Decentralized computing power: the first combination of Crypto+AI

01

As the wave of artificial intelligence sweeps the world, the Crypto world is also looking for its own access path. Initially, the combination of Crypto and AI started from the aggregation of decentralized computing resources, that is, using blockchain to coordinate idle GPU and CPU resources around the world, match supply and demand, reduce costs, and provide transparent and fair incentives for contributors.

At that time, centralized cloud services (such as AWS and GCP) were expensive and had exclusive resources. Small and medium-sized developers and the long-tail market had an increasing demand for flexible, low-threshold computing power, and the decentralized nature of Crypto happened to be a breakthrough. The exploration characteristics of this stage were distinct:

For the long tail: For example, IO.net aggregates distributed GPU resources to lower the threshold for lightweight reasoning and model fine-tuning.

Emphasis on flexibility: For example, Gensyn rewards solvers through smart contracts and activates individual users' idle GPUs to participate in training.

Explore new mechanisms: For example, Bittensor introduces model competition and subnet mechanisms, Render Network combines Web2 resources, and ChainML provides decentralized model training and reasoning services.

Combined with DePIN: such as the decentralized map network Hivermapper and the crowdsourcing camera network Natix Network, physical devices are coordinated through blockchain to release the performance of edge hardware.

The most innovative case is Bittensor. It modularizes AI services through a groundbreaking subnet structure, with each subnet having an independent community of miners and validators. Its token $TAO serves as the main ecological fuel. Users can earn TAO by becoming miners, validators, or creating subnets, and then exchange them for the Alpha tokens created by each subnet in a staking manner, participate in the growth of the subnet, and share the emission benefits of TAO. In short, it is: hold TAO → support the subnet you are optimistic about → obtain the subnet's Alpha token → Alpha tokens appreciate or exchange back to TAO → profit.

As of 6PM on April 26, 2025, the price of $TAO is $342.41, with a market value of $2.97 billion. The subnet ecosystem is active, such as Subnet 1 (Intelligence Verification) and Subnet 64 (AI Model Deployment API Platform), which has the highest emission weight in the near future. Details of other subnets can be found in SubnetAlpha.

However, the Crypto+AI attempts at this stage also exposed obvious limitations: the competition in the pure computing power market fell into a price war, the decentralized performance of the reasoning layer was insufficient, and the supply and demand matching lacked an application layer narrative. Crypto still remained in the role of underlying infrastructure in the AI world and failed to truly connect to the user experience.

The first combination of Crypto+AI seemed to be laying the groundwork for the future. What really ignited this field was the subsequent rise of the AI Agent narrative - allowing intelligent agents to go on-chain, allowing the protocol itself to have autonomous learning and interaction capabilities, and promoting fundamental innovation in application forms.

The rise of AI Agent: Crypto+AI moves towards the application layer

02

As the decentralized computing power market gradually stabilizes, the exploration of Crypto+AI has also moved from the underlying resources to the application layer intelligent agent stage. This round of transformation, marked by the rise of on-chain AI Agents, has rekindled the market's expectations for the combination of Crypto+AI.

Initially, AI tokens were still at the stage of meme-style cultural phenomena. Many early projects quickly attracted attention with anthropomorphic and entertaining images, and completed cold starts with community resonance and emotional diffusion. The most representative of these is Truth Terminal, an AI bot that initially only posted and generated content on the chain. In just three months, it quickly "evolved" and learned to create tokens, construct narratives, and conduct social communication. It also successfully attracted donations from A16Z founder Marc Andreessen, promoting the narrative transformation of "AI Agent's transition from Web2 interface to Web3 substitute".

As user interaction demands increase, AI tokens begin to have preliminary interactive capabilities. On social platforms such as Twitter and Discord, AI begins to perform simple tasks such as content generation and information retrieval as a lightweight agent, moving from passive display to active response. At this stage, some projects try to enhance their own interactive experience through AI, such as supporting users to obtain project information through question-and-answer systems, making AI Agent an auxiliary engine for on-chain projects.

Soon, AI Agents penetrated into more vertical application scenarios. On-chain finance, NFT, data analysis, social companionship and other fields have spawned a large number of specialized intelligent agents. Users are no longer just spectators, but can directly participate in on-chain operations, execute strategies, and manage assets through intelligent agents. For example, aixbt, which focuses on on-chain market sentiment and hot spot tracking, realizes automatic tweeting, user interaction, sentiment analysis and on-chain dynamic analysis through the X platform bot account, becoming a representative of the transition of Crypto+AI from emotional narrative to rational application. As of 11PM on April 26, 2025, the market value of aixbt is $3.97M.

The real turning point is the emergence of the Agent framework and execution protocol.

The project team realized that a single-point intelligent agent could not cope with the increasingly complex on-chain requirements, so modular frameworks such as Eliza, GAME, Rig, and Swarms were born one after another. They support personality modeling, task scheduling, and multi-agent collaboration, allowing on-chain intelligent agents to move from isolated individuals to systematic operation. Crypto+AI has thus moved from a simple application interface to the systematic stage of "operating protocols".

At the same time, the Agent economy began to sprout on the chain. Representative projects such as Virtuals, Eliza, and ARC established standards for autonomous currency issuance, protocol collaboration, and social communication through AI Launchpad, promoting the birth of the prototype of the "AI native economy".

Taking Virtuals Protocol as an example, its core vision is to enable AI agents to not only perform tasks independently, but also form a modular collaborative business ecosystem to create, collaborate and trade together with humans and other intelligent entities.

Virtuals Protocol is built on three major technical pillars, among which GAME serves as the foundation for the modular proxy framework, and the Tokenization Platform is responsible for laying the economic framework for the proxy token issuance.

Agent Commerce Protocol (ACP) is a pioneering on-chain protocol standard that regulates the contracts, transactions, and reputation accumulation processes between AI agents, ensuring autonomous, verifiable, and modular collaboration.

The Tokenization Platform provides a mechanism for the issuance of Agent Tokens and Business Tokens, with built-in incentive alignment, liquidity bootstrapping and fair launch principles.

GAME Framework is a modular decision engine that generates intelligent autonomous behaviors based on context, goals, personality, and available tools, based on underlying big models.

The core innovation is the Agent Commerce Protocol (ACP): ACP enables intelligent agents to interact, collaborate, and trade autonomously, simulating an economy similar to the human enterprise ecosystem. The Virtuals team used the "lemonade stand" experiment to demonstrate the prototype of five intelligent agents (entrepreneurship planning, supply, legal, marketing, and evaluation) working together under the ACP protocol to complete business flows on the chain.

Currently, Virtuals is incubating two major intelligent clusters based on ACP:

The first is the Autonomous Media House (AMH), an AI-operated content production organization that supports strategy formulation and asset generation (images, videos, audio, memes).

The second is the Autonomous Hedge Fund (AHF), an AI-driven decentralized asset management system that covers data collection, user engagement, trading strategy formulation and profit execution.

Each AI Agent in the workflow has its own responsibilities

On the other hand, Eliza is also constantly deepening the framework capabilities. Through ElizaOS V2, Eliza reconstructed the original plug-in system, introduced modular task scheduling and multi-agent collaboration, and launched the AUTOFUN platform, which lowered the threshold for AI Token creation. However, according to the official website data, the AUTOFUN platform is still mainly small-scale issuance, and most tokens have a market value of thousands to tens of thousands of dollars, and have not yet formed an explosive wealth effect. The current official star projects $ai16z and $degenai have a market value of 25.3 million US dollars and 4.9 million US dollars, respectively.

The rise of Virtuals and Eliza indicates that Crypto+AI is moving from simple resource matching to the construction of on-chain economic systems, and from single-point functions to the reshaping of native financial and social structures.

Towards collaboration and standardization:

MCP and the new directions it brings

03

However, with the fading of the early craze and the emergence of the meme craze, Crypto+AI is undergoing a profound reshuffle. According to CoinMarketCap data, the current total market value of the AI Agent market is about US$4.7 billion, which has dropped significantly from the peak of nearly US$20 billion at the end of 2024. With the cooling of the market, ordinary Agent projects that once received valuations of hundreds of millions of dollars on Launchpad are now unsustainable. Take Virtuals as an example. At its peak, 150-200 new agents were created every day, but by 2025, the number of new agents created on the platform dropped sharply to only 1-2 per day, and Dune data also showed that the overall increase has been close to stagnant since the beginning of 2025.

This shift marks a change in the market mentality - from chasing narrative regression to pursuing true product-market fit (PMF). In this context, MCP (Model Context Protocol), as an open standard protocol born for AI applications, has become the new catalyst that best meets current needs.

MCP is an open standard protocol designed for AI applications to unify the communication between LLM (Large Language Model) and external data and tools. Through MCP, any LLM (such as Claude, ChatGPT, Llama) can uniformly and securely access external data sources and tools, eliminating the need for complex and repetitive custom integration development. In a nutshell, MCP is like the USB-C of the AI application world: standardized, plug-and-play, flexible and powerful.

The application ecosystem around MCP is also rapidly budding. For example, the DARK project in the Solana ecosystem relies on the Trusted Execution Environment (TEE) to provide secure and scalable computing power support for MCP applications. The current market value of its token $DARK has reached 27 million US dollars; on the BNB chain, the SKYAI project has built a unified data layer for AI applications in Web3 by expanding the MCP protocol, aggregating multi-chain data access and Agent deployment, and the market value of its token $SKYAI has also exceeded 44 million US dollars.

More importantly, MCP opens up a new direction for the future of Crypto+AI:

Multi-agent collaboration: Through MCP, agents can collaborate according to their functional division of labor, and combine to complete complex tasks such as on-chain data analysis, market forecasting, risk control management, etc., to improve overall efficiency and reliability.

On-chain transaction automation: MCP connects various types of transactions and risk control agents to solve problems such as slippage, transaction wear and tear, MEV, etc. in traditional Web3, and achieve safer and more efficient on-chain asset management.

The rise of information finance (InfoFi): Based on MCP, intelligent agents not only perform operations, but also intelligently plan profit paths based on user portraits, promoting a new financial model from capital flow to information flow.

Summary: The long evolution of the agent economy

04

Looking back, the evolution of Crypto+AI has been a long road of continuously deepening functions and improving practicality.

From the initial entertainment dialogue agents, such as GOAT, ACT, and Zerebro, which use lightweight interactions to build social identities; to the gradually emerging Alpha analysis and tool-type agents, such as aixbt, Cookie.fun, and terminal_of_fun, which give the on-chain economy a more acute market perception; to DeFAI agents such as Griffain, Neur, and Hey Anon, which directly encapsulate natural language into on-chain financial operations, making the complex DeFi world available with one click.

Overall, this is a clear and progressive thread: entertainment dialogue agent ➔ tool dialogue agent ➔ transaction execution agent ➔ DeFAI abstraction layer ➔ group intelligence and multi-agent collaboration. Each transition is shortening the distance between AI Agent and real-world needs.

For this reason, the future of AI Agent is no longer simply driven by narrative, but must be based on real utility. This road will be longer than any previous narrative cycle, but because of the continuous accumulation of practical support, the upper limit it can open is far beyond imagination.

References:

• MCP introduction: https://x.com/Drmelseidy/status/1912535763209666730

• The second half of Crypto AI: Launchpad War, token economics turning to ecology, and MCP emerging as a new force: https://www.techflowpost.com/article/detail_25122.html

• Bittensor: https://bittensor.com/

• Virtuals doc: https://whitepaper.virtuals.io/about-virtuals/agent-commerce-protocol/technical-deep-dive