The Moat Is Where Your Data Lives

Transcript

[applause] >> I have a question, easy one to start with. Who has tried Fable 5 already today? Has anyone had chance to try Fable 5? Do you know what Fable 5 is? Anthropic came out with their new model, it's like Mythos light. I tried it, it's very good. Okay, another easy question. Who has heard of Filecoin? Can you raise your hand if you heard of Filecoin? Okay, that's good. Who has thought much about Filecoin in 2026? Can you show hands? Okay, great. Couple of people are really thinking about it. Well, I'm going to try to get you thinking about it a lot more in 2026, but we're going to mostly talk about big macro forces that are influencing all of the AI and infrastructure buildout. So, a little quick history lesson and we'll piece all of this together.

Historically, 10, 15, 20 years ago, the compute went was very, very cheap and it wasn't really the part of the stack that people thought about. You put the hard drives where the people were in Northern Virginia or in the Bay Area, and compute was not that big of a deal. You buy some chips from Intel, you buy some chips from AMD, no problem. But now we also have and we all know that that's not true anymore. Obviously, compute is the thing that everyone is talking about. Maybe the underappreciated part of the story is storage. As we think about the what is happening with AI over the next 5 years, we're going to be collecting a tremendous amount of data about the world. Data that in the past we've never, ever wanted to collect because nobody could

look at it, nobody could do anything with it. But if you're thinking about neurotech and robotics and course you've seen this with self-driving cars, there is so much data to observe in the world that's relevant to AI training, also AI inference. And if we really want to have AI permeate everything about modern life and actually progress modern life, we have to capture data about that. We also have to process data in real-time about that world that we live in. And so, when you think about what we've actually been capturing over the last 30, 40 years, it's a tiny, tiny fraction of the data that makes sense to capture in the AI era. And we have the only the capacity to capture a tiny, tiny fraction of that. We need much, much more storage capacity, and storage

is going to become a much, much bigger part of the story in a multiple different ways. So, this is the sort of entire story mixed together. In the past, the data was and the and the computer were kind of the same places because the data was the thing that was heavy. You brought the computer to the data. Now, as everybody has been talking about in the markets, the drive is to bring computer to where the power is. You have to find the power to actually power these GPU clusters, which means you're going to places like Wyoming, northern Norway, Finland, Ohio. Places that traditionally were not the interconnections of the internet, where computer and storage were located. And so, now all of that power is going to these places in the middle of nowhere,

and the storage is not there. So, there needs to be a solution for that. And also, it the power is all over the place. It's not just in one place. It's not just in 10 places. It's in 200 places, or maybe even a thousand places. And that's the future of where data centers are going. So, that means there's storage deserts, and this is already a thing. This isn't about the future. This is happening right now. Here's, as you could see scrolling on this list, major GPU clusters that are being built in the past year that are planned for the next year to come, and they're all being built very, very far away from where the storage is. If you need hundreds of petabytes or exabytes of data storage, and you don't want to pay exorbitant rates to the GPU cluster themselves or AWS, um you don't have an option that's

highly performant. You don't have an option that's close. And this is a real problem, and I'm going to explain a little bit why. If you A lot of people don't necessarily appreciate this cuz they're not thinking about so much. They're going to start thinking about it when they're trying to train their big AI models. When you have your data very, very far away from the GPU clusters, you're dealing with the laws of physics. It's not something you could solve with hardware. It's something that it depends on the speed of fiber optic cables, the number of nodes that you have to hop through. And if you have the data close on the same connection, physically close to the actual compute, you're going to get far more throughput.

Clicker's a little flaky. There we go. It's also extremely expensive. So, there's a whole economic side of this that is not being priced very well. Egressing data from a lot of these traditional buckets of data like AWS is extremely expensive. Why is it expensive? Well, because it can be expensive. It doesn't actually cost that much money to put a lot of data online and move it. But, AWS charges you a ton of money and the proof point of that is companies like Coreweave are actually giving huge, huge credits to move your data into Coreweave server so that it can be close. And get people to actually move their their data out of those buckets.

The problem is if you move your data into a Coreweave bucket, and we'll talk about this more in a second, and you really want to train your data on a Lambda server or on a Nebius cluster because that's what's available, you're not you're locked into that one vendor. This is just showing you that it's really, really important that your data is close and accessible really, really fast. It's not just about the cost of egress. It's not just about being frustrated. the cost of iteration and whether you actually win in the AI development cycle that you're participating in. If you can do more tests and you can run more checkpoints, if you can bring your data to where the GPUs are and iterate and test and try things and experiment faster, you're going to outpace your competitors. So, somebody that has their

data, their big data lake that they're training their data off of, 1,000 km away versus 10 km away or in the same building or on the same meet me room, um, is going to have a big, big competitive advantage versus that person that has the data really, really far away. Well, why, uh, why is nobody solving this? And the truth is that it's doesn't really make sense for a lot of the incumbents. The hyperscalers don't want to solve it because they're working at really, really big scale, and they need really big scale to justify and pencil these things out. Hyperscalers will spend a lot of time analyzing. They'll spend a lot of time building very, very big data centers, lots of big infrastructure. It'll take them a lot of time. They're also very focused on the GPU build-outs. And by the way, if you're a user, you still don't want to

be locked into one hyperscaler, or even one neo cloud, for that matter. So, that's where Filecoin comes in, and we're going to talk that a little about that a little bit more. If you're a local operator of storage, if you're in a data center as a collocated operator, or maybe you have a small data center, you have an opportunity to be where the compute is, and to actually benefit from that without actually buying your own GPUs. You don't have to go do the CapEx and and compete against CoreWeave, or against Nebius for GPUs. You can just be the data layer that's closer and faster and more performant and in the places where the GPU clusters actually are.

Um, just to add another layer on, if there's any lawyers in the room, I kind of doubt it, but maybe there are. Um, there's also the issue of of laws. So, where the GPUs are being built in places all around the world where power is available, where power is cheap, and that sometimes means that that data that you're going to use for your training, or for your inference, for that matter, which I think is going to be a growing reason to have data lakes. Big companies are going to have all of their data in one place that the LM LLMs can access. It needs to actually be in the same jurisdiction as where the compute is. So, if you have a cluster in northern Norway, or maybe, uh, somewhere in eastern Europe, you need to keep in mind that that data actually has to be in the

appropriate jurisdiction, and having a big, broad, distributed, global network where the storage is actually available close to the GPU clusters gives you that flexibility to make sure that you can match compliance with the performance that you want and with the things that you need to do so you can iterate more quickly. So, Filecoin actually already is this and this is the the big story that I think everybody needs to understand. We already are all around the world with our clusters and we built this over the last five or six years, um going back even more if you count for development before that. Um this is a really powerful asset. You don't have to do billions of dollars of capex because billions of dollars of capex have

already happened around the world and in a lot of cases these clusters of storage devices which are very hard to get um if you look at some of the press releases from the hard drive makers, they're actually have order books that extend out beyond 2027 at this point. Well, all that infrastructure already exists. So, the the network is there and it's available for use and as Filecoin that's what we're working on making happen. in 2026 is that this big already built network, this big capex build out, you don't have to make it, you can actually leverage what we have and what we're using.

We already talked about this a little bit but uh frankly, it's really important that you have an agnostic cloud, too. So, there are other options for storage out there. People are thinking about this. If you go to any of the big GPU neo clouds, they have storage services. They're very expensive for what it's worth, but also you're locked into that one cloud. Let's just say you move all your your data to a cluster in Ohio from one particular company, but they're all tapped out. They don't have a big cluster available for you to train your data. I don't know if anybody actually has gone and tried to reserve an instance on on one of these neo cloud's websites, but they have usually the oldest hardware and the smallest quantities available and everything else you have to reserve out for a month. So, you need to be able to move your data

very quickly between different clusters so that you can iterate quickly and be competitive and create the best possible product for your client, for your company, for whatever it is you're building. And so agnostic data storage is critically important. Having agnostic data storage that is really, really close to the GPU clusters you want to use, that's even better. So Filecoin, as I mentioned, is a network is a huge network that is kind of built for this. Um we have almost two exabytes of data uh capacity available around the world. We've shown in the past that the power of our network enabled by our crypto economics allows us to flex the size of that storage network up even more than that. We built a very, very big network very, very fast. Um we clearly know how to mobilize

lots of of of capacity. And so as we tap out these two exabytes, we'll be able to grow with it. And so, you know, we talked about the availability of storage before as being a major, major problem. Not only are we putting the storage where it needs to be at the performance level that it needs to be, but we're also able to tap into excess capacity that's not being used that's the being underutilized. And that's a really, really powerful force because every spare storage device, every hard drive, every NVMe really is going to need to be used over the years ahead. So everybody's talking about the constraints and the bottlenecks on compute. We're talking about the constraints and bottlenecks on memory and on RAM, but this is uh something that is starting to emerge and is going to become very hot because it's all uh

tied together and Filecoin as a network can uh be a global solution for that. There's a uh two ways to try out Filecoin in 2026. If you haven't tried Filecoin in a long time, I uh would really, really encourage you to give it a try because there's some really new exciting things. Um it's a lot easier to use if you've tried to use Filecoin in the past and found it a little confusing, maybe a little frustrating. Um it's it's not frustrating or confusing at all anymore, uh if it ever was. Uh File one is sort of our S3 on-ramp. So this This a really good way to just swap out your S3 endpoint and start using Filecoin services. We're live in the EU right now with a really really highly performant region and we'll be in North America

with a highly performant region very very soon next couple of weeks and we'll be in Asia in the next couple of months. So we're going global with that S3 sort of compatible layer for Filecoin and Filecoin on chain cloud which is absolutely developed for developers, for builders, for anybody who wants to use smart contracts, be programmatic with their their work and this is something that we've released over the last 6 to 12 months so you really should check it out. I really would encourage you to give it a try, put it to the test, benchmark it and tell other people about it.

So that's it. That concludes the talk and I hope that what you took away from this. I by the way I'd be happy to speak with you if you want to follow up. Feel free to send me a telegram but I think that data is the underappreciated part of the story of the generational AI infrastructure build out that we're undertaking right now and the solution is not in hyperscalers building big data centers or building data stores that locks you into a single provider. It's in activating a global network of highly performant but smaller and and mid-size data centers that can give you all of the performance and storage that you need right where the compute is and as

this whole trend of building GPU clusters all around the world including in the most remote places you can imagine but where the power is grows Filecoin is going to be a critical part of that story. Thank you very much. >> [applause]