2025 Internet Predictions (AI Included!)

Hello everyone and welcome to this week in net. This is our final episode of 2024. So happy holidays and end of the year for those who are celebrating. Today we're reflecting on the year, but focusing more on making predictions about 2025. I'm your host, Ront Mabasten, Lisbon, Portugal, a very cold today and rainy. Lisbon, Portugal, and with me, I have returning to the program, RCDO, John Graham, coming. Hello, John. Hello. I'm inside. Stay. I'm the rain. And funnily enough, just as you started this, the rain started really beating down on my roof. So I don't know if anybody can hear it, but I'm just underneath the roof and it is hammering down. So, actually, I don't think we can hear the microphone are good. So a very busy year, another AI year in the sense with a lot of AI things. Actually, the week we were recording, you were on Yau and finance talking about many things and AI, of course, was included on that regard. Yeah, absolutely. They wanted to chat about what's happening with AI because they're not everybody knows the that I guess they thought I knew the answer. So they asked me. Did you know? I mean, you know, we could talk about that right. We could talk about predictions for exactly. It's definitely not going away. That's pretty that way. So where is this going to be? There's going to be more AI things. For sure. Actually, 20 May 4 Mark, 12 first 14 birthday. And we also had a great birthday week in late September with a lot of announcements, cool things with a mission in mind. to help build a better internet. And you did 13 years at Clulford in December. I did 13 years in December. It's been a while. Yes. It's been a while. Yes. It was a very small company when I joined and now it's no longer a very small company. Although it's still small by some measures. So if you think about we've still a lot of growths do. There's a lot of things that we're going to be doing over the next few years. So yeah, lots more to do even though it seemed It's big to me. True, true. And big to me too. But on every guard over the years and I actually first found a Matthew Prince, our CEO, Blockpost from 2011 actually, doing some predictions. There's a few from you actually a few years after that, doing predictions for the incoming year in a sense. How easy it is or actually how difficult it is to do predictions for the incoming year. And how relevant it is even. So I think one of the problems with predictions is that they end up being either kind of linear with what has happened before. So you can kind of say this trend looks like it will continue. And those are not so interesting. What you really want is the big surprise ones, right? So you want to be able to a next year, there's gonna be an artificial general intelligence, or next year aliens are gonna visit the planet or something like that, because that would get people excited or scared or whatever, both of those. So it's kind of like your stuck between somewhat tedious and things that, well, maybe they're very, very hard to predict. I mean, just look at what's happened in Syria very recently where this sudden overthrowing of the government there if he sat back and said last year, do we think that's going to happen? People would have said no, but it did. And they're so hard to predict those things. But we'll give it a go. We'll give it a go for the end of the year. Actually, on that note, I was hearing a podcast, an episode that was like three days ago. And it was mentioning like a real expert on Syria, imagining that overtraying the government was not that possibility in the future. And when I was hearing the podcast, the government was already being thrown at the drone. So they really changed quickly when they changed it changed Greek. And on that front, I was like one of the things I did was I went to cloud player radar to look to see, you know, what was the effect on internet traffic in Syria, because it could have gone like two different ways in my mind. It could have been blackout, right? Or it could have being an incredible increase in internet use because maybe people will be communicating more freely as you've seen sometimes in in other situations. Um, neither of those things happen. Right. Show how good I have a prediction here. Right. We saw a sort of a bit of a drop for a few days. Um, but what's interesting, I think in that story is that there wasn't a blackout. And I think that indicates that, um, the government of Syria fell so fast that They didn't even get around to saying, well, maybe if we cut off the internet, we'll be able to get this thing under control. True. Actually, why not just show folks the internet perspective here? One of the things that I found interesting, and we did some tweets about this a few days before the government was over -drone. Internet dropped a lot. And then it picked up already this week. So with the changes being made. It was on Sunday, if I had not mistaken. So there was definitely a drop for a few days and then it picked up again. It picked up again, right? So it's a good thing. And many years ago, we looked at an internet outage in Syria. Well, you can probably find out on the Cloud Play blog somewhere, way, way, way back, which at the time the Syrian government blamed on a cable cut, but which we know wasn't because Syria has multiple cables in different locations because of its geographical location. And they all went down simultaneously. So that was a sort of slightly suspicious cable cut. Clearly wasn't. It was clearly government mandated. This time, no, that happened. Is this one? How serious are enough? The internet? That's right. That's right. It was like 12. They were saying that it was caused by a cable cut. But if you look at Syria's connectivity, because they have multiple lambourd because they actually have multiple cables across those land borders and it was suspiciously simultaneous. Well, going on, on the predictions now, before we go into those types of predictions, why not also try to review a bit of what was the year for a call for in very specific terms. We launched new tools, new improvements, developers had a lot of new tools to play with in a sense and build stuff with workers. More automation in the DDoS front AI audit that had a, this week we were recording at a very important update with a button where content creators can block the crawling of their content, right? So in an nutshell, how can we highlight some things of 2024 for Cloudflan? Wow, I mean, that's a big question. As ever, CloudFear is super innovative. We have weeks where we put stuff out and we're definitely not slowing down. There's my prediction for 2025. There'll be more stuff from CloudFear. A couple of things I think, though, we lost about a bunch of stuff around AI control. Because what I think a lot of our customers are seeing the ones who are in managing corporate networks, they want to get control on over AI tools. Who's using them, which ones are allowed, also the customers who are getting scraped by the AI scrapers, they want to make sure they've got control of that. So we lost a bunch of tools around AI. That was really important. We also launched a course stuff around workers AI. We constantly bring out the latest LARM models on the platform. And that has become come a really powerful platform and around this end of year period, you're going to see applications built on the Cloudflare platform hit the public consciousness with some pretty major brands involved. So that's coming up pretty soon. But a lot of other things, I mean, we talked about other things. I mean, the other area where I think has a long, long bet that Cloudflare at workton was, I always wanted them We had our Nvidia partnership many years ago. It's quantum -resistant encryption. And just the other day, Google announced a new quantum computing chip called Willow. We've spent about the last five or six years working with standards bodies, with other partners on quantum -risk -resistant encryption. And there's a lovely graph on radar showing the use of quantum -resistant encryption in the real world has really spiked up this year. And that's because we made it available to everybody and Google Chrome implemented it in the browser. And so a lot of connections are now quantum resistant if they're coming to Cloudflare properties. Makes sense. It was a clear increase in terms of percentage. Yeah, things like that. It is. And those are one One of the topics that a tech company has called for must be thinking of that before it goes to the news cycle in a sense. So it's now on the news cycle. There are folks worried, hey, what happens when quantum computers can decrypt what is in the internet in a sense, but there are already things in place for a few years that are gaining momentum to try to avoid that. the potential in the future. Well, we know how to avoid it, right? We have the algorithms. We now have the standards and we now have the implementation. And Cloudflare has been involved at all steps of the way to make sure that was part of what we do. And some years ago, we announced that post quantum cryptography will be included in all of our product plans. So no matter whether you're a free or a paid customer of Cloudflare, no matter how much you use your Cloudflare, post quantum cryptography is just part of what we do, because it's the same for the future of the internet and you know we have our mission help build a better internet this is possible. Exactly. Here's a sum up already in December reaching almost 30 percent of ATP as the request traffic being postponed and encrypted and as you mentioned it was between March and April that things started to change clearly. That's right as Google Chrome started to ramp up the availability in the browser we had worked with Google over the years to make sure these algorithms were suitable for use, we've done implementations, and we were ready for them and here we are. So there's a lot of connections are using that coming to Cloudfner, and of course the rest of the industry needs to catch up and have the same algorithms available. For sure. While we're sharing a screen, this is a 2014 -13 first of December 2014, so you were working the day before for me. It's me. It's you. A call there in 2014. Yes. Bigger faster and secure in this case. I kind of made up a word there. I'm much more secure. Is really a real word. So I was playing with English. I'm I'm a Portuguese. So for me, new words are make sense. I would be safe to cure or Portuguese in my cigar. Yes. or okay. Yeah. Yeah. Nice. So, exactly. We don't have like a real specific word for you say it better than me. Your accent's better than mine. I'm Portuguese. So I have years of that. There was a lot in terms of what was called for at the time. Also massive DDoS attacks. Yeah. Very small compared to the this day and age type of DDoS attacks. The expansion of data centers. all around and at the time hard, bleed, challenge. So it was a different internet, in a sense, at a time, although there are things that are pretty much the same. Was the year we introduced Project Gellie, who actually, keeping services. And years ago now, we did the 10 years celebration. You were very involved in that, in that year. Exactly. And also QLS, SSL, also an important thing at the time. And the time. Because L very important people keep control of their SSL private keys. We had shell shock. My goodness, that was a long time ago. I remember shell shock very well. Shell shock is a funny one. We decide is one of the vulnerabilities that we decided was so bad that we would roll out protection for all our customers, including the free customers who typically don't get the web application firewall. And we've done that for a few other things. We did it for local for J as well. We rolled out some basic protection for everybody. And Shell Shock was actually one where I remember the very clearly hard coding the Shell Shock, pathon into our Dua code. And I wonder if it's still there. Maybe it is. Maybe it is. A lot of changes in 10 years. So why not start also to project what's coming in 2025? In terms of the year it had, what are the He changes that we did in 2024 that you think could be really important for what's coming. But then they were not as important now that they were lost, but then so they will be in 2025. Well, I mean, earlier I mentioned AGI and an alien invasion, right? So those are my two big predictions for 2025, right? There'll be two sorts of non -human intelligence on it. So far from that, so bringing it back to actually sensible predictions. I think there are things that we've done on our developer platform that put us in a position where you're going to see breakout applications built on the platform. So whether it's using AI or not, the workers platform has reached now. It's been around for a long time now. This is not new, the workers platform, but we hear the services. 17. So now I can go back and find the blog post. I think it's a wonderful description of what it was, but that platform has grown to a point where there are many applications built on top of it. People are using workers in an incredible amount without realizing that we're executing the code for them in our data centers around the world. And I think that you're just going to see some major applications get built on workers and on workers AI, because the platform has reached a level of depth and with the right, you're going to develop an advocate, some examples, it's just going to be that's going to be the one in 2025 that I think you will hear a lot more about than we've been hearing about in the past. And it's 2017, I was correct. So seven years since we actually put in place to develop a platform and so many announcements in this since then. Exactly. It's at the time was starting small, but it's clearly today day, something really big with, I think over a 2 million developers are building with Cloudflare, which is a lot. And as you mentioned, potentially like big companies will start arising in terms of. Yeah, big companies and also small companies that scale because the thing that workers, workers, let's you do workers and workers AI is unless you scale something. So you can build a small version and get a big version for free, basically. I think that's really the beginning of the year. You're going to see some companies do sudden scale ups. If you think about how fast open AI scaled, that shows you the potential of a viral application that people want to use. And Kafka can help make applications grow very, very rapidly. On that note, this is something that San Altman has talked about not only about AGI that is coming potentially, and what that means, but also the first billion company, unicorn company with one person could be just around the corner. So these types of tools, not only a eye -driven, but workers driven could make a sense of there's less people that can do a lot of things in one company, kind of type of thing. Yeah, but by the way, that's also the history of technology in many ways. So if you remember back a few years, 15 years, something like that, you heard stories about individual developers who made personal a lot of money with a mobile example. And if you go back for that, when the web came long, you heard about small teams achieving a lot because they could reach a very large market. So the AI thing is just another step up in this. And so yeah, I agree with some of them that we're going to see some very small teams. use the leverage the AI gives them to do something really fascinating and large. So yeah, where there is some 2025 or 2026, we'll see, but there's definitely coming. The other thing we're talking about AI is that we're actually going to talk less about AI relative to how much we use AI. And the reason for that is we're going to be using AI without realizing it. And so these are, I love to get people is that on your phone, you can probably go into your photo library and you can probably search by something like cat or man or house or et cetera. That's an example of your machine learning or AI embedded into something that you don't think about. You just think, oh yeah, I can do that. That's a feature. And so you won't go around saying, oh, I'm using this AI in my photos app. You'll say, well, my photos app does that. And we're going to see that across the board as they get embedded into things, little by little, without us really thinking about it, we're just going to think about it as the way computers work a bit like if you remember when Google mail came along, one of the killer features of Google mail was search. You could, it didn't matter. You didn't previously, people were, I remember really well in Microsoft Outlook, for example, making folders and trying to categorize everything so I could find mail again. Google mail just said well forget about that to search a mail. I think AI is going to be a bit like this. It's going to be oh, yeah, well obviously my system knows how to read my handwriting because that's just normal computers do that True and it was a good search so making it good and AI will help with efficiency for for even on the search level It's it makes a world of difference use you're I think finding what you you're searching. Part of the reason next year is interesting is I think there's been a lot of, in some ways, overhyped expectations about what AI is going to do in 2024, as people tried to apply it, in particular, problems. It's not an instantaneous solution, right? You need to spend some time specializing it, training it, learning how to use the AI. So I think it's now that we're going to start to see these applications arrive, and they will just just be part of our daily lives, whether it's on internet or on your phone, AI is just going to be, oh yes, computers do that. They have some intelligence built. Exactly. Apple has announced Apple Intelligence by phases, so we will be incremental in a sense. Actually, in December, OpenAI started to do deployment every day of new things. One of the latest one was a better integration with Apple types of with Siri in a sense. So there's much more integration where folks that don't know anything about AI as you were mentioning are just seeing improvements in what they can achieve and not necessarily the perspective of, oh, this is a new AI too. It's just improvements. Yeah. And the interesting question is you and I are sitting here in Europe, okay. about these things that people are going to see. And that's an interesting question on all of this AI stuff is when will we see it? Because what you are seeing with some US AI companies is that they are denying the release of their AI products in Europe. So you store that with Sora from open AI, Apple Intelligence. My understanding is it's not until April of next year that we're going to see it in Europe. It's so available. So there is a gap in our ability to use some of these AI products. So when we talk about us, we have to bear in mind that you and I may be delayed in seeing these things and other parts of the world may see them ahead of us. True. On that note, not only on the AI front, on the internet front, there's a lot of talk. This was the reflections. I just did an episode with the list of stars I can just learn from our policy. the city, all about elections in 2024, but also predicting a bit on what's coming in 2025 in terms of policy. A lot of changes because there's new governments in town, all sorts of countries. On the technology side, is that something that maybe it's a opportunity because there's a lot of changes and other things that were you a bit on the regulation, more technology side perspective of 2025? Well, I mean, obviously things with regulation tend to play out over a long period. So we don't really know what the effect of European regulation will be on innovation, for example. To what extent are people in Europe held back by knocking out access to certain AI products? It's difficult for us to know the answer to that. I think it is a bit unfortunate because what has made technology take off over, I think, of the last 20 years has been the ability for people to access technology. And so computers got significantly cheaper. There was the open source movement which made availability of software to everybody. The programmers could look at other people's code and modify it and you saw many great innovations. So we're going to see, And it's going to be interesting to see how the European AI companies respond, whether this creates an environment in which they flourish, or whether creating a barrier means that local people have less experience with this stuff and that has some knock on effect. I'm not sure what the answer is. What I think is that restricting who access to knowledge is a problem and we need to make sure we're not doing that. Make sense, make sense. Let's see how it plays out. On that note, and we're talking about AI access, there's AI needs in terms of technology, in terms of data centers. We mentioned several times during this year in the blog and elsewhere. For example, we just around birthday, we announced Gen 12, Well, the generation of hardware that we have in our data centers with the CPU level, but the GPUs that AI definitely needs also there. What is your prediction on AI needs on that more hardware AI inference latency being quick to have the information you need from AI? Well, as I was saying here, AI is going to be a bit everywhere. going to realize where it is. So some of that's going to be on your phone. Like, for example, if you look at an Apple device, it has a dedicated processor for doing AI tasks. So you're going to see things there. You're going to see these big data centers where people are doing training. I think a lot of the focus tends to be on the M, you know, a metric is building a gigawatt data center or whatever. Right. You know, that that. And nuclear nuclear is now being used. Google. They're trying it. They're trying It's going to get power, right? And so, yes, absolutely. And so, you know, people concentrate on that. But I think the real thing is this sort of distributed nature of AI. So the AI chip, this in your phone, the AI cards or chips that are on our network everywhere in the world. I think it's 120 cities now where we have GPUs. And so, it's going to be a bit everywhere. I also, I'm very hopefully in 2025 we see an emergence of some really credible rivals to Nvidia, particularly for inference. You know, as it is, we've looked at lots and lots of companies for chips that are non -in -video because we're interested in, you know, having a variety and on network. We've done that for a CPU use in the past, right? We have Intel, MIMD, and we have some ARM as well. And so we're going to do the same something that you piece, because we're running a variety of different workloads, the different sizes. And so we'll want a variety. So I'm hopeful in 2025, we see some really good examples of those kind of processes become available. You've seen some major companies and Amazon has their training chips and Google has their own chips as well. Hopefully some of these things become more widely available, whether it's for them or for a startup. I think there's room here for improvement. There's room for low power chips. There's room for chips with a large amount of memory because of the size of the models. So we're still in a very, very active phase of innovation around how models get executed on what hardware or the algorithms should look like. They're all of the work around quantization. So I think that yes, there's a focus on, nuclear power stations powering giga what data centers makes sense. Good news story. I suspect the real news story will be about over a longer term about how we made this stuff faster, more power efficient, more memory efficient, more compact. And that will lead to an explosion of uses of AI everywhere. You don't predict difficulties on that note regarding access in 2025 on too much people trying to use something that is not clearly available in terms of the power. Do you think there's resilience there? I think that the fact that people want to push AI close to the device and on device gives you an incredible ability to distribute attribute the actual inference across the world. And that distribute the load required. I also think the fact that you have a meta who've really pushed out their open source model Lama, which is essentially becoming the de facto standard for building anything else. There are other open source models, but Lama is really taken over. Means that you don't have to have hundreds of companies building these foundational models. You can take these open source models and build on top of them. And so that takes some of the pressure off the training side of things. But then let's see, there's power requirements now. And we, you know, that will that will play out over the next two or three years. You're mentioning different players in the GPU space. And we were talking about year and review Lisa Sue. So the CEO of AMD, he was named in December, times seal of the year, which was interesting in a sense that was not Nvidia specific was actually AMD. Yep, on that regard. Well, I'm not sure that, you know, Genson that Nvidia can be CEO of the year every year. That's true. I mean, Nvidia is going to shoot someone else. I mean, Nvidia is growing and incredible and incredible. Right. you know, continues to do so. But of course AMD is doing a lot of interesting things and you know, we work very close. So that we use their CPUs extensively and we look at their GPU offerings as well. Well, another perspective that potentially you can try to predict is what for you is the thing that most people don't talk about about AI that you think could be really relevant in the next year and making a change. You already mentioned incremental increases that people don't realize. But what is that thing about AI models potentially or even what you can build with it that most people don't realize? For me, the big thing is the intersection between AI and privacy. I think there's tremendously valuable if some AI thing knew everything about my life because I could ask you a question. I could get it to book something. I guess there's something I know that restaurant I went to the week before last the one with the, you know, it's near Santo, she lives been and can you book that one again? And it could go figure that out if it knew everything about my life. Right. It could look in my calendar. It could look at my email or you know, wait, have I got any packages that haven't been delivered and I've lost track of, you know, because it's Christmas and I've ordered a few things, something could go like some AI could go do that. The problem is I'm unwilling to give some third party access to all of that stuff. It's very private. And so the interesting question will be whether there are local AI models that you run on your own computer or your own phone that guaranteed that data doesn't leave. And that would unlock something pretty exciting. And that's kind of what Apple is trying to do with app intelligence. why they have been cautious about what they've released and also why they spent so much time on the privacy aspects of it because they are aware that your life is on your phone. So I think the intersection AI and privacy is race -in -ifkant and not talked about enough. An on that regard Apple actually released their vision of what's coming in future years. Those types of interconnected perspectives They're not releasing everything all at once. They're doing it step by step. No, I mean Europe, we don't have anything almost. No, we haven't right now, but let's imagine what we've read about in the press because we can't actually try ourselves. I think what Apple is doing is what they do in general, which is they want to release something that is a working product that if very large number of people can use, right? I think about the size of Apple's user base. So they're going to spend time getting it right for that large audience. And that's why we've seen what they've released so far. And I think it will be interesting to see how it plays out. So Apple may not be the bleeding edge of AI, but they'll be the deeply integrated one. Regarding attacks, we already mentioned DDoS attacks. So we had a blog post about the evolution of 10 years of DDoS attacks. And there's other types of attacks. as Lock4J you mentioned that is still a three years since it was released or announced. It's still this year from our year review, the main one, vulnerability used in terms of types of access in a sense. What is the attack's perspective that you see for 2025? And how is AI enabling So the actual thing is kind of interesting, right? Because I think the biggest thing we see is it enabling fishing, where just people are writing better fishing emails. And the automation, right? The machine learning automation perspective? Yeah, I think that's probably the, that's going to continue to be a thing. It does look like AI is getting very good at hacking anything. I went to a very interesting talk in Switzerland a few months ago about company, which is programming the AI to do capture the flag competitions. And it's getting pretty good at them, but it's still not at the level of a good hacker. So the thing is a way off before AI starts hacking for us. We're still going to have to sit at keyboards and spend long hours doing that. But yes, AI is part of the mix, particularly helping social engineer things. And so So I think we'll see more of that. I'm going to see more of the rest of the usual stuff, right? Look at that DDoS, opposed to it, about the evolution of DDoS. I mean, DDoS is not going away. You saw it 10 years ago in that blood post you were mentioning. And we're going to continue to see DDoS just grow over time, just like, I guess, big number again. Why does it get bigger and bigger? Because the internet gets bigger and bigger. There are more machines to hack, more bandwidth to write available people build botnets and they can throw more bits per I can that to you. Make sense. Actually, the block was here, mentioning is this one bigger and better, how do you also attack sizes that have all over the last decade? And it's quite the evolution. As we mentioned, often it's a easy way for attackers to do attack sometimes. It's with other vectors of try to access. They use this one and others. But typically this is a very much used one for sure. And there's this perspective. Yeah, in terms of these. Oh, well, the actually the it's 5 .6 terabytes per second attack in October that we we saw. Terabytes for us be terabits. Terabits. Terabits. Yeah. And it was 309 gigabit. for a second in 2013. So then you just do a linear, linear prediction. Yeah, I mean, these sizes get bigger as the internet grows with. As you were saying in the radar report, the internet's grown by about 17 % this year in terms of requests. And we've seen like 20 % gross year on year on the internet. And so, you know, the internet's just growing. And basically, beginning to that, at bigger attacks. There's over the years, IoT Internet of Things perspective and healthcare and very specific industries that has over the years picking up and attacks on those are known and coming and those also a wider perspective of adoption in terms of zero trust architecture for companies. Do you, what do You're a prediction on those two notes. Zero trust architecture adoption and use and be secure with us. Zero trust is just the way things are going. We already do, all of us do zero trust all the time. We just get our phone out. We get our laptop out. We connect to some network and we log into services. And we are authenticated by that service. Zero trust is the corporate version about world. Everyone's already living in where these can control access to their internal resources. That's an architecture that's here. It's still not widespread. I mean, this is still a growing world. So there's a little bit more of it. And eventually companies that aren't doing it will realize how much it costs to maintain a corporate network with all the firewalls and all this kind of stuff and how inflexible it is. So I think that Cloudflare Zero Trust connectivity cloud approach is how everything is going to be architecture in the future. but for large companies, it takes time. And so they'll just be more of it over the next year and the year after and the year after. And on the IoT space, do you expect any changes, expansions on that regard? Well, I think we have got past the internet toaster madness. Lovely. I think people are doing There are a cluster as a smart host. I think that there was a mania for making everything smart at some point. And I think that a lot of people like things to not be smart because they just kind of work. Speaking of someone who has a home automation system, sometimes I wish I had copper cables with actual 240 volts flying through them like it just gets on or itself because sometimes it's smart can be painful. And so I think that the IoT devices that are out there are things like smart cameras and DVRs, weather stations, personal devices for doing health monitoring. I mean, these things are just going to be part of how we do stuff. Those are internet connected. But hopefully the, I really hope the internet toaster has died at death. I agree. I'm not a fan. But I remember a time where everyone was talking about that, it seems a bit of AI. Everyone was talking about that now people are using me that be watches or other things to monitor what they need to monitor, what is for them to monitor. And it's not as talked about still. Well, you know, I've actually gone kind of backwards in this. This is me, but it's a watch, right? With the watch, right? So I had Apple watch for a long time and I was getting all the notifications and I was getting all the health stuff and it was fascinating me and then I bought a Casio. Old school. So I'm wearing a Casio watch. It's a quite a nice Casio watch. I think it costs 200 euros maybe. So it's a high end for a Casio. But it doesn't have to. It's not smart. It's not smart. It has no internet connectivity. It has tells me the time. And it's sets itself from the radio signal. There's been around forever and it charges itself up as solar and I never have to think about it. And the only time I put my smartwatch on is the gym. And then it's interesting. Did I get into the heart rate zone for long enough, et cetera? Then it seemed to work out. You have a specific use case for it for it, which I like. I like the health stuff. I just, I don't want to think the buzzes on my wrist all the time. You mentioned already the need of thinking of energy efficiency, carbon footprint. That is something you think for 2025 will be quite important, not only for AI, but in general, for the internet. You know, carbon emissions for every customer are at carbon impact. How much carbon they say by using our service, because it's not offloading their servers, and we can be extremely efficient as we talk about in the gen 12 announcement, we tried to be more and more efficient over time. I think it's the carbon side of things is just one of those things that everybody is. So think about as part of how they build stuff, we have that, we have our green compute option, which is we'll only run your service in a data center with green power. So there are options there. Sure. And the future seems right there in terms of changes for what's coming. Well, this was great, John. And we have a nutshell, some predictions and some also summaries of what was 2024. Yeah. Yeah. Yeah. That's around inflation. Really, that AGI, maybe that will mean a lot, according to Sam Altman, a lot in terms of changes in our humans daily lives. Things are changing for sure. All right, as talking to you and as Bob Dylan would say, uh, time to our changing. They always are. And happy holidays for everyone that are selling us happy holidays, everybody. And, uh, happy new year. Exactly. It's coming. That's a wrap.