Is the investment trend a windfall or a bubble? What value is there in the AI + Web3 track?

Is the "AI + Web3" narrative a real technology fusion or another conceptual package? This article is based on an article written by TinTinLand and curated and contributed by Foresight News. (Synopsis: Can the AI track escape the stigma of hype?) Inventory of three AI Agent projects that embrace cryptocurrencies) (Background supplement: Grayscale News: The first quarter of 2025 is the most optimistic about these 20 major cryptocurrencies, focusing on DeFi, AI Agents, and Solana ecology) In 2025, the "AI + Web3" narrative popularity is still unabated. According to Grayscale's latest report, released in May 2025, the overall market capitalization of the AI Crypto track has reached $21 billion, nearly fivefold from $4.5 billion in the first quarter of 2023. Behind this wave, is there a real convergence of technologies, or is it another concept packaging? From a macro perspective, the traditional AI ecosystem has revealed more and more structural problems: high model training threshold, unguaranteed data privacy, high monopoly of computing power, black-box reasoning process, and unbalanced incentive mechanism...... These pain points are highly consistent with the native advantages of Web3: decentralization, open market mechanism, on-chain verifiability, user data sovereignty, etc. The combination of AI + Web3 is not just a superposition of two hot words, but a structural technology complementarity. Let's start from the core pain points facing AI, deeply disassemble those Web3 projects that are actually solving problems, and take you to see the value and direction of the AI Crypto track. AI service access threshold is too high and expensive Current AI services are usually expensive and difficult to obtain training resources, which is extremely high for small and medium-sized enterprises and individual developers. In addition, these services are often technically complex and require a professional background to get started. The AI service market is highly concentrated, users lack diverse choices, call costs are opaque, budgets are difficult to predict, and even face the problem of monopoly of computing power. Web3's solution is to break platform barriers through decentralization, build an open GPU market and model service network, support flexible scheduling of idle resources, and motivate more participants to contribute computing power and models through on-chain task scheduling and transparent economic mechanisms, reducing overall costs and improving service accessibility. Render Network: focuses on decentralized GPU rendering, also supports AI inference and training, and adopts a "pay-per-use" model to help developers access image generation and AI services at low cost. Gensyn: Build a decentralized deep learning training network, use the proof-of-compute mechanism to verify training results, and promote AI training from platform centralization to open collaboration. Akash Network: A decentralized cloud computing platform based on blockchain technology, developers can rent GPU resources on demand for deploying and executing AI applications, which is a "decentralized version of cloud computing". 0G Labs: Decentralized AI native Layer-1, which greatly reduces the cost and complexity of executing AI models on-chain through an innovative storage and computing separation architecture. Lack of incentives for data contributors High-quality data is the core fuel of AI models, but under the traditional model, data contributors struggle to get rewarded. The opaque and repetitive nature of data sources and the lack of feedback on how they are used make the data ecology inefficient for a long time. Web3 provides a new solution to formalization: a clear closed loop of collaboration and incentives between data contributors, model developers, and users through cryptographic signatures, on-chain rights confirmation, and composable economic mechanisms. Representative project OpenLedger: Innovatively proposed the concept of "Payable AI", which combines data contribution, model call and economic incentives to promote the formation of a data economy network for AI chain collaboration. Bittensor: A complete incentive system with TAO rewards, Yuma consensus mechanism, subnet precision incentives, knowledge collaboration, etc. as the core, directly links data contribution with model implementation results, and enhances the overall value contribution. Grass: AI data network collects user browsing behavior data through plugins, contributes to on-chain search engine training, and rewards users according to data quality, creating a community-driven data sharing mechanism. Model black-boxing, AI inference cannot be verified The inference process of current mainstream AI models is highly black-box, and users cannot verify the correctness and credibility of the results, especially in high-risk fields such as finance and medical care. In addition, models may be subject to tampering, poisoning, and other attacks, making it difficult to trace or audit. To this end, the Web3 project is trying to introduce zero-knowledge proof (ZK), fully homomorphic encryption (FHE), and trusted execution environment (TEE) to make the model inference process verifiable and auditable, and improve the interpretability and trust foundation of AI systems. Representative Project Sentient: Innovative model fingerprinting technology ensures that call behavior can be traced, improving the transparency and tamper-proof ability of model use. Modulus Labs: Using ZK technology to cryptographically verify the model inference process and realize the new normalization of "trusted AI". Giza: Using zero-knowledge cryptography to compute machine learning inference on-chain, thereby improving transparency and trust in AI model deployment. Privacy and security risks The AI training process often involves a large amount of sensitive data, and faces risks such as privacy leakage, model abuse or attack, and lack of decision-making transparency. At the same time, the ownership of data and models is vaguely defined, further exacerbating security risks. With the immutability of blockchain, cryptographic computing technology (such as ZK, FHE), trusted execution environment and other means, the security and controllability of AI system data and models in the whole process of training, storage and call are guaranteed. Phala Network: Provides Trusted Execution Environment (TEE) support to encapsulate critical computing in secure hardware to prevent data leakage and model theft. ZAMA: Focus on fully homomorphic encryption (FHE) technology, so that model training and inference can be performed in an encrypted state, enabling "computation without clear text". Mind Network: Build a decentralized AI data sharing and inference platform that supports privacy protection, and realize data security sharing and privacy computing through front-end encryption technology (homomorphic encryption, zero-knowledge proof, etc.). Vana: An AI identity generation application designed to give users back ownership and control of their data, ensuring its privacy and security. AI Model Copyright and Intellectual Property Disputes Current AI model training makes extensive use of Internet material, but often unauthorized use of copyrighted content leads to frequent legal disputes. At the same time, the copyright ownership of AI-generated content is not clear, and there is no transparent mechanism for the distribution of rights and interests between original creators, model developers, and users. It is not uncommon for models to be maliciously copied or misappropriated, and it is difficult to protect intellectual property rights. Web3 stores the model establishment time, training data source, contributor information, etc. through the on-chain rights confirmation mechanism, and uses tools such as NFT and smart contracts to identify the copyright ownership of the model or content. Story Protocol:...