All right. Thank you everyone for joining us. End of day one. I know we're about 35 minutes away from the live music, and dinners, and drinks, so thank you so much for sticking around for this last session. I can assure you, it, it'll be entertaining and informative. And it's my pleasure to introduce you today to Matthew Prince, CEO of Cloudflare. If you don't know him, I'm excited that you're here, and thank you so much for coming.
Thanks for having me. I feel like we're way back away from everybody, so yeah, here we are.
Sitting back. For those of you that don't, maybe they're not familiar with Cloudflare, why don't you just start with just a quick overview of what Cloudflare does? You introduced a new tagline called Connectivity Cloud. Let's talk about the new tagline and what that means, and what your mission is.
Yeah. So Cloudflare today runs one of the world's largest networks. We have presence in over 300 cities worldwide, where our mission is to help build a better internet. And when we talk about a better internet, that means an internet which is faster, more reliable, more secure, more efficient, and more private. And so that's what we're always looking to do. And, you know, at some point, people would ask us, like, "What- what are you doing?" 'Cause we launch a bunch of products that, in various ways, either compete with or complement the big hyperscale public clouds that are out there. And I think at various times people have said, "Oh, Cloudflare is the fourth, you know, public cloud," which actually I think is kind of insulting to the folks at Oracle, and IBM, and Alibaba, and Tencent.
But, it never quite struck us as exactly right, and I think what we see is... fundamentally, what we're doing is different than what they're doing at someone like an AWS, or a Google Cloud, or a Microsoft Azure, where if you look at someone who's on the product team at one of those companies, the core KPI that they're measured by is how much of a customer's data are they holding onto, have they, have they captured at some, at some level? And so at some level, they are almost like captivity clouds, where they're trying to kind of get all of your data in, in one place. And again, there's a role for that, and they've done some amazing things in what they've built.
If you talk to a product manager at Cloudflare, what they are measured by, the KPI that we pay attention to, is how much are we connecting various things together? How much are we making it easy for any device, anywhere in the world, for any cloud to any other cloud, to make it easy for that data to flow between those things and give a consistent control plane, consistent security, consistent availability, reliability, make it as efficient as possible? And so in that sense, we realize that where the traditional hyperscale public clouds are really like captive clouds, we are much more of a Connectivity Cloud.
Those two things actually work extremely well together, but I think that when you start from that framework, it helps frame how we approach a lot of problems differently, and how we have customers that, you know, are taking advantage of a bunch of the big hyperscale public clouds, but using us as that connective tissue that makes sure that everything works seamlessly together.
Well, it makes sense. You free up, you free up data, and you allow, you know, organizations to use the data across multiple clouds and release the captivity, I suppose.
Yeah, and I think that what we're seeing with, you know, everyone from small innovative startups up to the biggest financial institutions in the world, is that no one is single cloud. Everybody is multi-cloud in one level or another. You know, at Wells, you guys use Microsoft, but you also use Google. And the question is, like, if you're using these two things, and both of them are trying to say, "We want all of your data to be held captive in that," it turns out customers don't want that. They wanna be able to take advantage of each of those different bits. But if you're gonna do that, you need some level of a control plane which is consistent across it, and what does that control plane need to do? It's gotta have security, absolutely.
It's gotta be able to be performant, to make it so that you don't suffer some sort of performance penalty when you're using both Microsoft and Google. You've gotta be able to be as efficient as possible, drive down costs, help you save on that overall spend that you have with those hyperscale public clouds. And so oftentimes, when someone will sign up with Cloudflare, we'll be able to reduce the cost and the spend that they have on their hyperscale public clouds, and that's oftentimes where the dollars that we, we capture are. And then you wanna have that consistent across all of those things. And so I think that as the world becomes more complicated, it becomes more and more important for something like Cloudflare to be that unifying fabric that connects all of the different networks together.
That makes sense. You know, one of the, one of the things I was excited about having you, having you here at the conference to talk about was, kinda goes back to, your presentations earlier in the year at the Davos conference. You made some pretty bold predictions at that conference about, you know, predicting that the, the, you know, the, the market was heading into, you know, towards a recession, and obviously, it turned out to be a very accurate prediction, at the start of the year. So I'm, I'm... I was really excited to have you here, just be, you know, to hear your thoughts on-... how, how that's played out over the last 11 months, and, and where do you think-- what, what are your predictions for next year?
So I will quickly probably defer to a lot of people in the room who are a lot smarter on how the market looks. But I think from our perspective, we see some things that are really kind of interesting in terms of our view into the universe. So today, somewhere between 20% and 25% of the web sits behind us. And so, for example, in the last couple of days, as we've been going through both Black Friday and Cyber Monday, you know, we saw at the peak on Cyber Monday, we actually just published a blog post, where we had over 100 million request transactions per second that were flowing through our network.
And we can look at that, and we can say, slice it in a number of different ways to say, "Okay, here's a set of customers behind us that are, you know, consumer product companies. Here's a set of customers behind us that are financial companies. Here's a set of customers behind us that are, you know, B2B customers." And we can look at what are the patterns across all of those things and what's happening in terms of spend and in terms of how people are interacting. And I think, you know, starting in, you know, December of 2021, late 2021, we started to see that, you know, that go-go time that was happening for quite some time, felt like people were starting to pump the brakes a little bit, and things were slowing down.
I don't think that showed up in a lot of the financial results until quite a bit later. But that was that early indication, and I think for us, we are always looking at that data and being able to make investments across our business. So I think what we saw at the beginning of this year was, it wasn't gonna get a lot better, and I think it's been a really tough year across the space. I think that from our perspective, it feels like it's not getting worse right now. It feels sort of like it's stabilized, but at the same time, it doesn't feel like it's dramatically rebounded off the bottom.
Although, we did see, at least in consumer spending, that this was an uptick, a quite substantial uptick over last year in terms of the transactions that we can see. That seems like that's a positive indicator. What that will turn into in terms of B2B spend and how that affects other businesses, I think that's a harder thing for us to totally have our pulse on. But I think that it feels like it's not getting substantially worse out there. The beginning of this year, I think Q1, we were just, you know. Again, I think we were surprised by how much the sort of B2B spender, the IT spenders, really pumped the brakes.
If you look back at that, I think people were just super skittish with the collapse of SVB and some of the other banks that were out there. Everyone was waiting for what's the next shoe to drop, and especially in March, you know, the economy, and especially in the IT space, really ground to a halt. I think right now, again, it feels like it's not as bad as that. It feels like it's kinda flattened out, but I think people are still really skittish. There's a lot of things that you can look out in the world and say: What is it that might make the economy worse?
It's harder to point at those things and say, "That's the thing that over the course of, you know, the next 2, 3 quarters, is gonna make the economy substantially better." So I think we're kind of in a grind period, but in some ways, that helps us as a business because we can come in, we can say, "We'd help you..." We have a broad platform that we can consolidate vendors behind. We have ways that we can help our customers save money, and security is one of those just enduring things, that nobody wants to cut their security spend.
Well, you know, I think the traffic patterns you're seeing on Cyber Monday and Black Friday aren't necessarily reflective of a macro change, but it does seem like more of a permanent change in buying patterns from consumers not going into stores and buying things online now and leveraging the cloud and buying things, you know, on B2B sites like that are hosted on Cloudflare.
Yeah, I mean, we'll see how that turns out. But we're super proud that, you know, Shopify is a customer. They were powered on our platform. You know, Salesforce and their Commerce Cloud is a customer, they're powered on our platform. And so a huge amount of the online commerce that's out there relies on Cloudflare's network, and we're able to continue to, even in these times that are, you know, huge demands on our network, deliver incredible value to our customers and make sure that they can shine, and the customers can get whatever it is that they're buying at these times.
That's great. So, you know, next topic I wanna talk about was on the AI side. So we started the day talking about AI and using generative AI models. And I know you've, just from talking to you in your meetings today, you've had a lot of discussion on AI, but you, on your earnings calls, you've talked about really delineating the difference of how customers are using Cloudflare's network for AI purposes, more for inference versus training-
Mm-hmm
... models. I'm just curious if you could, you know, talk about, you know, why that's important. Why, why would they only use Cloudflare for inference, and what, what is the difference between training and inference?
Yeah. So AI really has three big steps. The first step is, you take a huge amount of data, and you pour it into a bunch of machines, and those machines crank away for a really long time, taking that data and reducing it basically into what is a very kind of fancy math equation, huge, complicated math equation. And then that creates this thing, which is a model. And in the financial services industry, you've got models, those models can predict all kinds of different things. It's not actually that much different than what that is doing. But you're taking a bunch of data, you're crunching it, and that's the training part of the AI space. And that's where a lot of resources are going into right now. A lot of people are experimenting with that.
If you look at, you know, what OpenAI is doing, what others are doing, they are building those initial foundational models that are out there. And the right infrastructure to build that is a big warehouse somewhere where power is relatively cheap, and where you can put a whole bunch of machines right next to each other and essentially have a giant supercomputer that's cranking out what that model is. That is not what Cloudflare is. That is much more what the hyperscale public clouds are, where they're, you know, even in AWS, they have a handful of these availability zones. But what that means is they've got a bunch of buildings close together, and they've got thousands and thousands and thousands of machines all right next to each other.
We have thousands and thousands and thousands of machines, but we're spread out across thousands of data centers all around the world in 120+ countries, 300+ cities around the world. So we're not really the right place to do that training, at least running on our infrastructure itself. That said, I think our opportunity in the training space is in helping that big set of data go to wherever there's enough resources today. Today, there is a real constraint on finding buildings full of GPUs that are out there. I'm trying to sort of give my best impression of Jensen, wearing my leather coat here, but, like, there is a shortage of GPUs today.
Our short-term opportunity in AI is because we're the Connectivity Cloud, we make it really easy to take data, not just from AWS East to AWS West, and do it without paying an AWS tax. AWS actually charges you to move data from one region to another. We don't. But we also make it easy to go from AWS to Microsoft, to Google, to Alibaba, to Tencent, to Oracle, to IBM, and basically chase around the world to wherever you can get the cheapest GPU prices.
And so if you look across the top AI companies that are out there, almost all of them or not, that's actually an overstatement, but we believe we are the, the most common cloud provider across all the major AI companies, and one of the big ways that they're using us is: how can they go find the cheapest GPU resources at any given moment in time? I think that that's an opportunity of the moment. I think it's an opportunity of right now. I do think that over time, there's not gonna have the same shortage of GPUs, and so being able to do that probably is not a huge business over, over the extremely long time.
But I think that same ability of being able to move data and do it efficiently and inexpensively, connect it, that's gonna be a big opportunity for us, and AI is one of the first places that's really manifested itself, really well. The second opportunity is once you've got that model built, you've cranked all the data, the model has gone from really big to really small. We believe. Then what you wanna do is use that model to do interesting things with it. So if you have an iPhone today, if you go to the Photos app on your iPhone and you search for passport, or you search for cat, or you search for dog, you've never tagged your passport in your photos, and yet your iPhone will actually pull that up, right?
Or it'll pull up a picture of a cat or pull up a picture of a dog, even though you've never tagged that. That's because Apple has gone and they've crunched a whole bunch of data on a whole bunch of people's passports and, and cats and dogs, and all kinds of things. They've loaded that model onto your phone, and then that model is able to actually take a new set of information and use it in order to do categorization. And we think that process, that's using a model on new data, that's called inference. And that inference process, we think, is gonna happen in three places, but the first two are the most important. So the first one is gonna be on the device itself.
And so, you know, people would say like: "Well, you know, maybe there's an opportunity for Cloudflare and things like driverless cars." Maybe, but I think if you've got a driverless car and it's driving down the street and there's a ball which is bouncing out from a yard and there's a young kid chasing after the ball, you don't want the car to have to go up to the network to make the decision on whether to slam on the brakes or not. You want that inference to be done on the car itself. So a lot of things are gonna be done on the actual device, on the actual... whether it's your phone or your car or whatever. But some models are going to be too big to be able to work on the actual device.
You know, I have no doubt that in 10 years you're gonna have ChatGPT-level quality on your phone, you know, effectively GPT-4 on your device, but by that time, we'll be up to GPT-20, and that's gonna be too big to run on your device itself, and it's gonna be better in some various ways, do something that the thing on your phone can't do. So the next best place is to have that inference run as close as possible to where you are. And that's better than going all the way back to Ashburn, Virginia, for two reasons. The first that everyone thinks about is it's just faster. Speed of light's only so fast, and having to send your data all the way across the world and all the way back is inefficient and slow.
So being able to answer that query as close as possible is actually better. And then the second reason, which is actually more important, is around the world, we're seeing all of this regulation around AI.
One of the things that governments around the world are insistent on is: how with AI do we not make what they see as the same mistakes that we made with the Internet originally? Which is today, too much of the Internet is concentrated, from the perspective of the global audience, is concentrated in the United States. Too much data flows back to the United States. How do we make sure that's not the case? And if Cloudflare isn't successful in this—today, 95% of the AI GPU resources that are being deployed are being deployed in the United States.
And so as I talk to leaders in Europe and Asia, all around the world, they're saying, "We wanna make sure that as our citizens are using these AI models, that they're distributed globally, and that the data about our citizens stays as local as possible." And so to that extent, we think that the second-best place for an AI model to run is in a network which is as close as possible to that end user, and that's what we've built at Cloudflare. So we've got, today, in over 100 cities worldwide, GPU resources that can run new models that you load to them, existing models, whether that's Llama, Stable Diffusion.
Any of these models that you've heard of, we've already preloaded them onto it, and they can be run locally, faster, more privacy and regulatory compliant, and then also, for a number of reasons, in a much more cost-effective way than shipping it all the way back to one of the traditional hyperscale public clouds. That's the second part of AI. The third part of AI is another piece, which is actually called fine-tuning, which is, how do you take data and take this generic model, GPT-4, and then make it yours? And I think our opportunity there is actually probably our biggest superpower because for our customers, their data is already flowing through our network.
So let's imagine that you are, you know, I don't know, a B2B software company, and you've got a knowledge base which is out there, and that knowledge base is connected to various products and features that you have. Cloudflare is already seeing as people interact with your dashboard and with your knowledge base, and we can see when somebody gets confused on a page, and they click the Help button, and then they go to a knowledge base article, and then they go from that one to another one to another one. And so what we can do is actually take that data, which is already flowing through our network, and then feed it back into those foundational models in order to effectively make them your own.
So if you wanna take ChatGPT but make it really smart about your B2B software company, Cloudflare makes that easier than possible to do that refinement and make that work, specifically for you and then run across our own infrastructure. So I think we've got a short-term opportunity around the opportunity around training, helping people move data to wherever there's excess GPU resources. Over the longer term, I think we're the sort of second-best place to run any API task, and the best place to run the most complicated tasks. And then we can help customers because their data is already flowing through us with the weights and the data that allows them to fine-tune those models and make them the most powerful for their own business. That's how we're thinking about the AI space.
It's a very comprehensive strategy. It sounds like you may have missed it this morning, but we had a guest speaker talking about AI and, you know, he had mentioned that, you know, I think more going back to your second point of keeping the models locally and you know, keeping the data close, building out this entire infrastructure stack that revolves around NVIDIA GPUs and MongoDB databases, vector databases, and the whole infrastructure stack that you need just to keep that data local seemed really complex and really expensive. And I was thinking, like, Cloudflare, which has Workers AI, offers the entire infrastructure stack as a service, which already embeds NVIDIA GPUs into it. So do you think.
Is that, is that what you're referring to, eliminating the need to go buy all this infrastructure just to run-
Yeah, we should just provide that for you. I mean, there's no way that if you're a global company, you can turn up infrastructure in, you know, everyone, everywhere from Luxembourg to, you know, Lagos and, and make that work, whereas we've already got that already there. And today, we're in over 100 cities worldwide. Our aim is to be in every single city where Cloudflare has presence, to be able to offer you that, that infrastructure. And so we've built that and made it available, and moreover, we've incorporated into the ecosystem in ways that, that make that work. So we've partnered with Hugging Face, for example, so that you can take any AI model that's on their, on their platform, and just with one button, deploy it out to Cloudflare's network and be able to, to run that extremely efficiently.
Then we're building out the tooling, so we have our own vector database, or we can plug into any of the vector databases that are out there. That's what you need for fine-tuning, that is, helping you get better and better over time. So that infrastructure exists. The secret to what we did, though, and what made this so successful is this wasn't something that we imagined, you know. We really started rolling it out a couple months ago, but we've been planning for this for at least the last six years.
And I remember, actually, if people have been following really closely, about 2.5 years ago, we did an announcement with NVIDIA, where we said, "We're going to be rolling out GPUs across the whole thing so people can do inference." And it was sort of a test balloon to see what would happen, and the answer was crickets. Like, there were a handful of things where people were like: "We want to use this, we don't want to use this," but the market just wasn't there yet.
But we learned a lot in that 2.5 years of what it was that people needed in order to really have an ecosystem where they could be valuable, and we were constantly looking, and as we deployed hardware, we were reserving space in that hardware to say, "This is where we're gonna put a GPU to be able to do this inference in the future." And so where if somebody traditionally has had to build out that infrastructure, they've had to buy the entire new AI server and deploy that.
We've actually been able to sort of say: "We're gonna take the existing network that we have and deploy GPU resources in that network." And so I think we've been able to do it in a much more cost-effective way, where, from a CapEx perspective, even though we already started rolling that out in Q3, it didn't have any meaningful impact on our, on our sort of forecast CapEx as a percentage of revenue, which, again, I think I'm really proud of our team for having the foresight to get in front of this and making that investment at the right time.
... after you've rolled these, you know, the service out, have you found any? I mean, a lot of people talk about using H100s as a critical component of, you know, to support an LLM. But is there— do you need H100s for every single model out there, or is there— can you downgrade any?
Yeah, it turns out that what sort of how much you have to be at the bleeding edge, or maybe a better way of saying it, which branch of the AI tree you wanna be on differs depending on what you're doing. If you're doing training, then it turns out having the latest, greatest GPU, the H100 or H200, which they've announced, but isn't coming out for some time, that makes a bunch of difference. When you're on the inference side, it turns out that it really depends on the model, what you want to use. And in some cases, there are models that run really well on CPUs, not even needing GPUs. On the other hand, there are some that need faster GPUs, some that need different flavors of GPUs.
And so we've got a great relationship with NVIDIA. We've partnered with them for quite some time. We have deployed their cards across it, but we don't think that our customers should have to think about what that underlying infrastructure is. When somebody uses Cloudflare, when they use Workers, they don't know if they're running on an AMD or a Intel or a Arm-based processor. We abstract that all away. We think the same thing is true from a GPU perspective, and so we'll have all different flavors of GPU resources that are out there. And we love what NVIDIA is doing there. It's a clear leader that's out there, but Intel's doing a bunch of really cool stuff in terms of what's coming out. AMD is doing a bunch of cool stuff. Qualcomm is doing a bunch of cool stuff.
Even folks that are a little bit outside the norm, like Apple, are doing some really interesting things, and we're staying on top of all of that. We think that over time, there's going to be different underlying silicon, which makes the most sense for different models, and our job is to optimize that and build the scheduler that gets the highest level of utilization across it, depending on what's the actual task that you're doing. What we wanna do is not only give you that best performance and give you that ability to, if you need it, run an AI task as close as possible. But separately, you—and the way that we price our AI model is there, there are sort of two different versions. There's the fast twitch neuron, right?
Which runs as close as possible, as, you know, you're, you're put right to the front of the queue. But then we have what I initially tried to call the slow twitch neuron, and our marketing team quite rightfully said that's not the right name for it.
So we have the regular twitch neuron, and what that does is it says, "Instead of you being right as close as possible, as fast as possible, let us figure out where in the world is the most optimal place for it to run, and in exchange, we'll price it in such a way that it's the most cost-effective way for you to do inference anywhere." And since we can move data across our entire network incredibly efficiently, that has allowed us to say, "You can run an inference task, and if you don't care where it runs, we'll find someone in our network where we've optimized it to have the best and most cost-effective performance." And then we've priced that in a way that is, you know, blows out of the water any of the other sort of inference tools that are out there today.
It's interesting. I also just wanted to wrap this kind of line of questioning up. You have, obviously, a lot of partnerships to build out this whole service with what you mentioned, Meta and LLM, all, all the different models, a lot of different, sounds like hardware providers, database providers. But the one that struck me as kind of interesting was, you have a partnership with, of course, Microsoft as well, too, which is, is a competitor at the same time. So, I mean, is there any risks to that partnership?
No, I think... I don't see them as a competitor in this space, and I don't think they see us as a competitor in this space. I think that they see that there's gonna be a different role to play for different things that are out there, and some models are gonna make sense to run on your phone. Some are gonna make sense to run on a network like Cloudflare, and some are gonna make sense to run back in some centralized, traditional, hyperscale public cloud. And so what we could do with Microsoft would actually help them have another place where they could have their models run.
And what Microsoft then could provide to us was they've actually built a technology which, you know, it is, it's sort of like the, the rail lines, the rails for the AI universe, where you can take a model, and it can be on your phone, and then it can easily, without the user having to think about it, transition up to the network. And then if for some reason we're not the right place, transition to the centralized public cloud and, and move seamlessly between that. They've built a bunch of tooling around that, and as we evaluated who are the right tool providers there, Microsoft was the clear leader in that space.
And so I think what we're trying to do is very much aligned with what they're trying to do, which is make it as easy as possible for models to run wherever the best place for them is. Some of those are gonna be on the phones. Microsoft doesn't have much presence there. That's fine. They wanna make it easy to go from the phone to the network, to the centralized public cloud, and back and forth. And in that sense, I think we're a, we're a terrific partner, and we offer something that they just don't have and would be very, very difficult for them to build.
That makes sense. You know, I think that AI is certainly an interesting use case. I know you've been working on it for six years, or a long time, but it really was just introduced, and it's got a lot of legs. We're just starting. But another use case built on the same exact network is Zero Trust and SASE. I wanna ask about that as well, too. It's still leveraging your exact infrastructure, your advantage of the network, but how do you see Zero Trust in companies that are trying to deploy, you know, a new architecture and moving away from that legacy network infrastructure to Cloudflare?
You know, I think and same thing is true with AI, same thing is true with, you know, almost every product that we build at Cloudflare. We create our own problems, and then we solve them for ourselves as Customer Zero, and then those are the things that turn into products at Cloudflare. And so Michelle, my co-founder, hates history lessons, but I was a law professor previously, and kinda like history lessons. So the history lesson of Cloudflare is, you know, we started and launched, and for the first seven years, we were really focused on a set of products that were all about, you're putting something online, and we need to protect the front door of your business.
And so, you know, that meant that we had to launch products like DDoS mitigation, web application firewall, rate limiting, load balancing, API protection, bot management. All of those things were very much around good things are trying to come into your business, bad things are trying to come into your business. We wanna give you the best experience for the good things and keep the bad things out. What happened as a result of that, though, was, you know, we woke up one day and we had, you know, the State Department and the FBI and the Central Bank of Brazil and, you know, some of the biggest financial institutions in the world, some of the biggest health providers in the world, that were using us, and all of a sudden, every bad actor on Earth was trying to break into our systems.
And around that same time, Google had released a paper on what they called BeyondCorp, which was this idea that the traditional model of kind of a castle and moat of security didn't make sense. It was really born out of their own experience of being attacked largely by the Chinese state actors in the Chinese government. And they said, "We have to rethink how security works," and they built a different model for themselves, and we were close with that team, and so we were trying to learn from them.
At the same time, we were going out and evaluating everyone who was in the market, and our team would look at them and be like: "Listen, they've got the right idea, but they've got the wrong implementation." Time after time after time, whether it was from a security gap that was out there, whether it was from some sort of performance gap, whether it was just because it wasn't built in a way... Like, almost all of the other Zero Trust vendors that we talked to actually slowed things down, and we're like, "Our developers aren't going to put up with that." I mean, I think at Wells, you guys, you still use some VPN from who knows when, that...
Everyone in the audience who's from Wells is laughing, 'cause, like, I don't know how you all put up with that, because our team would never tolerate that. And so we, at some point, you know, we're not, we're not radical in this. We don't think we have to build every solution out there. We worked with a lot of great vendors that are there. We're, you know, we're not the people that build the endpoint solution, endpoint security solution. We love CrowdStrike. We use them all the time. We're not, you know, there's a lot of things that we don't think we have to build, but in this case, we were like: We don't trust anyone else. We have to build this ourselves. And so that's how we started this journey. As we said, we're gonna build it.
We had the network already, and it turned out that the way that you pay for network resources is you pay for the greater of in versus out, and all of the services we were providing already had a lot of out, but not much in. And so all of a sudden, we were like: "Oh, well, we can add this, and it doesn't actually even change our underlying cost infrastructure." And we deploy it, and our team was like: "Wow, this is way better than those other things that we evaluated out there. It actually doesn't slow things down. It speeds things up." And we did things to make sure that it just worked extremely well. And as we would demonstrate it to people, I would literally just pull out my phone if you want to see this.
Like, I pull up our wiki. I'd be like: "Let me show you how it- how I access my wiki from a phone." And I'm like, click into it, and it's got posture authentication on my phone. It's got, you know, an integration with an identity provider. It's got, you know, two-factor authentication enforced on even applications that don't have two-factor authentication built into it. It's got a centralized kind of repository. And I'd show it to customers, and they'd be like: "Whoa, that's cool! I want that." And so that was the point at which we said: "Oh, we're gonna, like, go and build this and sell it to other customers." And it's been, I mean, successful beyond, you know, our expectations, where we can go in and sell it...
And the other thing which is powerful is, again, because we pay for the greater of in versus out, you know, if you look at, like, a Zscaler, we have enough capacity, we'd add at least 10 Zscalers to our network without changing the underlying operating costs of our network in any meaningful way. And so in that case, we can come into a customer and say, like, the ROI is really incredibly substantial, and moreover, we can bundle the two different solutions together. And so if you think of the products we built for the first seven years of our life as protecting the front door of your business, the products we built for the second seven years or the first, the second, you know, 10 years of our life is about protecting the backside back door of your business.
But as one bundle, we can come in and say, "Yeah, you get WAF and DDoS and rate limiting and WAN optimization and your own network and you know all of the access and gateway and other controls that you need, CASB," and that's all one bundle. And by the way, it's cheaper than it would be if you were buying it from any of the other vendors out there. More and more customers are saying that's an incredibly compelling value proposition, and we keep winning customers away from a lot of the kind of first-generation Zero Trust vendors that are in the marketplace.
... I mean, I think it's a matter of perspective. To your point, you know, if you're a user like Wells, and you're used to waiting five minutes to connect to your network, and you don't know how slow that is, it's not slow until you realize -
Oh, I mean, I-
It could be faster.
... It can be. I'm tempted to just... It's too far away, but find me afterwards, and I'll show you how quickly, you know, we can pull up any resource. And for me, you know, I remember we did a convertible debt offering, and I was in Rwanda as we were going through this. And I remember that we had to log into the various banking systems of various bank providers that were helping us with it, and one of them used our solution, the other one used, you know, a much more traditional VPN.
And I just literally couldn't get into the one, whereas the other one, it was fast, it was incredibly... It was it just worked. Because we've got a presence in Kigali, and we can do all of, all of the sort of everything we need to do there and from a security perspective, and then put it on, you know, our own backbone, which runs back across, and it's more secure than you can get across, you know, just about, you know, any MPLS connection that's out there. And that's an incredibly valuable ability for us to deliver real value. And again, it makes me and my ability to do my job anywhere in the world, you know, or even from my phone like good luck with logging into the, you know, Wells VPN from your phone.
Like, that's something that we can do extremely well, and that is what more and more of our customers are just coming to expect. And so we're just a lot faster than whatever your traditional VPN is, and we're a lot faster than any of the sort of first-generation Zero Trust providers that are there, with the same level of security, in fact, a better level of security, that they can provide. Most of the sort of first-generation Zero Trust vendors use us as a vendor for their DDoS mitigation, for example. And so I think that shows you we can do more than they can, and that's, like, a great way... I think over time, Zero Trust is just a feature of the overall network security product, and that's exactly what we're offering.
Well, it sounds exciting. We've talked a lot about a lot of different use cases you have, a lot of interesting growth opportunities going forward. I know we're out of time here, and everyone's gonna get ready to go to the next events, but is there anything you can share with us in terms of 2024 predictions? What are you excited about next year? Whether it pertains to Cloudflare's growth drivers specifically or just, you know, spending trends next year that you think are interesting-
Yeah, you know what?
Changes in tech.
So, I think we're in for a bit of a grind over the next little bit. Again, as I look forward, it's not clear what the event is going to be that kind of restores optimism in the space. And again, at risk of being the downer before drinks... Like, the election makes me really nervous. Almost any outcome in the election makes me incredibly nervous. We've gotten to a point where there aren't a lot of institutions left that anyone trusts anymore, and I think that has set in place a real, you know, set of kindling that, with the wrong spark, could go in a bunch of bad ways.
I don't care if you're a Democrat or Republican, I don't care who you're ultimately gonna vote for, you know, at some level, all of us depend on having some trust and foundation in some underlying system which is out there. So I'm proud of the fact that our team, you know, we volunteer our resources to any state and local officials that are helping administer elections. More than half of U.S. states rely on Cloudflare today in order to protect their elections infrastructure. We're doing the same thing for campaigns, and again, totally across the board, anyone who is running for any office can get our services to make sure that their campaigns are protected. And I think that it's that work which can hopefully restore some foundation in trust that's out there.
Because, you know, I, I've had the opportunity to meet with a bunch of the, the people who, you know, have really thankless jobs right now, who are local elections administrators, and they're getting... I mean, they used to be afterthoughts, now they're getting death threats all the time. They're worried about their family. They're oftentimes saying, "You know, this isn't worth it." And so I think us being out there and saying: We're gonna provide our services, and we're gonna help you out, 'cause it's basically you against the GRU. And this isn't like a... This is like someone running, you know, Polk County, Tennessee, and it's them versus the GRU, and I think that we can help hopefully even the odds there and help those folks out. The one note of optimism is...
There have been a lot of head fakes in this space, but I am hopeful that the change in tone towards China maybe is positive. And that's something that we're watching closely, and we've seen... We watch sort of Chinese hacking attempts, and how that changes over time. When Obama and Xi struck the sort of we-don't-hack-each-other accords, you actually saw Chinese hacking drop. We haven't seen that happen yet, but I think that's something that we're gonna be watching closely, and if there really is a shift in policy, that feels like the only kind of positive macro sign that I've seen in the last, unfortunately, in the last little bit.
But it could be a really positive one if essentially the world wakes up and says, "Listen, there are a very small handful of nations that are rooting for absolute chaos that are out there. There are a couple of us that should be the adults in the room, that should be rooting for stability and predictability." And China and the U.S. actually have more aligned in terms of caring about stability and predictability than just about anyone else, and if the agents of kind of calm can prevail over the agents of chaos, I think that would be that's the optimism that we might need in the world.
Well, I'm also optimistic about your chances of getting rid of our VPN and-
That'd be... I think I have a dinner-
Good luck with that.
... with your IT team next.
Yeah, I know you do.
Good luck, though. I will say that you are the only bank that was on the cover of our IPO that has not yet become a customer, so, you know, no pressure.
Okay.
Anyway.
I wish I could help.
Yeah.
Thank you.