Tokenizing GPUs and Their Yields to Create a New Asset Class for AI and Compute, with Kony @ GAIB (Audio)

[00:00:03] Hello, everybody, and welcome to the Crypto Hipster Podcast. This is your host, Jamil Hasan, the Crypto Hipster, where I interview founders, entrepreneurs, executives, thought leaders, amazing people all around the world of crypto and blockchain. And I have another amazing guest for you today. And I'm really excited about this interview. So let's get into it. He's from Hong Kong. His name is Kony. He is the co-founder of GAIB. Kony, welcome.

[00:00:30] Thank you, Hasan. And sorry, thank you, Jamil. Really honored to be here. So I'd love to chat with you more on what we do. Awesome. Awesome. So thank you for joining me. And yeah, so I want to find out first about your background. What is your background? And is it a logical background for what you're doing now?

[00:00:50] Sure. I have a background for both in crypto and traditional finance, as well as like on the technology side. So I entered the industry like from a while back then. Before crypto, I was in traditional finance doing a lot of the credit research, equity research, and private equity investments.

[00:01:09] And then when I entered the industry, I was mainly at the asset management firm doing a few things like SPAC, hedge funds, VCPE, real estates, which at that time, I got super interested into crypto. That's why I joined the industry and jumped into a crypto exchange, top five at the time to focus on M&A, mainly helping the exchange to expand outside of China.

[00:01:33] That's why we look at M&A on exchanges, wallet companies, eSports, gaming, phantom platform, anything they can bring in uses would fall under my radar. And been there for some time. Until then, like there was some change in management of the exchange. That's why I left. And after I left, I went back to the old firm, the SMN firm that I first joined to help start a crypto VC fund together with my partner there.

[00:01:58] It was only two people, me and the founding partner. So we bring, we bring a team from two people to five. And then basically help with most of the things, including fundraising, research, due diligence, due sourcing, relationship with other VCs, accelerators, and also like student clubs as well. So I really loved what, what crypto is bringing to the world. And also I'm a big fan of like AI.

[00:02:25] So that's why I have also like been keeping a closer eye on what AI and crypto merge. So for the fun side, for crypto VC side in the past two, three years, my main focus was to look into infrastructure investments. So that's why we invested into Eigenlera, Babylon, Risk Zero, Linear, Sui, Polyhedra, so on and so forth. These like deep tech crypto infrastructure projects.

[00:02:52] But other than that, myself, I like every aspect of that. So I also like, we also look into the perspective, including DeFi Infra, like MEV, supply chain, intempo.goes, and also crypto and ad merge came in around late 2023. That's where it caught my eyes. Because I'm a big AI fan. I've been following with, on AI for quite some time.

[00:03:14] I view my own checkbook, trading bots, and even AI agents with long-term memory long before the craze we are seeing these days, around like trajectory 3 and 3.5 launch. So pretty familiar with the whole workflow of like, how do you launch an AI agent? How do you work with them, et cetera. And it's actually around like late 2023, when the whole market, crypto market and AI market merged. I was like, what's so interesting here?

[00:03:41] Like, what are the projects doing between the crossovers of the two? So it was at that time that I studied most of the things, like talking about data model, decentralized compute, AI applications on crypto, et cetera. And that's one thing very interesting. There was nobody talking about finance and AI and crypto crossover, which intrigued me.

[00:04:05] Because given my personal background in finance, obviously the thing I saw, the opportunities I saw in the market was, in every asset class, there's got to be a finance company that service the whole industry. For example, in airlines, there's aircraft leasing. In current, there's aircraft leasing. Like, this current, this product leasing company. And then in other asset class, we have a market for gold. We have a market for oil.

[00:04:32] Why don't we have a market for GPU asset class, which is the most important infrastructure that the AI era needs, right? So it was at that time I had this idea. Well, there's got to be some place for a finance company that focus on this in the market. So I talked to my co-founder, Alex. So he is basically the AI and semi-controller guy.

[00:04:54] So he's more like, oh, his family runs ReTech, which is the top seven global chip maker these days, making a lot of the microchips like Bluetooth, Suncut controller. And then he himself also runs an AI cloud company, basically servicing computation power to companies out there that want to launch their AI services.

[00:05:16] So I was asking him the question, do you see there's a need for exactly like a finance company or company to help these AI cloud companies in their sense of finance? And he was like, dude, I have been thinking about this for so long. Like, you know, let's do this together. So that's why like two minutes into our call, we have decided to start a company, use some time to polish the idea, fine tune the architecture.

[00:05:40] And also, I also found my CTO afterwards, which is he has a PhD in cryptographic research, a long DeFi expert that has joined the space from 2016 onwards. Perfect guy that I go to because I can leverage his expertise to actually leverage on all the DeFi, DeFi building blocks to achieve the goal that I want to do. So what getting back, right?

[00:06:10] So that's why, you know, my background and the team's background, we have a very diverse background to cover both like to cover all the three areas, AI, crypto and DeFi. So making it a good marriage of the three. So can allow us to do what we do here, which is to become like the economic layer for AI compute future. Okay. Fascinating. Okay.

[00:06:38] You said, you said you put out leasing. You know, I don't think of I don't think of AI and leasing together. So I want to find out more about that a little bit. So you told me what Daib is all about, right? And you are transforming AI infrastructure investment. So I want a two part question here. One, how are you transforming that investment? And two, what's going to be the role of leasing in AI? Yep.

[00:07:09] Yeah. I just answered a second question first. So like, well, the leasing is basically what I was trying to say in every asset class. You will have a company there to help you finance the assets, right? For example, the airline, they don't really own aircrafts. Like most of them are being leased to them. Like people help, basically people help them finance that, right? So that's the whole point of me talking about leasing.

[00:07:34] Because there's always got to be a finance company to help you finance the asset purchase, run an asset operation there. So it's the same thing for AI. Because AI relies on a lot of the cloud companies and data centers. They require a huge amount of GPUs to be up and running, to be set up. So we need to have a lot of capital. So we need finance. And what Gaiif is all about, how do we transform AI infrastructure?

[00:08:00] Well, as I mentioned earlier, we are the first economic layer for AI and compute. Make it in simple terms. We transform these GPU-backed assets into yield-generating opportunities. Meaning that we also have a product called AI dollar. So Gaiif's a synthetic dollar. So the investors can actually invest into it so that they can earn yield from the actual AI-powered compute there. And the problem that we are trying to solve here is pretty simple.

[00:08:28] Mainly two problems in the market, in the crypto market and the blockchain market. One is there's a lack of real yield in the market. You know, as Vitalik noted in the crypto industry. We have to break the elbows. The yield has to come from something external. You know, we can't just rely on token subsidies and, you know, continuous emission of the tokens there to make people buy into that. It's not sustainable.

[00:08:55] And then the second problem that we want to address is this lack of access to the AI infrastructure investments. So other than indirect ways of buying and feed their stock, what are the ways you think you can capture the AI infrastructure growth? Well, I think that's a way for most people to do it, right? So exactly these two problems is the reason why Gaiif is helping us solve.

[00:09:22] Number one, we want to bring in the actual real yield into crypto. And this yield has to come from something really sustainable source, which is the GPU's revenue being paid by the actual AI demand out there. This company renting the GPUs, paying them money to our cloud company partners. So these are like money-making assets. And we want to bring these actual cash flow into the market. And the second thing is about the direct access, right?

[00:09:49] So as I mentioned earlier, there's no market to give you direct access to invest into GPU, these asset class. And through Gaiif, through us tokenizing the GPUs and also the yield, we can create a new bank product on chain. So it can give you direct investment into the AI infrastructure, which is super important because it's all the upstream AI applications. They all need that. It's basically the engine power of all AI applications.

[00:10:17] So that's how we capture the value in the AI supply chain, right? And on top of that, we build DeFi. Like other than tokenization, financialization, we're going to build DeFi on top. So bringing the physical assets on chain is the first step. Second step is bringing liquidity, creating an economy. So that's why we'll be integrating with the other lending point protocols, even setting up our own option futures trading,

[00:10:42] the other tiers and other structured products so that we would like to create products with different risk-reward profiles that feed everybody or anyone's appetite in terms of the risk, the investment needs, the appetite there. So there's a whole goal of what we want to do, to change the AI infrastructure investments successful people. Wow.

[00:11:07] So I might get into the real yield versus fake yield later. But you said something really interesting because if I talk to some common lay people, my neighbors or friends or whatever, they believe that AI is the future. So how do you access it? I said, well, I use chat, I use chat, GBT, or I could buy NVIDIA stock. Or if you're in crypto, you're either buying iExec, RLC, or Render.

[00:11:37] But you're not actually being, you're not able, like no one talks about being able to access that core infrastructure like you're talking about now. So why is it important to provide investors with the direct means of accessing AI and benefiting from its growth? Like what's that going to open up and why is that super crucial? Yeah. So exactly as you mentioned, right?

[00:12:04] When we ask, tell most people, like, how do we invest into AI? They talk to you, they talk to you exactly. Let's go buy NVIDIA stock. Well, that definitely gives you some ways to trade stocks like other companies you want. But where's the direct access? So that's exactly what we want to do. Provide a direct means for them to invest into it. And why is providing direct means to invest into AI infrastructure so important? It's actually very simple.

[00:12:31] If you want to grow through the whole asset class, always invest in something that captures most of value. The benefit from the pie, the growth of the pie. And what is the most important thing in the whole AI ecosystem? I would describe the supply chain or the value capture chain. It's actually a smiling curve.

[00:12:55] Meaning on the right-hand side, these are basically application companies, including Meet Journey, Runway, and the other, like, newer, manless AI, like these more application level of AI companies. They're fighting their product market, they're making more money, their pie is growing, right? And then on the left-hand side, what are these? These are AI infrastructure companies, including all the cloud companies, including all the data centers,

[00:13:25] including NVIDIA, who's making the trips there, right? But these AI infrastructure, no matter how big our application grow, they still need it. So basically, they're enjoying the growth of the whole market, the growth of the pie. And no matter how the market changes, I still think the small curve rule actually makes sense. Meaning they capture most of the value right now. If you look at how much Meet Journey makes, while they're good, many people are using them.

[00:13:51] But then the money they're making versus the cloud company versus the NVIDIA is really a small percentage of that. And I do believe they will grow someday. They will definitely be bigger. If you look at the whole SaaS supply chain, it became like a pyramid to an inverted pyramid. Meaning the application captures more value than the infrastructure side. But if you look at the AI pyramid, the AI supply chain, it's still like a pyramid shape.

[00:14:19] Meaning the infrastructure, which is at the bottom, captures most of the value there. And that's exactly what we want to unlock, allowing people to have means to invest in it. And that's the famous saying, meaning in the AI era, compute is a new currency, right? Sam Altman, Elon Musk all say about this. And what's so important about firing up this currency are the AI infrastructure. These are computation assets. And it's hard for you to invest.

[00:14:43] The reason why it's hard for you to invest is because all these enterprise-grade GPUs that OpenAI is using, Brock Elon Musk exists using, they all come from a huge amount. Meaning, number one, high cost. One H200 machine, AC costs $200,000. There's no layman that you and I, we can just buy a machine, get home, service the other cloud companies that it needs there. It's impossible. So that's why most of it, we don't have access to it.

[00:15:12] But through us, through tokenizing these GPUs, making it fungible, making it tradable, people can now invest into these asset class. And also, if you look at the market, how is the growth is going? Well, a lot of the capital, a lot of the money is still going to infrastructure investments. Manifestation 7 investment into AI in front per year is $300 billion. $300 billion into it.

[00:15:37] And Alibaba just announced a $50 billion plan to invest into AI in front in the next few years. Trump, when he got signed to office, President Trump who has gone to office, mentioned about the Stargate $500 billion initiative with software and OpenAI. So it just tells you, we are in a market that is really demand driven on the AI infrastructure. There's a lot of demand to invest into this market. We're still early.

[00:16:04] We want to provide access to a market that's growing, access to a market that's early and access in a direct way instead of indirectly. So this is like what we are doing, what we're focusing here. Let's see. All these opportunities that people haven't had before, you're now unlocking and you're tokenizing at the source creation utility level.

[00:16:32] But you're not just, let's see if I understand this right, you're not just creating tokens. You're actually creating options, alternatives, trading opportunities, a whole entire finance market around the previously unaccessible parts of AI that people wouldn't be able to touch before. Exactly. Exactly. The whole point that we wanted to do is, you know, back in the days, gold was super liquid. High trading evictions, hard to trade them.

[00:17:01] You have to move physical gold to where you trade it, right? And afterwards, different types of products or instruments just came out. Now you can trade gold on ETF, you can trade gold on futures, you can trade gold on index fund, or you can trade gold on some other structure products, right? This is a huge amount of financial market that comes out. Like we have a market for oil, for gold, even for orange. Why don't we have a market for GPU asset class?

[00:17:26] Or even when the other computation assets will come up, there might be TP and you like, we want these asset classes. People want, why can we trade it? Why don't we invest into them, right? That's the whole point of ours is to break down barriers, allowing and giving access to people. I like it a lot. I think it's genius.

[00:17:47] I want to find out, you know, besides investing, besides finance, like, you know, AI has come in, has been around for 120, 140 years, right? At least the discussion of it has been around for that long. It comes in waves of every 20 years, you know, and then it stops and then it goes back and then talk about it again, then you stop. And then now it seems to have historically more momentum now than it's ever had before. Right.

[00:18:15] So what do you see as a future of AI infrastructure? It looks like it's got legs now. So how do you see, in addition to the finance, what do you see possible? Yeah. I do think for the AI, like, not only AI infrastructure, right? So, but the whole industry. That's a two key thing. One is definitely we're still in the early stage of growth. Before Tread2B was out.

[00:18:42] The reason why we had a term called Tread2B moment, where we get a historical number of people signing up to use Tread2B on OpenAI's website. The reason because of that was before it was launched, we had no idea that AI could be that powerful. Like, imagine all those customer service chatbots that you were using, talking to online for your customer service when something goes wrong. They always give you the same answer, right? Irrelevant, really dumb answer.

[00:19:12] They was like, what are you saying? And also all those that use a menu, click, click, click, click. Like, when other Tread2B launched, right, we realized, oh, the machine can actually think. Like, we're using, we're actually hearing, like, intelligence details, right? So, that's a very important thing. And afterwards, in the past, like, two or three years, there's a long, long debate between, like, open and closed. Like, open source versus closed source.

[00:19:37] We saw a lot of efforts by different giants in the market just pushing for, like, open source of the development there. They have their own benefits in terms of how they develop. I don't comment on the both. But I do see that, you know, innovation has been driven over the past two years for all the competition on the model's intelligence, on the model's performance, on the AI infrastructure changes there.

[00:20:03] And the second thing about AI infrastructure, the future I'm seeing is decentralization. Decentralization. Decentralization in the means that before Tread2B was launched, the most common cloud that we are using globally, I mean, right? GCP, like Google, Microsoft Azure, and also Amazon AWS. But then, after AI was launched, there's actually more specific cloud infrastructure being spun up just to meet the AI needs.

[00:20:32] There are more local, regionalized cloud companies that is coming to provide more specific services to the companies there. The reason behind is to actually surface these AI models, they sometimes in production grade, they require quite a high performance and also low latency. Meaning you don't expect the machines to actually give you a response, like, after a minute. They want fast responses to have better UX.

[00:21:00] So in that case, the cloud has to be closer to where the company is based in, you know, to provide the best experience. For example, if there's a cloud company in the States, there's a cloud company in Asia. For an Asian company, definitely no doubt he will go for the Asian cloud to get the best experience for the AI models and AI services there. So there will be more and more cloud companies coming up in a more design-transmator.

[00:21:27] So there's a lot of data monitoring against, like, always seeing the top three cloud dominance in the market in terms of how they service the AI, how they service the compute. So this is very good because it allows for a whole decentralization of more localized cloud, more infrastructure to be built. So who get benefited in the end is you and I. Like, we are the consumers. We get benefited. A quick number that I can give you for references.

[00:21:56] I think a year ago, like, if you actually run a GPU on AWS, let's say H100 GPU, which was at that time one of the more top-tier ones, say, of our GPU, it requires $10 per hour on AWS. And then versus on the more smaller, up-and-coming, growing cloud companies, they're only offering you, like, $3. So there's a huge gap between the cost and between, like, the cost of using GPU, using computation power.

[00:22:25] And what benefits it can give is with cheaper cost, with cheaper hosting of these AI models or even training of the AI models, we as consumers can actually get more performant AI services, smarter AI applications. Even more specific domain can actually be benefited. They can actually train the models with a lot of cost, right?

[00:22:47] So we have been hearing about OpenAI using million dollars to train the ChattuVD 4.0 and a lot more newer, like, models require less money. Part of that is because we have a lot more GPUs at a cheaper cost or more available to them to be used. So a lot more reasons. And even on the model side, there's more advancements to the model improvements there.

[00:23:11] So in the future, I do see that AI infrastructure having exponential growth. And also, there will be more and more decentralized than what we're seeing these days. I'm glad you brought up the comment about decentralization because there's a lot of people out there who think that AI and crypto butt heads. You know, I talk to crypto and AI founders all the time who see a synergy.

[00:23:39] But there's a lot of people who don't think there's a synergy, think they're opposites. So how do we help get those people to understand the usage of AI and crypto together as beneficial? I mean, like, I was trying to give an example is what we're trying to do here is also to contribute to decentralization of AI infrastructure.

[00:24:03] I was basically mentioning for a cloud company to actually set up, they require a huge amount of capital. For example, a really small cluster of GPU, 1,000 cars of age 200, which is now the state-of-the-art GPUs. So they require like $30, $40 million. It's a lot of capital compared to the other businesses that you run. And to actually fund these initial capital, they're basically three ways.

[00:24:31] Either you're super rich as a company through a revenue, through your profit, or you go to VC, country capital, and say, hey, can you invest into us through our equity investments, right? We get the money to buy the GPUs and then get the service out and make money. And the third way is to get that debt financing. Blow money from banks, private credit, hundred and hundred individuals, private lending, et cetera. Most of the cloud companies out there, because we're still in the media market,

[00:24:58] it's very hard for them to get the third, like debt financing. Most of them will go to VC, get expensive, like equity investment, or relying on their own capital. So what we're providing them is a new way to get money, get capital, so that now you can grow faster. You can come to us. We help you grow. So the whole goal for us is we service more and newer AI cloud companies in this space,

[00:25:27] so that number one, we're helping the market to grow. Number two, we're seeing more and more decentralized cloud being, more and more localized and regionalized cloud companies being set up, and exactly these other people they want to work with and help them to grow. So in this way, GuyGee is helping the decentralization of AI. And talking about the other perspective, right, a lot of people are really, really reconvicted when they hear about crypto and AI crossover.

[00:25:55] They think the two cannot coexist. But if you actually look at the interesting thing about blockchain, what it's giving to AI is definitely one thing, is the business model. So what Web 1.0 and what 3.0 did differ is because of the token economy. Web 1.0 is also open source, but without a business model, everything becomes very good, which is a good thing.

[00:26:23] Like TCP, IP, all these protocols, everybody's using them without paying money. And Web 3.0 is we're having a decentralization right now, but also because we have an economy, a token economy, the business model can grow, can be more sustainable, so that we can keep on incentivizing innovation here. And a lot of people are giving tries on this way, including like Fuloco that's still tokenizing data ownership, tokenizing model ownership, tokenizing contribution by developers.

[00:26:52] They aim to push for more open source, more decentralized contribution to the market. So yeah, a lot of innovations and different perspectives of AI, blockchain, crossover. You said you're doing two things. It sounds like you're doing a third thing, and that is to dissolve the capital, the intense capital requirements to access this industry. Exactly. That's what we want. Yeah. And an additional capital channel.

[00:27:22] Yeah. So say I don't have this capital. And say, I don't know, I went to a couple conferences over the past year, and I was talking to chip makers in crypto, like Litecoin miners and Bitcoin miners. And they said that their chips they use now to mine Bitcoin can later on be used to access and to be able to build AI. You know, if I have a miner in my house,

[00:27:51] I'm able to use those chips, right? So, you know, what's the future of the chips and how can they be used to further expand and grow, you know, this infrastructure network? Very, very interesting point. It is exactly a lot of the cloud miners, a lot of the miners, the BDZ miners, like big, big miners, like Big Tier, Big Name,

[00:28:18] like all these mining companies are trying to pivot to AI because they see that AI, number one, they have demand. Number two is the AI, they have like revenues, like sustainable revenue there. And that's why they also want to diversify some of the reach from just mining Bitcoin. And the point I'm talking about chips. Well, mining chips and AI chips are slightly different in the way how they design.

[00:28:46] Like mining chips are, most of the latest, latest sort of mining chips, they are using like ASIC, ASIC architecture, meaning like application-specific chips. So the main function that how they design is to solve the puzzle on a beacon blockchain, you know, to mine the network. And then for most of these like AI chips, they rely on GPUs. And a key difference there is GPUs is highly optimized

[00:29:15] for parallel processing. So, which is needed by AI models to keep it running faster and keep it to run more efficient. So undoubtedly, like in design, they're the same, right? They also have the compute power there. But most of the mining chips are, like they're trying to be more suitable

[00:29:42] for mining, for doing AI computation there, but they're not designed that way. So a lot of the like big, big miners, they still buy NVIDIA GPUs there. Some of them, they design their own. For example, BigDeal, they have their own TPUs that was designed to do AI computation tasks there. Yeah, most of them see you buy NVIDIA GPUs to set it up. So there's some differences in terms of the architecture, how they're using. Yeah. So that's why in the future,

[00:30:11] in terms of the chip's future, we'll be seeing more powerful chip. If you listen to Jensen in the GTC, like which is like two weeks ago, I think. So he was saying the future is a power limited economy, meaning the more efficient the machine, it will win. Power is something that is scarce and is limited in terms of capacity. So that's why the whole point of NVIDIA

[00:30:39] for making chips is not only to make it run smarter, run faster, but also make it run more efficient so that it can reduce the computation power there and subsequently the cost for running these like AI models, for training these AI models there. And this was also reflected by H100 and H200, which was like the two version. H100 was launching, I think, 2023, and then H200 was in last year. So H200 and H100,

[00:31:08] they have similar computation power, like how fast they can solve a puzzle and how fast they can do a computation. But then the efficiency, the power cost for H200 is actually lower than H100, which means that efficiency is being improved. So subsequently, the running that, the cost also improved. So that's, that will be the, I think that will be the trend going forward. Increase, increase efficiency to reduce cost, reduce power cost or increase like quick power,

[00:31:37] computation power debt. So yeah, it's very interesting. With that, I would definitely like to see a lot more powerful chips coming up so that it makes a lot of generations, a lot of computation power for us. Because what I believe is two things. One, we can only do so much with AI because we can only have so much computation power. This is exactly where like all the open AI, big, big tech giants, they want full of computation power debt. And the second thing is

[00:32:08] a very interesting thing. Like the invention of Steam Engine doesn't reduce the usage of coal, but it actually stimulates that. Meaning that the more efficient AI model comes up, the more efficient chips comes up. It doesn't really, you know, reduce people's usage of computation power or AI. But vice versa, we will actually see an explosion of more AI applications, more interesting applications will be spun out. So yeah, I'd love to see this future.

[00:32:38] And I do believe that we are in that direction right now. Now I know what you mean when you said earlier, fake yield versus real yield. The explosion in usage and the efficiency creates real yield. Exactly. Yeah. Cool. Awesome. So I want to thank you very much for your time today. I enjoyed this conversation a lot. So thank you. I have one last question. How can people find out more information about you, about Gaiib? How can they do that?

[00:33:09] Yeah. You can follow us on Twitter. Just search Gaiib's underscore AI. And then you can also check our website, Gaiib dot AI. We have more documents about what we do, about AI dollar, about our pre-deposit bot that we're opening up in April. So stay tuned, follow us, and get more compute. And I do want to like, give me a little bit at time. It's the famous thing I like in Dune that I want to say it because, you know, Gaiib's name comes from Dune. We take inspiration from it.

[00:33:37] And also because like, just similar to Dune, Spice is the most important asset in the universe. In our time, in our AI era, compute is the most important asset. So he who controls compute controls the universe. Awesome. Thank you very much for your time today. Thank you, Jemmo. My pleasure. Sure. Thank you.