Editor’s Note: On April 21, at the highly anticipated RWA-themed forum of the 2026 Hong Kong Web3 Carnival, Dr. Xiao Feng, Chairman of Wanxiang Blockchain and Chairman & CEO of HashKey Group, delivered the opening keynote speech and officially launched the 2026 Tokenomics White Paper. (Related reading: HashKey Group Releases Third White Paper in the Web3 Economics Series: Rebuilding On-chain Finance and Tokenization Infrastructure for the Agent Economy Era) In his speech, Xiao Feng provided an in-depth analysis of business models integrating AI tokens, blockchain tokens, and privacy-preserving computing technologies such as fully homomorphic encryption. Below is the full transcript of the speech:
Good afternoon, everyone!
After the information overload from yesterday and this morning, we understand many of you may be feeling a bit overwhelmed. This afternoon’s session focuses on RWA, and to kick things off, we will officially launch the 2026 Tokenomics Whitepaper.
Looking back, HashKey began publishing this series of whitepapers in 2023. In the 2024 version, we specifically introduced the "three-token model," comprising equity tokens, utility tokens, and non-fungible tokens (NFTs). HashKey Group is actively implementing this model ourselves: we have our own utility token, issue NFTs during specific events, and maintain an equity ownership structure listed on the Hong Kong Stock Exchange.
Over the past decade, we have found that for the most fundamental blockchain base protocols, a single token may suffice; however, when targeting applications and B2C users, relying solely on a single layer of tokens—such as utility tokens—is insufficient to establish a robust, comprehensive, and effective economic incentive mechanism. Utility tokens are primarily used for community incentives, while equity tokens serve to incentivize founding teams and shareholders, forming the core of our Token Economics Whitepaper v1.0.
The whitepaper has now been updated to version 3.0. This time, we focus on how the agent economy, driven by AI agents, integrates with cryptocurrency and blockchain. Therefore, my talk today is titled: “Innovation in the Agent Economy Model”—exploring how combining AI tokens, blockchain tokens, and privacy-preserving technologies such as zero-knowledge proofs and fully homomorphic encryption will bring disruptive innovation to the agent economy.
From a business innovation perspective, the two main characteristics of blockchain technology are:
- First, it is a trustless, permissionless open network. Participants do not need to undergo KYC or sign contracts beforehand, which is the most fundamental technical characteristic of blockchain-native business activities. However, this alone is clearly not enough.
- The second characteristic of blockchain is public transparency. It’s hard to imagine the massive issues financial institutions—such as banks—that have extremely high requirements for privacy and compliance would face if they were to move their entire business processes onto the blockchain: it would amount to “data exposure.”
Neither financial institution data nor highly privacy-sensitive medical data can operate directly on a public, transparent blockchain.
On the other hand, AI has indeed unleashed tremendous economic creativity. Especially since AI has evolved from large models to AI agents, the industry is actively exploring how agent-based economies could unlock more than ten times the commercial value. However, the same issue of data transparency arises here. Under conditions of complete data transparency, the AI agent economy suffers from significant flaws—and addressing these flaws requires privacy-preserving computation technologies.
Looking back at the development of blockchain, nearly 16 years since the launch of the Bitcoin network in 2009 have fully demonstrated its significant commercial and economic value. However, due to the public and transparent nature of public blockchains, many compliant business applications cannot run directly on them. In response, in 2015, traditional banks and regulatory authorities around the world introduced the concept of "consortium blockchains" or "permissioned blockchains." The emergence of consortium blockchains has, to some extent, alleviated privacy concerns—only authorized nodes are permitted to join and share data within defined permissions.
However, this mechanism has significant flaws. Over the past decade, we have seen two major consortium blockchain organizations: R3, involving global banks, and Hyperledger, led by IBM. Yet, after ten years, neither has produced applications with broad commercial potential. As a result, an industry viewpoint emerged: consortium blockchains may not be blockchain at all. In that context, this perspective held some merit.
But today, with the rise of tokenization and the tokenization of traditional financial assets, consortium blockchains are making a comeback. According to our understanding, nearly all major global banks are already operating a permissioned blockchain internally. However, these internal bank permissioned blockchains are often single-node systems used solely for internal verification, which we refer to as "private blockchains."
Why does it return?
When a globally renowned bank offers tokenization services to its customers—such as tokenized deposits—it does not need to address trust issues or rely on third-party nodes for validation. Customers already trust the bank and simply use tokenization technology within the bank’s existing account system to complete cross-border transfers from New York to Hong Kong in just two minutes. Without tokenization, such a transfer could take two days. As a result, private blockchains have led the way in revival.
However, private blockchains also have limitations. When customers of two different banks conduct cross-border, interbank transfers, a broader network beyond a single private blockchain is required, bringing consortium blockchains back into focus. For example, SWIFT is collaborating with nine major global banks to explore how blockchain and deposit tokenization tools can solve cross-border, interbank fund transfers. This marks a renewed interest in consortium blockchain technology. Yet, a core challenge remains in interbank collaboration: how much of your internal data are you willing to share with your partners?
At this moment, a breakthrough has been achieved in a new technology: privacy-preserving computation—including zero-knowledge proofs and fully homomorphic encryption. These technologies enable computation while preserving privacy, producing results from encrypted data that are identical to those from plaintext computation. These technologies have existed for some time; I recall that as early as 2016, during Ethereum DevCon in Shanghai, speakers mentioned zero-knowledge proofs, formal verification, and other related techniques. However, to this day, they have not been widely adopted, primarily due to performance limitations—their efficiency and cost remain insufficient to support commercial deployment.
However, according to my understanding, fully homomorphic encryption chips are expected to be launched in the second half of this year, with performance reaching approximately 1,000 operations per second. This is already sufficient to meet the needs of certain commercial scenarios, as many use cases do not require real-time computation and can tolerate delays of 10 minutes or even an hour. With fully homomorphic encryption, the integration of “blockchain tokens + AI tokens + privacy-preserving computation” can enable truly disruptive innovation in the business models of agent economies—each component is indispensable.
We can envision that when privacy-preserving computation algorithms become efficient and cost-effective enough to support large-scale commercial use, blockchain technology will experience another resurgence—届时 we may no longer need private or consortium chains. All data, once encrypted, could be directly uploaded to public blockchains, with robust privacy protection technologies sufficient to meet the highest global compliance standards. This is very likely to occur within the next three to five years. Yesterday, Ethereum’s founder Vitalik also shared Ethereum’s roadmap for the next five years here. He noted that Ethereum does not need to compete to become “the fastest chain,” but should instead remain committed to decentralization and security. Within blockchain’s “impossible trinity,” Ethereum focuses on the first two principles, leaving performance challenges to be addressed by hardware acceleration, algorithmic optimization, and Layer 2 and Layer 3 networks tailored to specific use cases.
Next, I’ll provide an example to illustrate how these three technologies work together to create new business models.
Taking a hospital as an example, medical data is highly valuable, yet demands extremely strict privacy protection. In future token economic models, all businesses will become "token factories." With technologies such as fully homomorphic encryption, hospitals can convert medical data into tokens. Anyone can access these tokens to compute required feature results, but no one can ever obtain the original raw personal privacy data.
Only when these three technologies are combined can traditional business models be truly disrupted and the ultimate form of the agent economy be reached. If only “AI Token + Privacy Computing” are present, the business logic still holds—hospitals can still innovate—but their market reach cannot expand globally. Digital products and services are inherently global in nature. Without blockchain, demand sides would still need to negotiate and sign agreements with hospitals offline and make payments through traditional banks. This clearly does not align with the operation of a “Token Factory.”
What should a true "Token Factory" look like? It should leverage blockchain’s permissionless and trustless nature to transform hospital data into AI tokens and make them openly available worldwide. Any demand party should be able to access the data 24/7, just like using the Bitcoin or Ethereum networks, without cumbersome agreement signing or KYC processes. Each data request consumes tokens, and token consumption automatically triggers payment to the hospital. The perfect integration of these three elements is the ultimate vision of the agent economy.
The same applies to individuals. Suppose your years of health check-up and medical data have been encrypted and stored on the blockchain—you could directly send a request to global insurers: “Here is my encrypted data; use actuarial models to compute my information within a homomorphic encryption environment and offer me the most cost-effective, personalized insurance plan.” Under this model, the way financial services are delivered will be fundamentally transformed. No longer will you need insurance brokers or intermediaries; you won’t be tied to any single financial institution, yet you become a potential customer to all of them. You can seek your own “optimal solution” across the entire network through a trustless, permissionless approach.
Here, I also wish to clarify a common misconception: many people regard AI Token as the currency unit of the agent economy. In fact, AI Token is not currency—it is the “means of production” in the agent economy. Jensen Huang of NVIDIA once proposed a five-layer structure of Token Economics (electricity, computing power, large models, algorithms, applications), which describes the production process of AI agent products or services. However, when you want to purchase or consume these services, you must pay with currency—and that currency must be digital currency. AI agents cannot use traditional fiat money; they require a currency that is programmable, infinitely divisible, and capable of real-time settlement.
Imagine an AI agent calling a hospital’s data API—it cannot follow the traditional banking transfer process of “I’ll send payment first, and you provide the service after it clears tomorrow.” It must enable real-time value settlement. Moreover, such machine-to-machine interactions may require payments of just a few cents or even fractions of a cent per transaction, which traditional banking payment systems cannot support due to their high transaction fees. Therefore, programmable digital currency is the true “blood” flowing through the agent economy.
In addition, the agent economy will give rise to new digital assets. In this system, not only currency but also assets need to be tokenized. Why tokenize? Because without tokenization, machines cannot read or utilize them. Existing currencies such as the US dollar, Hong Kong dollar, and Chinese yuan are not programmable; only after tokenization can they be recognized and accessed by AI agents.
To summarize finally.
The AI agent economy is a trustless, permissionless business model that will cause commercial costs to plummet. In traditional societies, maintaining business trust requires an enormous and expensive system—including accountants, lawyers, courts, public security agencies, and even prisons—all designed to prevent and punish commercial misconduct. The entire society bears the cost of operating this vast system. However, in a trustless blockchain-based business network, these costs will be entirely eliminated.
Under this new low-cost, high-efficiency business system, entirely new asset classes will emerge. The crypto world includes “native digital assets” like Bitcoin and Ethereum, as well as “twin digital assets” formed by tokenizing real-world assets. Similarly, the AI domain will also give rise to its own twin and native digital assets. Native AI digital assets will proliferate significantly over the coming years, creating entirely new business models and demanding a completely new financial services infrastructure.
The financial system we know today was designed for humans, but in the future, a new generation of financial services—and even an entirely new capital market system—will emerge, specifically tailored for AI, machines, and AI-native digital assets.
That concludes my presentation for today—thank you all!