AI and web3 are the two tycoons that humans have entered the next technological growth point in recent years. After ChatGPT brought a disruptive AI experience, on-chain AI also quietly completed the gorgeous turn from meme to infrastructure, becoming the most certain new track of sustainable explosion in the minds of web3 people.

At the just-passed Hong Kong Consensus Conference 2025, the integration of AI and web3 is also a hot topic. Whether it is the main venue or the sub-venue, the discussion on AI x web3 Incessantly heard. The Biteye team went deep into it and learned about many potential AI projects, bringing you the frontline sharing.

1. AI infrastructure 1. AI Agent launch platform and framework

With the popularity of $AI16Z and $VIRTUAL at the beginning of the year, AI Agent launch platform and framework AI infrastructure construction has been very popular in the past six months. These projects provide developers and ordinary users with a low-threshold platform to own and use AI Agent, and are one of the key directions of this round of AI projects.

0G Labs is the first decentralized artificial intelligence operating system (deAIOS) that connects computing resources, data and models by building an AI-proprietary Layer 1. , build a distributed AI development ecosystem. Developers can use modular tools provided by 0G to build more efficient and transparent AI Agents while receiving fair reward distribution. Users can contribute computing resources and gain benefits to the decentralized network by running nodes.

DeAgentAI is an innovative platform focusing on decentralized AI Agents and is committed to promoting the development of multi-agent technologies (Multi-Agent Systems). Through DeAgentAI, users can create, manage and coordinate AI Agent networks and apply them to various scenarios, such as business automation, data analysis, personalized recommendations, etc.

Autonomys Network is a decentralized infrastructure stack designed to achieve secure and autonomous human-machine collaboration. Anyone can join the network to become part of the node by just a solid state drive (SSD). Users can create exclusive AI agents that can act independently, such as booking services and managing funds, coordinate meetings, and even participate in on-chain governance.

Gaia Network is a decentralized AI infrastructure platform that supports the distributed development and operation of AI agents and applications, and integrates distributed through blockchain technology. Storage, computing and data verification to solve AI privacy, scalability, and accessibility issues. Rather than building a centralized server, GaiaNet builds a distributed network of edge computing nodes controlled by individuals and enterprises to host fine-tuned AI models based on node operators’ proprietary domain knowledge and expertise.

Questflow A decentralized multi-AI Agent network, where users only need to describe their needs, and the AI ​​agent network can complete tasks independently. Multiple AI agents can perform tasks seamlessly and leverage the collective intelligence and capabilities of these AI Agents to achieve efficiency and speed that a single AI model or humans cannot match.

2. Decentralized AI

Decentralized AI is the ultimate goal of on-chain AI. At present, many projects are in the direction of computing power, data, models, etc. This goal is constantly working hard, hoping to break the monopoly of large companies over LLM through decentralized means and help the public gain ownership of data and models.

Vana is committed to building a decentralized user data sovereignty platform to turn personal data into financial assets. Vana is an EVM-compatible Layer 1 blockchain that allows users to contribute and verify data through the blockchain, generate tokenized assets for AI model training, and through Proof of Contribution and privacy protection technologies ensure data security and quality. Yesterday he received investment from YZi Lab with CZ acting as an advisor.

Hyperbolic is an open-access AI cloud platform that provides users with affordable and scalable GPU resources and AI services by integrating global computing resources. The platform supports AI reasoning services, on-demand GPU access, and monetization of idle hardware, helping businesses, researchers and individuals reduce costs and utilize computing resources efficiently. Hyperbolic provides services to well-known on-chain AI projects such as Virtuals and AI16z, and revealed in this meeting that it will launch its own AI Agent framework to furtherPromote the innovation and application of AI computing.

OpenLedger is a next-generation network focusing on AI and blockchain, providing decentralized economic infrastructure to support developers to obtain high-quality data and fine-tune Special Language Model (SLM) and deploy it as a paid service. SLM is its core advantage, designed for blockchain, DeFi and smart contracts, and at the same time, it uses the Open LoRA framework to efficiently run thousands of fine-tuning models on a single GPU, reducing costs and maintaining high throughput. Combining Datanets data networks and Model Factory, OpenLedger provides complete solutions for data aggregation, model optimization, and secure deployment.

IO.NET is a decentralized computing platform that provides on-demand access to GPU and CPU clusters, designed to eliminate users' expensive hardware or infrastructure demand. Through a decentralized physical infrastructure network (IOG Network, composed of independently operated distributed hardware nodes), users can instantly access distributed GPU clusters through IO Workers, suitable for high-performance computing, AI model training and other scenarios, without the need for self-directed Purchase expensive hardware.

Aethir is an innovative platform focusing on providing distributed cloud computing infrastructure, with Aethir Earth being a bare-metal GPU cloud designed for AI computing tasks that provide high Performance computing capability; Aethir Atmosphere is a cloud GPU network optimized for the gaming industry, supporting low latency, high-quality gaming experience. Users can also add their GPUs to the Aethir platform to become contributors and earn profits.

MinionLab is a decentralized network of autonomous AI agents called "Minions" that run on user devices and are used to go from the Internet. Data mining is real-time, and device owners can receive MINION token rewards by supporting the network and running these autonomous data mining agents. By leveraging idle computing resources, MinionLab provides a more resilient, cost-effective and scalable way to mine data while driving the AI ​​industry towards decentralized and user-driven development.

GAIB is an economic layer solution dedicated to the AI ​​and high-performance computing sectors, aiming to create aA brand new asset class and economic system. Its core philosophy is to treat GPUs as assets and computing power as currency. GAIB solves the current problems of computing resource allocation and financing through decentralized computing networks, segmented ownership and blockchain technology (L1/L2), while meeting the huge demand for high-performance computing resources in the AI ​​era.

Kite AI is a decentralized Layer 1 blockchain platform designed for the AI ​​economy, designed to use its innovative Proof of AI (PoAI) Consensus mechanism to unlock fair access and rewards to AI assets (data, models, and agents). Kite AI is committed to building an open, collaborative ecosystem that allows data providers, AI developers, and end users to benefit from the decentralized AI economy.

Automata is designed to provide middle-layer privacy protection and trackless computing capabilities for decentralized applications (DApps). It provides developers and users with an efficient, trustworthy and decentralized environment to maximize data privacy and user autonomy.

Public AI's vision is to create an open and transparent AI data platform that supports the collection and annotation of multimodal data, including text, audio, video and map data . At the same time, it provides efficient and low-cost AI-assisted data annotation services. The AI ​​assistant will initially annotate the data and then conduct a comprehensive review by the AI ​​validator to ensure the quality of the data. Adopt a fair and transparent Proof of AI consensus mechanism so that everyone can participate in the AI ​​training process by contributing data.

3. One of the important challenges facing the development of verifiable AI

All the most important challenges facing AI development is the opacity of the training process and the inability to guarantee the accuracy of AI output results. At present, many projects hope to achieve verifiability of the AI ​​training process through technologies such as ZKP and TEE to ensure the reliability of the AI ​​output results.

Phala Network is a decentralized cloud computing platform that provides trusted privacy computing and AI inference services for on-chain applications. Its products and services include confidential computing networks based on trust execution environments (TEEs), support for smart contracts, AI model training and inference, and data privacy protection. Phala Network has natural advantages in scenarios where high privacy is required, such as DeFi, DeSci, NFT markets and on-chain AI Agent development, and can be achieved through TEE-technology ensures data security and computational verifiability.

Brevis is a decentralized computing engine that provides verifiable off-chain AI and blockchain computing, combining Zero Knowledge Proof (ZKP) to improve privacy With efficiency, focusing on financial data privacy protection, supporting the expansion of smart contracts and Web3 applications. Its products provide a secure computing environment that ensures the privacy of on-chain AI inference and transactions while maintaining verifiability for DeFi, forecast markets, and privacy-driven AI inference.

Verisense Network is an innovative platform focusing on decentralized data verification and trusted AI, helping developers verify data sources and ensure the authenticity of training data and Integrity; at the same time, it supports audit and verification of the AI ​​decision-making process, provides transparent training data and inference processes for artificial intelligence models, eliminates the "black box" problem, and makes AI decisions more credible. DeFi, forecasting markets and compliance monitoring suitable for AI models can ensure the reliability of on-chain data.

2. AI use cases: potential and expectations

Compared with the rich AI infrastructure, there are still relatively few outstanding AI use cases projects, except for us who are familiar with the recommendation In addition to the special robot AIXBT, the speakers of this event also mentioned that @BukProtocol of the Virtuals ecosystem can realize travel booking business, NBA star commentator @HeyTracyAI, and DeFAI Agent@askjimmy_ai under the ARC framework, etc., which is the application of AI Agent More possibilities are provided.

Narra is the Gamefi AI Agent platform on Berachain. It can use Narra's AI engine to generate real-time dynamic narrative content, interact with players, promote the development of stories, and in Provide a personalized experience in gamified scenarios. In addition, Narra also supports the generation and interaction of AI-NFTs. These NFTs are not only digital assets, but also have intelligent features and dynamic interaction with users.

AI Travel is an AI-powered travel assistant that can help users automatically customize different travel plans through chat; it can also help users call hotels. Order the right room at the best price; and provide ordering and price comparison services for activities such as hotel flights through browser plug-ins.

HeyTracyAI is an AI Agent focusing on sports commentary in the field of basketball, participated by NBA champion Tristan Thompson, which can provide real-time analytical and predictive insights to the game.

AskJimmy is an AI Agent platform focused on the finance and trading fields. Its main goal is to create a decentralized multi-strategy hedge run independently by the AI ​​Agent fund.

3. The transformation of traditional projects to AI is the general trend. Many traditional web3 projects in this event also embraced AI and announced their respective AI turnover plans.

Honey chains such as Sui, Near, Flow and Aptos actively participated in AI-related meetings during this conference. They said in the discussion that the emergence of AI Agent It is of great help to simplify the complex blockchain interaction process, and can introduce more users into the world of web3 and solve the interaction problems that have been plaguing the industry before. The tokenization and mutual cooperation of AI Agent also allow everyone to see more possibilities for AI to be on the chain. In this regard, my husband's chain has expressed the development goals of AI first, and will fully support the development of AI from the aspects of underlying architecture, account innovation, etc. At the same time, through activities such as hackathons, developers are encouraged to innovate and develop AI applications on the chain.

Eigenlayer, which has focused on Restaking services in the past, also said in its co-sponsored AI sub-conference that it is working to build a decentralized trust layer (Decentralized Trust). Provide verifiable cloud services (Verifiable Cloud) to provide on-chain proof for off-chain operations such as AI training and reasoning and prediction, and help the development of verifiable AI Agents.

4. Challenges and the Future

As a hot topic of this Hong Kong Consensus Conference 2025, the discussions and sharing about AI and Web3 are fierce and exciting. While looking forward to a beautiful blueprint, many guests also mentioned that the development of on-chain AI still faces many challenges, including insufficient model reliability, ambiguity of prompt word intentions, storage and hardware restrictions, privacy and security issues. These challenges not only bring technical difficulties to the industry, but will also give birth to huge innovation opportunities. In the long run, everyone is full of hope for the development of on-chain AI and looks forward to the use of infrastructure.The further improvement of the implementation, use case innovation and community collaboration will jointly promote the integration and prosperity of AI and Web3.