated model validation, a real time threat detection and access control mechanism, which, you know, the idea is that it ensures an enterprise can scale AI safely. Right?
It does integrate with Cisco's security cloud and, therefore, provide some sort of cross domain AI protection, which, you know, makes sense the way people use, technology today, multi cloud environments especially.
And, I believe based on what I read in, Cisco's press release, we're gonna be launching or they're gonna be launching in March. And, yeah. So we'll see.
Yeah. I mean, I think I may have mentioned this before on the podcast, but I was, listening to a keynote speech at a conference by the chief revenue officer officer at OpenAI, and she made the point that one of the interview questions she always asked people is, how do you feel about AI from its, you know, security standpoint and its potential impacts on society? And her her point was if people don't wrestle with that, don't, have a bit of a challenge with that where they see the positive that AI could bring to our society, but also the potential downsides, the potential impacts on things like data leakage, like you said, or even using this as an attack vector.
You know, people can now use AI to be much more targeted in their phishing attempts. There's all kinds of, you know, security issues that come along with AI, which is why a lot of people are kind of opposed to it. Right? Anyway, her point was, if you don't wrestle with that, then that's a bit of a problem.
Right? That should be something that you're you're concerned with. Right? And I think Mhmm. I'm I'm I'm excited to see some investment in that side of, AI things.
Right? I think article's a little light on details on exactly how all this works. So I'm sure as as it's released and people start to use it, we'll start to hear more about it. But I definitely think, you know, investing in in research and products that are helping us secure AI is where deploying a lot more.
This is, much needed for sure.
Oh, yeah. I agree. I wish that there was a lot more meat out there on how this works. You know, being a nerd, you know, I wanna know how it works.
But, also Yeah. I wanna understand if this is truly an effective mechanism to protect against some security threats and vulnerabilities or if this just be the only answer.
That's the thing. So I wish I could provide a better analysis on, like, what I think. Is this thing effective? Is this, is this new product, something that we should be paying more attention to?
But I will say, Cisco is not part of the AI conversation that much. You know, we've been talking about NVIDIA and DeepSeek and Mhmm. You know, a lot of and OpenAI. So there's even Microsoft and and and Meta and and Google are part of the conversation, but you don't hear Cisco's name nearly as much.
And so that's interesting. Now, you know, the thing is, though, Cisco is integrated in enterprise networking, enterprise tech stack, you know, very very much so. And so it does make sense that they have a a really great opportunity to, integrate this kind of a mechanism where they already are and into the tools that they already have that are presumably already deployed out there. So it's an easy lift, I think.
Whether it works or not, I don't know. We'll see. I would love to look into this more, as as we learn more from a technical perspective.
Yeah. I know. A hundred percent.
Alright. Next up is an article from investing dot com from, yesterday, January twenty eighth, as we're recording this. Juniper shares tumble on DOJ lawsuit concerns. So we've reported before, and I think most people in networking are probably aware, that HPE had entered into an agreement to acquire Juniper Networks.
It had gone through most of the legal and regulatory gates that it needed to to be approved for that merger, acquisition to go through. The EU, the European Union, and the UK regulators had approved it. And, Phil, I don't know if you remember. I wanna say it was, like, last fall maybe.
It's been been a little while now Yeah.
That that that they approved it. And when the deal was originally announced, you know, analysts seem to think that was actually going to be the hardest one to get through because the European Union, their regulations tend to be a little bit more strict when it comes to anti competitive stuff than than the US. And, you know, it's been for the last few months a lot of speculation on why it's taking so long for the DOJ to improve it, to approve it. Sorry.
See how this most recent article that the DOJ has concerns on this has been, you know, had a major impact on Juniper's share prices and so forth. But, you know, I think it'll be it's gonna be interesting to see. I mean, the the hint in this seems to be that maybe this deal won't even go through, that the deal may be dead. The DOJ is not going to bless it.
That'll be as you can see, there's penalties built in this. So if people aren't aware, when an when an acquisition like this takes place, there's money that's put into an escrow. Somewhere like when you, you know, have a house and you have a mortgage and you pay your taxes and insurance through an escrow account, they create an escrow account when one of these acquisitions is going through where they put money into it. In this case, there was eight hundred and fifty million dollars that HPE put into an escrow account.
And if this deal doesn't go through for regulatory reasons, Juniper gets that money. So, yeah, it may, on the surface, seem like, oh, this is bad for Juniper if this falls apart, but they're gonna they're gonna get something out of it. They're not gonna walk away with nothing. So that'll be interesting.
I mean, interesting to hear Well, sure. Sure. They they're not gonna walk away with nothing. They're gonna walk away with almost a billion dollars, but that wasn't the plan. I mean and if it was, man, that's, like, you know, plans within plans. Right? Alluding to Dune if anyone's a Dune fan.
But, certainly, that's not the original intent. No.
For sure not.
Yeah. The interesting thing is if it was a it was an unconditional approval from the EU commission. Right? Mhmm.
So I would have expected it to be the other way around. Like you said, they're usually more strict and stringent, and it's a little bit easier on the US side. And that's the, it's the inverse here. So this is a legal thing with Department of Justice.
So, you know, who knows? I don't know. And we have a new administration now. We have new perspectives, new leadership.
So I I don't know where this is gonna go. But, certainly, you know, I'm sure investors among those at HP, but investors in the market in general, the tech market are going to be paying very, very close attention to this because this is a big deal. We're talking about one of the heavyweights of the networking industry and then, of course, HPE being one of the the, you know, the biggest tech companies, period.
Mhmm.
And you also gotta remember that, Juniper, more so than HPE, but both of these companies are actually vendors to to the US federal government too. Right? So they have some vested interest to protect, critical infrastructure that they rely on for all of our defense agencies and all of our government agencies to communicate and run their networks and so forth. So there's a bit of a vested interest that the DOJ and the, you know, the US federal government has in making sure that if this acquisition goes through, they're not left high and dry and don't have support and and the ability to purchase new hardware and get hardware software upgrades done, hardware replaced when there's failures, all of those type of things for the for the equipment that they have in their networks. Yep. So there's a lot of angles to this one too.
Alright. We'll move right along. Next article is on, some some tariffs. We alluded this a little bit earlier that Trump has placed on, chips being exported to Taiwan.
Specifically, Trump pushed hard for ASML, who is a company who manufactures well, I understand it, Phil. They manufacture equipment that manufactures chips. They don't actually manufacture the chips themselves, but they manufacture the equipment that companies like TSMC, who fabricate the chips, actually make the equipment that makes this. So that's one company. It's a Dutch company. I think I believe based out of Amsterdam that Trump is wanting to place additional tariffs on.
Interestingly enough, they actually just released their earnings, after the market closed. I believe it was yesterday, and their earnings look really good. They actually beat their estimates. So kind of a mixed bag here for for ASML.
Their stock jumped twenty sorry. Not twenty percent. Ten percent, after they had strong q four earnings. So it's interesting that, these earnings are coming in the same week that all of the DeepSeek news that we talked about earlier in the podcast is coming.
So, again, it's a a strange market in that there are companies that are still continuing to make profit while DeepSeek is being able to presumably train their models using, lighter weight and less resource ASICs. So still more growth in this, in this part of the market, I think, that, we're gonna see come. So Yeah. Yeah.
I mean, I agree. It's interesting. Sure. But, one thing that I I'm I'm noticing is that there's a stark contrast between the numbers here, ASML in particular.
Right? Their q four twenty twenty four earnings. And, of course, also, I don't think we we discussed that their, their twenty twenty five revenue outlook is very strong. Thirty one and a half to thirty six point eight billion dollars is what they're, predicting, and they're citing continued strong AI and semiconductor growth.
And this is ASML, I grant you. But that is in stark contrast to the the implication that some would say that we don't need these high end chips anymore because of, you know, DeepSeek r one coming out. So there is a contrast between the numbers and the predictions, in this case, of ASML and what many of the talking heads in tech media and social media and that sort of thing, are saying.
Alright. Last up this week is, an article from CNN dot com talking about, the TikTok ban that took place. So TikTok went offline for a few hours, right before Trump's inauguration, right at the end of the Biden administration, then came back online. And, our own Doug Midori has done a little bit of reporting on this on what we see on the Internet as far as the traffic goes. You know, TikTok did go offline, came back online. Interestingly enough, when it did come back online, ByteDance, who is usually their number one CDN that they use to deliver that traffic, was not one of the places that traffic was being delivered for the first few hours that TikTok was back online. And then, you know, ultimately, it did come back online.
Trump has pledged to help keep it online. And then, of course, if you're following this ongoing saga of TikTok and its ownership with ByteDance and Chinese parent company there, Trump is help pushing to try and find an American company that would buy these assets away from from ByteDance. There's still, you know, more to come on that, but it does look like they're starting to become some companies that are interested in that.
Yeah. I'm I don't use TikTok, and then my children are not allowed to use TikTok in my house, but I have been following this. And, I do remember it was an executive order that president Trump made, I think, like, immediately after his inauguration.
It was actually before his inauguration.
They they allowed him Biden actually deferred the decision to Trump and allowed him to make the decision on what he wanted to do around this. Yeah.
Well, what he wanted to do was delay enforcement of the divest divestor ban law.
Right? And then Yep. Exactly. And that's when TikTok's you know, as the application itself, the web page, began to be available again for for US users in particular. And remember that it wasn't like we had, American service providers, shut things down. TikTok shut itself down, you know, in in accordance with the new law and the new order.
I don't know. And maybe that's a political victory for president Trump, because he gets to claim responsibility for bringing back TikTok. Many people are happy about that. I don't know. But, we have discussed the nature of TikTok on this podcast several times and its, potential for data exfiltration, use for, you know, specifically China, and doing whatever kind of, reconnaissance and information gathering. So there there are still concerns there. So I'm not really sure how to interpret this exactly.
Yeah. The one interesting thing I took away from this specific article was a statement by, I guess, the spokesperson for TikTok saying that, they will continue to work with Trump on a long term solution that keeps TikTok in the United States. So most of the news that I have read about this, they've been pretty uncooperative, it seemed like, that they have been defending that they're not, under Chinese government control, you know, that this is all of the speculation around data privacy and so forth is all hearsay and, you know, have been Yeah. Right. Yeah.
Sorry. Did I say that out loud?
Yeah. Yeah. No. I mean, you know, that that's an interesting one. I was thinking about when we were talking about at the top of the podcast around DeepSeek and and some of the data privacy concerns there, I think, you know, it's important to know or to think about that the Chinese laws and the Chinese culture do not really have this concept of intellectual property rights and copyrights like we have in the United States.
Right? It's considered part of their culture that if you have developed something, you've learned something, you've built something, that everything is open source. It's kind of the the way you think about it. I'm probably doing a really poor job of articulating the the the cultural thing there, but that's fairly well ingrained not only in their laws, but also in their culture.
Like, they think it's really bizarre that in this country, we allow companies to protect what they've built instead of making them share it with everybody. Right? And so they take that sort of approach to their international way of interacting too. So not saying that's good or bad, not trying to make a political statement.
Just something that I think we as, IT practitioners and networking folks just need to be aware of, that that's the way that culture works.
Yeah. Yeah.
I mean but do you you remember that the law that banned TikTok, that was bipartisan in congress, and it held a lot of support.
A lot of Americans were in support of it as well with the exception of, like, much younger people. Mhmm. I don't know exactly what age you wanna attach to that, but, certainly, you can imagine the, the primary audience for TikTok. Right?
And so that is interesting to me, considering that president Trump spearheaded, this reversal.
I'm not sure what the political strategy is there, but I know that the solution that he is working towards in particular is co ownership with an American company. ByteDance will be a co owner with an American company solving that problem. And I don't wanna say sidestepping any laws because you're fulfilling the law. It's not really sidestepping in the sense that you're on you're you're fulfilling its requirement.
So that's how that's gonna work. But, this is kicking the can down the road as well, because who's to say that American companies aren't necessarily, also infiltrated by nefarious, you know, people from from from China that are trying to also, you know, exfiltrate information from there. Also, you know, just having co ownership doesn't necessarily solve the technical problem of data exfiltration, within, you know, the code of the app, being able to do that and send information to back back to China. So that that's still an issue.
It might give a little bit more oversight by somebody in the US. We have a little more detail and data on what's going on, but it's not gonna be a hundred percent transparency day one. That would be naive to think that that's gonna happen.
Yeah. But, you know you know, American companies like Apple, for example, they were quick to, you know, go go right along with it. You know? They Apple removed it from the App Store, you know, like, instantly. Right?
Yep.
And there was no issue there was no issue there. So but, I mean, the thing is, the purview of a president does not allow him to just completely revoke a law passed by congress and signed by a president. Right? You can't do that by virtue of an executive order, exactly. So, so we'll see how this plays out in the short term because I know I know president Trump loves social media. So, I'm sure this is gonna be something that pops up.
Yeah. And the other thing to remember is Trump does have his own social media platform too. Right?
So Oh, yes. That's true.
Some potential for, you know, financial gain on his end.
I'm not saying that's the reason this is being done.
There's always that angle that follow the money. Right? Like, you always gotta follow the money a little bit when trying to figure out what's going on in politics.
Yeah.
Alright. And now for upcoming events. First, we have Nanog in Atlanta, Georgia next week, February three through five. We're gonna have a few people from Kentik there, including Justin. So if, you see him, please make sure to stop and say hi.
We have, Apricot in Malaysia in a few weeks, February twenty seventh. Justin, I believe you're gonna be there as well.
The Missouri nug, held by the, USNUA. It's gonna be, next week as well, February sixth in Saint Charles, Missouri. And the Connecticut Nug in Hartford will be February twenty. If you're not familiar with the NUGS, the networking user groups that's part of the USNUA organization, and to see if there's an event near you, go to the USNUA website, click on groups and events or something like that. But you go to the website and look for events, you'll find it, and see, if there is one near you coming up soon. So for now, thanks so much for listening. Those are the headlines.