Interview: The Round Trip

Compiled & Organized by: Yuliya, PANews

In the increasingly capital-intensive competition of the AI supply chain, physical assets such as GPUs and robots are becoming the world's most scarce and valuable production factors. With GAIB announcing its TGE on November 19, this RWA × AI × DeFi model has also attracted higher external attention.

In the new series of Founder’s Talk produced by PANews and Web3.com Ventures, host John Scianna and Cassidy Huang invited Kony Kwong, co-founder and CEO of GAIB, to delve into the story behind the founding of GAIB, how to design a robust risk control model, and how to turn computing power into a tradable asset.

*Note: This video interview was conducted on October 30, and some data and dynamics may differ from the current situation.

The Founding of GAIB — Bridging the Gap Between AI and Finance

PANews: First, please introduce yourself and the impetus behind founding GAIB. We know you were a venture capitalist at L2 Iterative Ventures (“L2IV”) and have a traditional finance background. What experiences led you to discover this unique problem in the current AI field?

Kony: Before entering the VC field, I worked in traditional finance, including credit research, equity research, and traditional investment banking. Later, I developed a strong interest in cryptocurrency, so I joined a large exchange, responsible for its overseas expansion, leading mergers and acquisitions of other exchanges, wallet companies, esports platforms, and more.

Eventually, my partners and I founded a cryptocurrency VC fund focused on infrastructure investments, such as ZK, Layer 1, and cross-chain MEV projects.

The real turning point that prompted me to establish GAIB was at the end of 2023 to early 2024. At that time, the crossover between AI and Crypto began to emerge, and everyone was exploring the possibilities here. As an AI enthusiast, I conducted in-depth research. Since the release of ChatGPT-3.5, I invested a lot of energy, even building an AI agent myself. There were almost no mature frameworks available, and everything had to start from scratch, such as how to vectorize data, how to search, how to implement RAG (Retrieval-Augmented Generation), and how to establish long-term memory (the context window was very short at that time).

I found that these two fields I loved were merging, so I began to study the Crypto × AI track in depth. However, at that time, most companies in the market were doing:

• Decentralized computing power markets (like Akash and Render);
• Decentralized data labeling (using blockchain as an incentive mechanism to encourage public participation in data collection);
• Some focused on AI agents.

But what surprised me was that almost no company approached it from a "financial" perspective. This was unusual; as someone with a finance background, I knew that the most important thing in the early stages of an industry is to solve the "capital problem." The entire AI industry, especially in AI infrastructure, is experiencing exponential growth in computing power demand, which directly leads to exponential demand for computing assets like GPUs. This niche is extremely capital-intensive; building data centers, purchasing GPUs, and deploying infrastructure are all very costly.

I thought, why don't we do something here? For me, the core spirit of blockchain is to create new markets and new types of assets, which is exactly what DeFi and the "DeFi Summer" have taught us. Thus, I came up with the idea of providing financial services for AI infrastructure, as this part captures most of the value in the AI supply chain.

I shared this idea with my co-founder Alex. His background leans more towards semiconductors and AI; his family runs one of the top seven chip manufacturers in the world — Realtek. He also operates a cloud service company called GMI Cloud. He personally felt all the pain points I mentioned: as a startup cloud service company, obtaining capital is extremely difficult. There are two reasons:

• First, GPUs and computing power were still a very new asset class at that time (a year or two ago), and no one really understood it;
• Second, cloud services themselves were also new.

So, Alex and I hit it off. We believed we should establish a financial company focused on the AI field. This had happened in all other industries, whether it was coal, real estate, or any other sector. Thus, we decided to found GAIB last year.

Investment Returns in AI Infrastructure and GAIB's Business Model

PANews: Indeed, the AI field requires huge capital expenditures, and the return on investment (ROI) period can last for years. How do you view the investment return situation for AI infrastructure like GPUs? Additionally, how does GAIB collaborate with cloud service companies to help them shorten their capital turnover cycle?

Kony: In fact, most people may not realize that GPUs, as a type of infrastructure, have quite considerable profitability:

• First, large AI or gaming companies require a lot of GPUs and typically sign 2–3 year long-term contracts with cloud service providers to ensure stable computing power. This provides cloud service providers with robust cash flow.
• Second, on-demand AI computing power is very scarce. For example, a few months ago, when ChatGPT launched a new feature that could transform images into specific styles, it led to a computing power shortage even for OpenAI itself, indicating that the market is still in a clear supply-demand imbalance.
• More critically, the typical investment return period for GPUs is actually 12 to 18 months. This means that from a revenue perspective, the annualized ROI can reach 50% to 100%, which is excellent compared to other asset classes.

Regarding the second question, who we collaborate with. My co-founder Alex's company GMI Cloud is naturally one of our first partners and the starting point of our project. However, in the nine months that followed, due to the exponential growth of the market and the surge in demand, we received a large number of collaboration requests and built a strong project reserve. Currently, we are collaborating with over 10 "Neo Cloud" providers and Nvidia cloud partners globally, covering regions such as Thailand, Taiwan, Singapore, Hong Kong, Japan in Asia, the United States, Canada in North America, and Norway, Iceland, and Denmark in Europe.

We tend to collaborate with companies that are typically Nvidia cloud partners. For those unfamiliar, this means these companies have passed Nvidia's review and have the license and capability to provide software and hardware services. More importantly, they receive official recommendations and preferential treatment from Nvidia in obtaining the latest GPUs and quality clients. Therefore, most of our partners are Nvidia cloud partners.

This leads to our unique advantages in this field:

• First, our founding team is itself an operator in this field, and we deeply understand its economic model and GPU business.
• Second, we have a network of cooperative clients.
• Third, our capital solutions are more flexible, better meeting their needs in terms of underwriting time and duration.

As a result, our project reserve for collaboration has been continuously growing.

How Does GAIB Ensure Returns and Control Risks?

PANews: You mentioned collaborating with cloud service providers around the world, but the cost structures for electricity, data centers, etc., vary greatly across different regions. How do you ensure a relatively consistent ROI across different partners? Additionally, you also mentioned collaborating with Nvidia cloud partners; does this mean you have a set of standards to ensure the credibility and operational capability of your partners?

Kony: Yes, ensuring that ROI is completely consistent across different cloud service providers is very difficult because they each have different fee and cost structures. For example, the costs of electricity, facilities, and data centers in Asia are completely different from those in the United States.

Therefore, when transacting with these cloud service providers, we focus not on costs but on net cash flow recovery. We assess:

• If we give the other party $100, when and how will they return that $100 to us?
• All factors, including electricity costs, facility costs, and even depreciation costs that may affect asset value.

As for the transaction structure, it depends on the specific agreements we reach with them. Sometimes we adopt a fixed-rate model, requiring an annual fixed return rate. In other cases, we prefer to invest directly in the assets themselves and then take a certain percentage of the total revenue generated, such as 50% to 70%. In this model, our protection is stronger.

This boils down to the actual structure design of the transaction and our experience. We can require various protective clauses, such as prioritizing the recovery of all investments and returns before the other party distributes any profits. In summary, we set up various protective mechanisms to ensure our funds are prioritized for repayment.

Additionally, we have two hard standards when transacting with these cloud companies:

• There must be real assets as collateral, and they must be over-collateralized. When we provide funding, there must be physical assets as support. More importantly, we require over-collateralization. For example, if the other party has GPUs worth $100, we might only provide $70 to $80 in funding, meaning we have at least a 1.3 to 1.5 times over-collateralization rate. This way, in case of any issues, even if the asset value is discounted, our principal still has a relatively safe buffer.
• The company must have signed clients and a good payment record. This ensures that the GPUs we invest in are not just sitting idle burning money but are being effectively utilized and can continuously generate cash flow to repay the funds we provide.

If a company does not meet either of these two standards, we will not proceed with transaction negotiations.

From "Spice" to Tokenization, GAIB's Core Philosophy

PANews: This risk control model sounds very robust. Additionally, we are very interested in the name GAIB; it seems to be related to the famous science fiction novel "Dune." Can you explain the origin of this name and how you view the position of GPUs in the AI value chain?

Kony: Yes, we are all fans of "Dune," and the name GAIB is indeed inspired by it. In fact, it is also an acronym for GPU, AI, and Blockchain. You could say we are a "Global AI Infrastructure Blockchain" (Global AI Infrastructure Blockchain) platform.

This analogy is very apt. In the universe of "Dune," "spice" is the most precious and important commodity. Similarly, in our current AI era, computing power is everything. Whether you are using ChatGPT, Claude, or Perplexity, the core foundational unit you are discussing is computing power. Therefore, the role of computing power is very similar to that of "spice."

As for the position of computing power in the AI supply chain, I like to describe it using a "smile curve." This means that value is primarily concentrated at both ends of the curve.

• One end is the application layer, as they control pricing power and users. However, currently, the curve of this market is not steep enough because most applications only started to become profitable this year; previously, they were burning money without forming large-scale commercial applications. But we are seeing more and more applications emerging.
• The other end, which is the left side of the curve, is AI infrastructure, including GPUs, data centers, and even robotics manufacturing companies.

Regardless of how the application layer evolves, they will rely on core AI infrastructure. As I mentioned earlier, whether you are using the ChatGPT model or the Claude model, it ultimately relies on the underlying GPU chips to provide power.

I like to use the "Visa card" analogy: no matter which bank issues your Visa card, every time you make a transaction, Visa earns a little money from it. GPUs are similar; every time you call any model or use any AI application, GPUs are running, providing computing power, and generating revenue from it. This is why we focus on core AI infrastructure, as this part has enormous expansion potential with the continuous growth of the application side.

How Does GAIB Turn GPUs/Computing Power into On-Chain Assets?

PANews: Financializing AI infrastructure sounds like an excellent entry point. So, what are the next steps? How do you plan to build a complete financial stack on this basis?

Kony: That's a very good question. Within GAIB, we refer to ourselves as an "ecosystem" because we are the bridge connecting off-chain RWA with on-chain DeFi economics. The process we handle for these assets is mainly divided into three steps:

• Asset Digitization: We must first convert these physical assets into digital form. Otherwise, their data, asset value, and other information cannot be reflected and utilized on-chain.
• Asset Financialization: After the assets are on-chain, the next step is to transform them into useful financial tools or products. For example, can they generate revenue? Can they be used as collateral for lending? Based on this, we develop different products.
• Injecting Liquidity: Once these assets exist on-chain, if they have no utility or trading channels, they are essentially just a pile of useless data. Therefore, we have been expanding the uses of these assets, including integration with lending protocols, DEXs, derivatives protocols, etc., to truly integrate them into the on-chain economy and form a closed loop.

These are the three things we are working on. Through this core infrastructure, we can handle any type of AI infrastructure asset. We started with computing power and have proven that this path is feasible; we have successfully tokenized approximately $30 million worth of assets on-chain.

Now, we are preparing to expand into what we believe is the next big trend — robotics. If you view AI as the "brain," then robots are the "body" that interacts with the physical world. Similar to GPUs, robots have physical hardware and are about to undergo a significant transformation in their profit models. The future robot models will be completely different from the large mechanical arms in traditional manufacturing and will become more consumer-oriented.

We recently announced a partnership with a Nasdaq-listed company, Primech, which primarily produces cleaning robots. We are exploring the tokenization of these robots because they have adopted a new business model we call "Robot-as-a-Service" (RaaS). In this model, we have both hardware assets and can obtain stable monthly income, which is a perfect model for us to create a product that provides users with stable AI-related returns.

AI Dollar, GAIB Token, and Ecosystem Outlook

PANews: This sounds very exciting. You mentioned integration with lending protocols; DEX integration is relatively permissionless and easier to achieve. But how are you advancing in the lending market? Can you share some specific collaboration agreements?

Kony: In the lending market, we are about to integrate with Morpho and many other similar protocols. Additionally, there are various types of lending protocols available on different blockchains that we can utilize. For example, on Plume Chain, there is a lending protocol specifically designed for RWA, so we are working hard to integrate with as many blockchains as possible to allow our assets to gain as broad an application as possible.

PANews: Last month, there were some cases in the NFT space that creatively released liquidity from "illiquid assets." I was wondering if someone could create an "AI strategy," using these tokenized AI assets to profit from transaction fees and then reinvest the earnings into more AI infrastructure?

Kony: That's an interesting idea. This is also one of the reasons we are launching an AI-supported stablecoin or synthetic dollar — which we call "AI Dollar." The goal of launching AI Dollar is to make it a universal "umbrella" covering all assets on our platform. The value of AI Dollar will be supported by all the different types of tokenized AI infrastructure assets we introduce, including computing power and robotics.

This way, users have a unified unit that they can easily use to obtain returns, and it can be integrated into all the DeFi protocols they desire. Therefore, AI Dollar is a single entry point we provide for users to access the entire world of AI infrastructure.

PANews: How can users earn returns through AI Dollar? Do they need to stake it on your platform?

Kony: Yes, similar to other models. For AI Dollar, you can stake it and receive a staked receipt version. This staked receipt will continuously generate returns from the underlying computing power and robotics assets.

PANews: So, what is the vision and role of GAIB's native token in the entire ecosystem?

Kony: The GAIB token is a crucial element in our entire ecosystem. It is not just an ordinary governance token; it has real utility.

As I mentioned earlier, GAIB is an infrastructure platform. One of the core parts of building this infrastructure is our node network, which we call the "verification network" or "node coordination network." This network requires all tokenized GPUs to continuously run a node and report data to our network to ensure that these assets genuinely exist, operate normally, and generate returns.

To ensure the security of this network, we need users to stake our GAIB tokens. We employ mechanisms from some re-staking protocols. This means that GAIB tokens provide economic security for the network we offer.

Secondly, the GAIB token is, of course, at the core of all incentive measures in our ecosystem. Whether it is additional returns, extra incentives, extra rewards, or DeFi integration activities, all the behaviors we encourage will be driven by the GAIB token.

Therefore, the GAIB token is central to GAIB's operations, both at the technical infrastructure level and in terms of governance and incentives.

PANews: Finally, we see many decentralized computing power providers in the market, such as Io.net and Akash, but they seem to focus more on Web3 cloud infrastructure. Do you think GAIB, which focuses on serving Web2 market companies, will intersect with these Web3 projects in the future?

Kony: I think their reasons for existing are different. Decentralized computing markets like Akash or Io.net are intended to be aggregators, bringing together various idle resources, whether consumer-grade GPUs or institutional-grade GPUs, and providing users with a unified API to access this computing power.

This model does suit certain users because it may be cheaper for small-scale deployments or small-scale use cases. However, if you need to deploy at scale, such as requiring thousands of GPUs to train a large model or provide production-level services, you may still need to talk to those large traditional cloud companies or emerging cloud companies we are collaborating with.

So I believe the market is large enough to accommodate their respective niche products and services.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。