Web3 has seen a number of innovation trends over the last half decade, each with a strong appeal and market. While some of these trends have faded, others have held strong (staking, for example), evolving into more mature features of the Web3 experience.
RWAs have been an interesting trend in the last few years, showing a hint of what might be possible with tokenized real world assets. However, much of these use cases are focused on financial assets, acting as an on-chain holder for another financial asset. While this can certainly have its place, it has been a let down to many in the market that we aren’t seeing a much wider diversity of these use cases. There are glimpses of new use cases here and there, and when they appear it sends the message that we have a lot left to discover in the world of RWAs.
Even Vitalik Buterin has expressed this sentiment, sharing on X his desire to see much more diversification for the way RWAs are used. The good news is that this diversity may be close to exploding, as the field of AI seems to be invading every other industry. Blockchain is no different, and AI is making its mark here as well. We can see many different uses for AI in the Web3 industry, but we are starting to realize that the AI industry itself could benefit significantly from Web3 also. RWAs have massive potential here, and stand to gain perhaps the biggest use case of all: tokenizing the key assets of the AI industry. Let’s dig into how RWAs can benefit the AI industry, how AI-related RWAs can help to greatly diversify a portfolio, and how this type of tokenization might work. To gain further insight on the topic, we’ve asked Cloris Chen, CEO of Cogito Finance, to share thoughts on the intersection between AI and RWAs.
> Allow people to buy shares of that company in the form of tokens
— vitalik.eth (@VitalikButerin) August 26, 2024
Isn't this an RWA?
(I support RWAs, I want to see a much greater diversity of them onchain so that apps can use them without being subjected to the systemic risk of one single issuer or asset class)
RWAs Need to Move Beyond Finance
As mentioned above, RWAs today tend to focus far too much on traditional financial assets. This isn’t to say that these assets aren’t good, and that their tokenization isn’t a good idea (the assets are good and so is the tokenization). RWAs tied to low risk assets such as T-Bills can be an excellent part of a diversified portfolio. The fact that T-Bills are unrelated to the crypto market further insulates wild swings, giving the stability you want in a portfolio. All well and good so far, right?
The problems occur when the RWAs become proxies for these financial instruments, giving the illusion of diversity without actually being all that diversity. Instead, we need to find additional ways that tie RWAs to non-financial assets, and then incorporate those assets to portfolios in ways that provide a solid balance. For the AI industry, these assets include the AI models themselves, datasets that are used for training the models, and the GPUs that actually run the models through processing power.
With these three new assets in RWA form, they can help to further balance a trading portfolio. But where do the stand in terms of yield?
According to Chen, “Tokenized AI assets typically carry a high risk/reward profile, but it’s essential to consider their market context—specifically, the rapid growth of AI services. These assets are already in high demand, driven by the exponential rise of AI services. What’s more, AI is here to stay, so tokenizing and investing in such assets will become a common occurrence long-term. While a tokenized AI economy may face lower liquidity during its nascent stage (as compared to traditional financial markets) it creates an emerging market, attractive to a growing number of investors.”
Tokenizing AI
In the field of AI, the possibility of tokenization not only offers unique opportunities for a portfolio, it is something that is sorely needed from the standpoint of the AI assets themselves. GPUs can be prohibitively expensive, especially if maximum usage is needed to train, but not necessarily operate, an AI model. This creates a major efficiency problem, and one where a collection of linked GPUs working in synch, taking in many different jobs, can provide the best usage of the hardware. If the GPU can be tokenized, this usage can be monitored and monetized. Further, the higher cost of a GPU (or many, if the processing required is heavy) is incredibly burdensome, but fractional ownership of the RWA can help to distribute both the costs and the rewards.
Chen discusses how AI models can also be tokenized: “One key advantage is how they solve the issue of enforcing intellectual property rights, which has been notoriously difficult in traditional formats. With tokenization, AI researchers can monetize their models by listing tokenized versions on a marketplace, embedding IP rights directly into the token itself.”
This is a critical advantage, being able to fully control an AI model’s processing without the risk of giving it out and having the IP stolen. With fractionalized ownership, members can buy and sell their holdings, which creates a fully liquid market even though the physical assets are not changing hands.
The datasets needed to train the AI models will need to go through a change in how they are managed as well. Traditional data brokers are used to an uncomfortably strong level of control over their data and processes. With Web3, the process is much more democratic. Chen says of data brokers, “Many traditional data brokers operate in a way that gives them too much control over the data. However, as data providers demand more transparency and control over how their data is monetized, these firms will need to adapt to meet the evolving standards of the market. In other words, we do expect possible partnerships between traditional data brokers and tokenization platforms.”
Looking Ahead
AI is a revolution in its own right, but has found its way into many industries. Web3 is no different, and is fully making use of AI as it looks to redefine what an RWA should be. However, Web3 is able to offer many benefits to the AI industry as well, introducing elements such as fractional ownership, programmable tokens that can evolve to match the useful life cycle of an AI model, and the ability for communities to build and run data sets. We will continue to watch closely as these two technologies find their way, hoping that they will continue to intertwine and find many more benefits to come.
