Scaling DePin: Storage, Compute, Edge Services by Paul Wagner

Transcript

I think so. Uh, okay. We've got uh the right size crowd here. I'm gonna kick this off if there's a a few people in the room that are willing to answer questions because I'm going to have some questions for people. If you're a developer in the space, go ahead and come up. Uh, I'll do a quick introduction. So, I work for the Filecoin Foundation. My name is Paul Wagner. Uh, I've come from about almost 18 years of experience in enterprise development, sales. Uh I've also run a data center management company. So I have a deep background when it comes to heavyweight infrastructure and uh currently based in Austin, Texas. I also actually lived in the Middle East as a student uh about 20

years ago. So I have some background. I lived in Kuwait during the height of a very certain period of that time in the Middle East. Nevertheless, uh there's probably a handful of people in the room that use cloud services both for their projects, their development in the past. Um I'll probably query on the crowd at some point about that. Nevertheless, the crypto space has tried to uh we'll call it upseat the cloud services that are available. It's been successful in some areas, but there is no other way to say this that big cloud is big and it's only getting bigger. The capex that

Amazon is investing globally in 2025 is supposed to be $75 billion. Uh number two, which is Azure, which is near in revenue parody, is investing similar. And then you have your smaller services below that. Uh the Chinese clouds are also quite big, but if you were to add up all of their revenue, they would only be as big as Amazon as a whole. Nevertheless, if you add up all the revenue from these these companies, it's somewhere between 350 to 400 billion dollars just in cloud services. It's a massive market. It's growing somewhere between 15 and 20% a year. It will only get bigger. Now, how

does that apply to the crypto space? And not only that, the Maya region also has its own growth targets that are also very important. If you've lived in this region, you probably have gone through some of the changes that were available to you, specifically where Huawei Cloud was trying to make a big splash here, but the large North American companies have come in and footprinted themselves across the main region and Azure seems to be the number one. Now does that matter? It might matter for you if you're concerned about where your infrastructure is. Uh some of the the mayor region is misrepresented. So uh the large cloud services are putting

infrastructure in those regions. But also from a historical standpoint, the companies in the main region often build their own infrastructure because they prefer to have sovereignty over it. and they may be purchasing some of your traditional services that would require that. Oracle, IBM, if you've been to Internet City, you can see the large uh big cap tech companies that are represented here and they've been here over 30 years. So, they do a good job. Now, in the blockchain space, there is a term that has been circulating around. It's called Deepin. Who coined the term? It's probably likely a VC firm. I don't want

to name names, but when you Google it, you'll see it. You have probably seen large graphs like this maps, who's who, who's surviving, who's maybe thriving. Uh, Filecoin has been around since 2020. Some of our tech even prior to that. Uh, we're somewhat uh resilient is the best way to put it. And I'll explain why the storage industry from a deepin side, the decentralized physical infrastructure is often more resilient than what we've seen in the last say 18 to 24 months of large swasts of GPU deployments that will likely either get consolidated or they will make their money and close up

shop because the turnover in the GPU market is very high. Nevertheless, there are some longstanding companies on this list. I'm sure all of you um have seen the upper left Helium. There's they they've also switched over in the United States. You can get a mobile phone now that uses Helium data. Very effective. There's a lot of competition in that space, but they're making moves. There's some new newer launches in the geospatial nets. Tomorrow there is a deepin day. uh you can ask me afterwards if you're more interested in this space and the heavyweight infrastructure, some of the IoT use cases. There's a whole specific day for this that's been somewhat tr traveling

around the blockchain conferences and you can learn a lot more and talk to the people there. A lot of their services are live. You can read and write data to them. You can buy their services. You can provide infrastructure. That is one benefit of the deepin space is it tends to be a bit more physically oriented. So it's a little easier for most people to grasp from day one figure out if there's a product market fit or if there's an interest on their side. So at the end if you're interested I can send you the link or you can Google it. Uh I believe it's about somewhere between here and Marina.

So about 20 minutes if you're around. Some CDM networks that are on there. We're going to talk briefly about those because I do want to talk about storage, compute, and edge services. I'll briefly touch on GPUs, but probably not uh in depth. I think that's a market that has been become quite saturated. And if anyone's tried to consume those services, you've seen the price of GPUs per hour go down by about 50% in the last 18 months. That's that's pretty clear uh especially if you're looking at the web two alternatives on runpod or vast. Okay, I want to talk specifically about compute. Uh this is somewhat

filecoin leaning in that we work with all of these providers. There are other great providers out there, but I know that these work. We have ecosystem companies that are building both on top of them using them trying to supplant their services that might be in AWS, GCP, Azure, Huawei cloud or bare metal. Akos is largely what you might consider a what I like to call a tier 2 cloud provider similar to OVH htzner or some other vulture is another one that I'm sure people have consumed services with.

They were largely first to market in this space. Uh they are doing I think advertised somewhere between 2.5 and 3 million ARR and they're getting paid on all of these services. They have a few locations around the globe, but they are largely in a few handful of cities. Uh, but they will work for a lot of your services. They also offer what I would like to call professional services migration. So, if you're trying to move out of one of the major cloud providers, one of the benefits that the DPIN providers will do is they will hold your hand on it. You're not going to get that if you're a small-time thousand to $5,000 a month purchaser of services on AWS and trying to move to GCP. You're

not big enough. But because the desire to win is there, they will provide those services. On the Akos side, they have your standard services like EC2 if you need compute, storage all-in-one. It's great for if you need to run basics web services or an application that works on a standard client server model. You can turn up services relatively quickly. I'd also liken them similar to Lenode, which is another great example, but they also do provide GPUs. Another one is Node Shift.

Node Shift is an aggregator. They basically are going out and vetting deepin compute providers and doing what they can to basically ensure the service level agreement is solid and then they will then take a rake of the service on top and they also provide CPU, GPU and storage. again their console from a developer experience very similar if not easier to what you would get from I would probably say OVH or Vulture very easy to start a node get an SSH login get started cost savings on most of these are aiming somewhere between 70 and 80% on a

monthly basis so I would assume a lot of people either if they've raised money or they have received some form of credits from AWS. We'll eventually run out of those credits. These would be my my top picks just because I vetted these services. So, I've saved that time for you. There are other ones out there, but I know that they have a customer service arm, so you can get a hold of them if you're going to migrate a,000 to $2,000 of spend per month over to these services, which is why I like to highlight them and put them up there. But again, it's up to you to determine them. At the end, I'll have this the slides available for sharing. So, if you want to check them out, you can. They all accept credit

card payments, which is also very nice. You don't have to go buy their token. Usually, they're offering a discount on their token, but from the deep side, why I also like these providers, um, just uses their token to pay their providers like we do on the Filecoin network. We would love to have everyone pay and fill but it's just not practical. So we uh I would say the successful deepin providers or networks will obuscate that from the user. Uh Fluence is a company that has really really made some headways in our ecosystem. Um they're also sponsoring DPIN day tomorrow and the founders will be there. They have started off in a serverless environment. So if you have

Lambda functions that you would need to migrate, they are a perfect provider for that. They are also launching an EC2 replacement. So they'll go head-to-head with both a Kosh and node shift over time. Um they also do a very good job of vetting their providers on the network to ensure that they have a high quality of service and a service level agreement that can be met that's equal or on par to a 49 infrastructure setup. much like AWS GCP Azure etc. Wrong slide. All right, I am going to highlight a little bit of where I have to lean, which is the Filecoin side. The global storage market

is a massive market. This graph would include all of your major cloud providers. That again goes the major US providers, the major European providers, and the major mostly Chinese providers. And it would also include your traditional services that you might consume on Box.com, dropbox.com, mega, etc. So why is the storage market so big? Well, it's one of the largest dependencies for any application developer. It's also a very mature market. It's in the terms of computer science or computer hardware, computer development. It's been around since about the n the the first hard

drive came in around 1956 is usually the number. Um but from an evolution of markets they've done a really good job of basically ensuring you don't have to delete anything. So the market continues to grow. That being said, uh if you are in the actual guts of the business, the storage provider business at the hard disk level, your margins have been squeezed from somewhere in the mid 90s around the the low teens all the way down to about 1 to 3%. It's an incredibly efficient business. It's a triop in that there's only three providers in the spinning disc space

globally. That's uh two of them are from the United States and one is from Japan. Most if not all of the hard drives in the world, 50% of them are made in Southeast Asia. The rest are made in China and some parts of uh Japan as well if they're high-end. The SSD market is slightly different. There are about seven providers with uh four of those seven being North American as well. Their margins tend to be a little bit higher. Uh but again that's quickly being squeezed down. So the business you want to be in is not actually building hard drives because it's in incredibly complex. It's incredibly efficient and it's incredibly locked

down. But where you do want to be is why is there a $200 billion TAM that's growing between 15 and 20% per year? It's solely because it's the highest product margin business in the cloud space. So when people are building on top of storage and reselling storage, they are often considered to be the highest margin business. Uh not only is this true when you look at the major cloud providers, if you can get deep into their 10ks, but also everyone knows this internally when you talk to a solutions architect at any one of the major cloud providers or a salesperson or a salesperson at any major value

added reseller around the world. Selling EMC, veh or any of these backup services is where you want to be as a salesperson. And those jobs don't come up very often because they're great ways to laugh your way to the bank. Now, when you look at some of the other, we'll call it guts that you can resell into the market in the deep space, CPUs and GPUs tend to be lower uh margins. They also tend to have a a recovery period that's a little shorter. The recovery cost period in storage is usually a minimum of 18 months uh closer to 36. That being said, the lifespan of a hard drive or your

unit of production in the case of we'll just call it storage or CPU or GPUs is usually five years or more. So the lifespan is very high. Now what is the lifespan of a GPU in the current marketplace? one year. Jensen Hang is dialing that dial down as as much as Wall Street will push. So, they're they're nearly on a one-year product life cycle. I'm I'm ready for the H300s. I don't need them, but I'm happy to sell them to you if you need them at the right price because they're not available in in this region.

I know. Uh that being said on the CPUs they tend to have a recovery period somewhere around 24 months as well but the lifespan on CPUs is usually about 36 months from the major cloud providers. In larger markets, you can often find huge deals of used CPUs and GPUs on the market, and they are usually being resold through satellite or we'll call them hidden companies that are basically selling out the used equipment out of AWS, GCP, and Azure. So some of the actual second tier cloud providers are buying that

equipment and basically lifting and shifting and reinstalling it. So that is why often you'll see a product life cycle or new CPUs or GPUs in the console of AWS much faster than you would anywhere else because they have a very good system of moving out old hardware. Even in US- East1, which is the the largest footprint of any cloud provider in the world, you'll often see that move around. There's a a law that I like to highlight again why people get in this business. And it's by a famous bank robber in the 1930s and the story goes he was walking out of the courthouse in in handcuffs and they they ran up to him

and said, "Willy, Willy, why did you rob all those banks?" And he replied with, "Because that's where the money is." And in the again selling the guts of a data center where I like to say the money still is is in storage. And you can see this from the list of nond deepin storage providers just in the web 2 space. Some of these companies are 20 years old which in the IT space or just in the general technology space is almost a lifespan that anyone can dream of. And again from the sales perspective knowing this um having sold this equipment and also services I would much rather be selling that than anything else because usually the renewals on

those deals are most people sign a minimum of three three-year deals because they know they're going to store it that long. But to give you an idea of a market that maybe you have an idea of and why it would be this way is Hollywood is one of the largest archivers in the world and they use one university. Uh I will not name the university because they're a client of ours but that university has an internal archive department that does $80 million a year in ARR just storing archived films. They also have a digitization portion of it where they will store both the physical copy and the digital copy.

Their terms on storage are six and 20 years and they are paid upfront for those costs. It's an amazing business to be in. Every CFO's dream in the SAS world would love to see six and 20-year contracts paid upfront. And just to give you an idea of what they do, they do a lot of really good customer service and they just resell certain cloud services which is again is what a lot of the DPIN providers that I find to be successful are doing. They are not building out the networks. They are building application layers on top to consume these networks and do it much better than the network ever would have done as a whole.

Now to highlight some of our providers on the network that have built on top of our roughly four exabytes of storage. We have three that I'd like to highlight. Sriracha is one. They use some of the original IPFS technology and then they uh partner it with Filecoin for their tiering of that data. Now they have a very simple lightweight API. It is its own API, but it's very well documented. And thankfully, it's all been trained on the major uh GPT providers. So, if you actually ever have any problems with it, you never have to use their documentation. They have full Ukan support, which a lot of developers in the blockchain space prefer outside the

blockchain space. We're not quite there. Uh we can talk about that tech if you'd like to. I just want to highlight that. Lighthouse, one of our providers, they've been around uh nearly three years. They, I believe, have something somewhere between 1,00 and I think 1500 customers that use their API on a daily basis, so they very high IO on it. Um, it's smart contract native, which is very important. Uh, they're usually at all of the hackathons, so they capture that value right there on the spot. um they have a little bit different pricing model wherein they're trying to do longer term storage and trying to you to pay upfront for that. So you'll commit

and then that also helps with the way they run their business. Uh Sriracha runs more on a monthly basis much like pay as you go. We have AAV which is one of our S3 native providers. They also have an option to run your own node. So if you need to have somewhat of a local copy next to or adjacent to your infrastructure, you can have an S3 native IO consumption there. That makes it much easier than having to write externally. And then you can opt to have that data replicated outside of your environment. Not necessarily like a CDN. We'll get to that, but just for an additional hot copy or as the industry likes to say, N plus3. There's a couple

different terms used to if you're a big bank, for example, you will probably be required not only by law, but by your CTO or your CISO or your CIO to have three copies. And one of those copies will have to be far outside a minimum of say 500 kilometers from your headquarters or or US- East one or wherever your data center might be. even if you're using eraser coding or whatever. Usually in in North America, I believe GDPR has some form of requirement for replication as well. But I'm not an expert on those standards, but I'm just uh more of an expert on what I see in the market. Uh because

normally the the data administrator uh at these companies, not their full-time job, but also they will lose their job if they lose the data. So we offer these types of native services because a lot of the people that have built on top of our network come from a very specific background. They come from the data storage background. They know what the customer is demanding and they know how to sell into these markets. All that being said, data migrations are one of the most difficult sales processes in the entire tech industry because one once something is working, people don't want to migrate out of S3. The usual

reason that they would come to us or come to any of these deep pin networks as they should is for cost savings. And the cost savings usually that we can provide on storage usually start at around 50% per month but we can go as is as as high as 70%. Just because we have a different cost structure also we're not offering usually in the deep space I will say no one is offering the grade of service that AWS is going to offer on the durability and availability. uh but they can come very very close and I think that's usually sufficient for 80% of the use cases out there unless you're running a high

frequency trading application or something like that. Other than that generally all of these services will look and feel just as good as um what you would need. Now the edge services market is somewhat I could go back to the original slide. However, this is a very difficult market to crack. Uh un unfortunately, it is also one of the ones that has the best use cases for decentralized nodes that might do something as basic as a CDN. Now, getting all of that to work properly has been a huge undertaking. And Titan is a team that's built on our network and

they also run their own nodes has been able to do this and supplant what I would consider to be a competitor to Bunny. Bunny CDN is uh usually what I would consider a tier 2 provider over Cloudflare and AI. So they're able to provide lots of lightweight nodes to do call it file sharing, file streaming all around the world and they do a really good job of showing you where it is and you can specifically point to a specific region. They have one customer that everybody knows about. It is Tik Tok.

Tik Tok uses their services in the Asia-Pacific region. And they also have another product that this is an interesting market in the DPIN space, IP leasing. Why would you need IP leasing? Why would you need IP hopping? Well, if you're trying to do something on the pin testing side or you're trying to do something nefarious, it's good to change your IP. That's a highly profitable service. You could reach out to Titan if you need to use it. But I will tell you that that service I am highly shocked at the margin on IP leasing because there's a there is a fixed number of IPv4 space around the world. I think it's like 4.3 billion IPs. I don't know what

percentage of them is unused. Nevertheless, being able to lease an IP for a minute is actually a service that people demand. Either whether it's for penetration testing or again, maybe you have some type of trading bot or you're doing some type of AI scraping, they they provide a really good service for that. That's one of their products. In general, again, Titan and a lot of the other edge service providers are usually saving you about 70%. whether or not you can get the performance uh related to Cloudflare or Aami. Again, this is where I would say you can actually get a hold of somebody and yell at them if the service is not working, which is one of the benefits of

where the current DPIN infrastructure is. It's not all chat bots that provide no customer service. So, the customer service on the deep end side is where the market is actually winning over some customers because a customer that's paying $3,000 a month will never get a support person in AWS if they have a problem. It's just not possible. AWS doesn't need they can lose $36,000 a year. Just going back to again this map, I predict this map will change. Some will fall off, some will stay, but there are

some anchor logos on there. And those are the ones that if you have questions that you know who's who, who's going to survive because anybody that's going to migrate again, a couple thousand dollars worth of spend doesn't want to have to do it six months from now. So what I've done here is like I said I vetted those ones on there. We we have our own internal system to vet these as working systems. But there will be changes in this graph over time. But there will be logos that will be here 12 24 months from now. Especially because deepin is a very heavy uh not only like weight investment, it's also a heavyweight installation. And the type

of people that get into it tend to have very long-term visions. They also tend to come from backgrounds where five uh and 10year planning cycles are much more common. So generally I find the deep pin networks to be resilient. I don't find them to be as interesting and speculative as probably some VCs might. But if you look at uh the Filecoin network for example, over $2.5 billion dollars of hard drives CPUs GPUs, and network cables, networking equipment in capex has been installed over the last four and a half years.

Some of the GPU networks did that in less than 18 months. So before people make those huge hardware purchasing decisions, you would hope they would go through their ROI calculations, which on an H100 is about I think they needed to be at around $3.5 a month if you were trying to pay it off in a year. Clearly, if anyone's followed that market, you're down to about $2 an hour now per card. So the payoffs uh recovery time is usually about 24 months there. So, why would you use deepen? Uh, outside of the fact that I I assume a lot of people in the room probably are using major cloud services, they're

either locked in to those credits, uh, and they're just under pressure to build an application, launch a token, and get the damn thing working. I don't really blame anyone for being in that position. I only blame the GCP salespeople for hooking you into their ecosystem and then making it very difficult to migrate thereafter. So if you start with one of the major providers, which some of these major providers also, if you are already committed to spending a couple thousand a month, they will knock those discounts down even more. Uh most people in the blockchain space don't have very large cloud bills. I can tell you I we we are privy to a number of these bills. Most

people are spending between $1,000 and $1,500 a month. I would not advise them to move. We most of the deepin capture is going to be two standard deviations away from that portion. But there's a big enough TAM that they can not only capture that market and continue their operations and continue to invest. Now, what are some other reasons to use DPIN? Uh, you could just use it as a backup for your EC2 instead of having to spin up multiple regions or multiple regions in storage in AWS. There are customers that like to do that. Just have it as the standard term is business

continuity. If you're in that that realm, a lot of the deepin services are really good for that. They're cheap. CTO's like to see that or CFOs like to see that. One thing they do offer as well is localization of services. There are markets, specifically Titan in the CDN market has nodes in countries that the major cloud providers do not. Therefore, they're actually a better option for a number of developers that might need services in Saudi Arabia or certain parts of India for that matter where there are data transport costs that are quite high, but they can be very very cheap at the consumer level.

It's good enough to serve out a local app. I like to say there's maybe less vendor lock. Uh it's definitely in the favor of all of these networks to design it so that you stay and pay. But usually the vendor lock is it feels far less like capture once you're logged in. It feels as if it was designed to if you want to move your service move your service but if our service is working you'll continue to stay. The last option is the self-hosted infrastructure. I had pointed out that there were a number of providers and I'm sure I'm missing some of them that if you need a local node to read and write from for performance that then might

replicate to the greater network you can find that in the deep space which is much harder to find uh in the storage space unless you're using something like minio but some of the second tier providers uh in the storage space usually don't have that option so back blaze and wasabi you would just be writing to a cloud endpoint. They don't actually have an onprem option that you might require. There are a number of markets especially in the storage space and we will probably see in the edge compute space that would prefer to have their or edge compute also being GPU prefer to have their inference more local in the building in and maybe even have those models local and just have

them backed up to a remote location. you can uh get a hold of me here. Uh happy to share these slides out if you have any questions. I have a swath of knowledge that can't all come out in one presentation. Uh but I'll also open it up to any Q&A. If you don't have Q&A, we can do it after here on the ground. So, any questions? Am I overtime? Under time, probably under time a little bit. It's all right. Paul Wagner again, send me a message. Thank you very much.

[Applause]