Interview with GAIB founder Kony: Breaking the "capital dilemma" of AI infrastructure, how does GAIB turn GPUs and robots into income-generating assets in the DeFi world?

Interview: The Round Trip

Compiled & edited by: Yuliya, PANews

In today's increasingly capital-intensive competition within the AI supply chain, physical assets such as GPUs and robots are becoming the scarcest and most valuable production factors globally. With GAIB announcing its TGE on November 19th , this RWA × AI × DeFi model has garnered even greater attention.

In the new "Founder's Talk" series of "The Round Trip," co-produced by PANews and Web3.com Ventures, hosts 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 the complete business path to turning computing power into a tradable asset.

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

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

PANews: First, could you please introduce yourself and the impetus for founding GAIB? We know you were a venture capitalist at L2 Iterative Ventures (“L2IV”) and also have a traditional finance background. What experience 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 cryptocurrencies, so I joined a large exchange to be responsible for its overseas expansion, leading the acquisitions of other exchanges, wallet companies, esports gaming platforms, and more.

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

The real impetus for me to found GAIB came in late 2023 and early 2024. At that time, the crossover between AI and Crypto began to emerge, and everyone was exploring its possibilities. As an AI enthusiast, I conducted in-depth research on this. Since the release of ChatGPT-3.5, I've devoted a significant amount of time and effort to it, even building an AI agent myself. Back then, there were almost no mature frameworks available, so everything had to be built from scratch, such as how to vectorize data, how to search, how to implement RAG (Retrieval Augmentation Generation), and how to build long-term memory (the context window was very short at the time).

I discovered that my two favorite areas were merging, so I started digging into the Crypto × AI track. But at the time, most companies in the market were doing the same thing:

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

What surprised me was that almost no companies approached the field from a "financial" perspective . This is unusual, as someone with a financial background knows very well that the most important thing in the early stages of an industry is solving the "capital problem." The entire AI industry, especially in AI infrastructure, is experiencing exponential growth in computing power demand, which directly leads to an exponential demand for computing assets such as GPUs. This sub-sector is extremely capital-intensive; building data centers, purchasing GPUs, and deploying infrastructure are all extremely expensive.

I thought to myself, why don't we do something about this? To me, the core spirit of blockchain lies in creating new markets and new assets, which is exactly what DeFi and the "DeFi Summer" have taught us. This led me to the idea of providing financial services for AI infrastructure , as this captures a large portion of the value in the AI supply chain.

I shared this idea with my co-founder, Alex. His background leans more towards the semiconductor and AI fields; his family runs Realtek , one of the world's top seven chip manufacturers, and he also runs a cloud service company called GMI Cloud . He personally experienced all the pain points I mentioned: as a startup cloud service company, securing funding is extremely difficult . There are two reasons for this:

First, GPUs and computing power were still a very new asset class at the time (one or two years ago), and nobody really understood them;
Secondly, cloud services themselves are also very new.

So, Alex and I hit it off immediately. We believed we should start a financial company focused on AI. This has happened in every other industry before, whether it's coal, real estate, or anything else. And so, we decided to create GAIB last year.

ROI of AI infrastructure and the business model of GAIB

PANews: Indeed, the AI field requires huge capital expenditures, and the return on investment (ROI) can take several years. What are your thoughts on the ROI of AI infrastructure like GPUs? Also, how does GAIB collaborate with cloud service companies and help them shorten their capital turnover flywheel?

Kony: Actually, most people may not know that GPUs, as an infrastructure, are actually quite profitable .

First, large AI or gaming companies require a large number of GPUs and typically sign long-term contracts of 2–3 years with cloud service providers to ensure stable computing power. This provides cloud service providers with a robust cash flow.
Secondly, AI computing power available on demand is extremely scarce . For example, a few months ago, ChatGPT launched a new feature that could transform images into specific styles, which at one point caused a computing power shortage even for OpenAI itself, demonstrating that the market is still in a state of obvious supply shortage .
More importantly, the typical return on investment cycle for GPUs is actually 12 to 18 months . This means that, from a revenue perspective, the annualized return on investment (ROI) can reach 50% to 100%, which is excellent compared to other asset classes.

Regarding the second question, who do we partner with? My co-founder Alex's company , GMI Cloud, was naturally one of our first partners and the starting point for our project. But 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 pipeline of projects. Currently, we partner with more than 10 Neo Cloud and Nvidia Cloud partners globally, covering Thailand, Taiwan, Singapore, Hong Kong, and Japan in Asia; the United States and Canada in North America; and Norway, Iceland, and Denmark in Europe.

We tend to partner with companies that are typically Nvidia Cloud partners . To an uninformed audience, this means these companies have passed Nvidia's vetting process and possess the licenses and capabilities to provide hardware and software services. More importantly, they receive official recommendations and preferential treatment from Nvidia in accessing the latest GPUs and high-quality customers. Therefore, most of our partners are Nvidia Cloud partners.

This brings us to our unique advantage in this field:

First, our founding team are operators in this field themselves, and we have a deep understanding of its economic model and GPU business.
Secondly, we have a network of partner clients.
Third, our capital solutions are more flexible and better meet their needs in terms of both underwriting time and duration.

Therefore, our pipeline of collaborative projects has been continuously growing.

How does GAIB guarantee returns and control risks?

PANews: You mentioned partnering 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 return on investment (ROI) across different partners? Also, you mentioned partnering with Nvidia Cloud partners. Does this mean you have a set of standards to guarantee the credibility and operational capabilities of your partners?

Kony: Yes, it's very difficult to ensure that the ROI is completely consistent across different cloud service providers because they each have different fees 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, our focus is not on cost, but on net cash flow recovery . We will assess:

If we give someone $100, when and how will they reciprocate with $100?
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 agreement we reach with them. Sometimes we use a fixed-rate model, requiring a fixed annualized rate of return. In other cases, we prefer to invest directly in the asset itself and then take a percentage of the total income generated, such as 50% to 70%. This model provides stronger protection.

This comes down to the actual structure of the transaction and our experience. We can request various protective clauses, such as priority in recovering our entire investment and returns before the other company distributes any profits. In short, we will set up various protective mechanisms to ensure that our funds are repaid first.

In addition, we have two hard criteria when doing business with these cloud companies:

There must be real assets as collateral, and it must be over-collateralized . When we provide funding, there must be tangible assets to back it up. More importantly, we require over-collateralization. For example, if the other party only has a GPU 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 ratio. This way, in case of any problems, even if the asset is sold at a discount, our principal still has a fairly safe buffer.
The company must have existing signed clients and a good payment history . This ensures that the GPUs we invest in are not just sitting idle burning money, but are being used effectively and generating a continuous cash flow to repay the funds we provide.

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

From "spices" to tokenization, GAIB's core philosophy

PANews: This risk control model sounds very robust. Also, we're interested in the name GAIB, which seems to be related to the famous science fiction novel *Dune*. Could you explain the origin of this name and your views on the role of GPUs in the AI value chain?

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

This analogy is very apt. In the Dune universe, "spice" is the most precious and important commodity. Similarly, in our AI era, computing power is everything. Whether you're using ChatGPT, Claude, or Perplexity, the core building block you're talking about is computing power. Therefore, computing power plays a very similar role to "spice."

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

On one end is the application layer, because they control pricing power and users. However, the curve in this market isn't steep enough yet, as most apps only started to truly profit this year, having previously been burning through cash and not yet achieving large-scale commercial adoption. But we are seeing more and more apps emerge.
At the other end of the curve, on the left, is AI infrastructure, including GPUs, data centers, and even robotics manufacturing companies.

Regardless of how the application layer evolves, they will all rely on the core AI infrastructure. As I mentioned before, whether you're using ChatGPT's model or Claude's model, they ultimately depend on the underlying GPU chip for power.

I like to use the analogy of a Visa card: no matter which bank issues your Visa card, Visa earns a little money every time you make a transaction. The same is true for GPUs; every time you call any model or use any AI application, the GPU is running, providing computing power, and generating revenue . This is why we focus on core AI infrastructure, which has enormous expansion potential as applications continue to grow.

How does GAIB turn GPUs/computing power into on-chain assets?

PANews: Financializing AI infrastructure sounds like a great starting point. So, what's the next step? How do you plan to build a complete financial stack on top of this?

Kony: That's an excellent question. Internally at GAIB, we refer to ourselves as an "economy" because we act as a bridge connecting off-chain RWA with the on-chain DeFi economy. Our process for handling these assets primarily involves 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 used on the blockchain.
Asset financialization : After assets are put on the blockchain, the next step is to transform them into useful financial instruments or products. For example, can they generate returns? Can they be used as collateral for lending? We develop different products based on this.
Injecting liquidity : Once these assets exist on-chain, if they have no utility or trading channels, they are essentially just a bunch 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're doing. Through this core infrastructure, we can handle any type of AI infrastructure asset. We started with computing power and have proven this path is feasible, having already successfully tokenized assets worth approximately $30 million onto the blockchain.

Now, we're preparing to expand into what we believe is the next big trend—robotics . If you think of AI as the "brain," then robots are the "body" that interacts with the physical world. Much like GPUs, robots have physical hardware and are poised for a massive transformation in their monetization models. The future of robotics will be completely different from the large robotic arms of traditional manufacturing; it will become much more consumer-oriented.

We recently announced a partnership with Primech, a Nasdaq-listed company that primarily manufactures cleaning robots. We are exploring tokenizing these robots because they are using a new business model we call "Robot-as-a-Service" (RaaS). This model allows us to have both hardware assets and a stable monthly revenue stream, making it a perfect fit for creating a product that provides users with consistent AI-related revenue.

AI Dollar, GAIB Token, and Ecosystem Outlook

PANews: That sounds very exciting. You mentioned integration with lending protocols, and DEX integration is relatively permissionless and easy to implement. But how are you progressing with the lending market? Can you reveal some specific cooperation agreements?

Kony: In the lending market, we are about to integrate with Morpho and many other similar protocols. Furthermore, there are various types of lending protocols available to us on different blockchains. For example, Plume Chain has a lending protocol specifically designed for RWA. Therefore, we are working to integrate with as many blockchains as possible to make our assets as widely applicable as possible.

PANews: Last month, some cases emerged in the NFT space involving creative strategies to unlock liquidity from "illiquid assets." I'm wondering if anyone could create an "AI strategy" that leverages these tokenized AI assets to profit from transaction fees and then reinvests the profits into more AI infrastructure?

Kony: That's an interesting idea. It's one of the reasons we launched an AI-powered stablecoin, or synthetic dollar—which we call the "AI Dollar." The goal of launching the AI Dollar is to make it a universal "safety net" covering all assets on our platform. The value of the AI Dollar will be backed by all the different types of tokenized AI infrastructure assets we've introduced, including computing power and bots.

This gives users a unified unit that they can easily use to earn rewards, and it can be integrated into any DeFi protocol they want. Therefore, AI Dollar is a single gateway we provide to users to the entire world of AI infrastructure.

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

Kony: Yes, similar to other models. With AI Dollar, you can stake it to obtain a staking certificate version. This staking certificate will continuously generate returns from the underlying computing power and bot 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's not just a regular governance token; it has real-world utility.

As I mentioned earlier, GAIB is an infrastructure platform. One of the core components of this infrastructure is our node network, which we call the “validation 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 are real, functioning properly, and generating returns.

To ensure the security of this network, we require users to stake our GAIB tokens. We employ certain restaking protocol mechanisms. This means that GAIB tokens provide economic security for the network we provide.

Secondly, the GAIB token is, of course, at the heart of all incentives within our ecosystem. Whether it's additional earnings, extra incentives, additional rewards, or DeFi integration activities, all the behaviors we encourage will be driven by the GAIB token.

Therefore, the GAIB token is at the core of GAIB's operation not only at the technical infrastructure level, but also at the governance and incentive levels.

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

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

This model may indeed suit certain users because it can be less expensive for small-scale deployments or small-scale use cases. However, if you need to make large-scale deployments, such as those requiring tens of thousands of GPUs to train a large model or to provide production-level services, you will likely still need to talk to the large traditional cloud companies or emerging cloud companies we are working with.

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