From hype to reality, where is the future of Web3 AI?

Original Title: Post-AI Agent Bubble: Where's the Real Value in Web3 AI? Original authors: @Defi0xJeff, @steak_studio Chief Original text compiled by: zhouzhou,

Editor's note: After the AI Agent bubble burst, Web3 AI enters a new stage: moving from hype to infrastructure and practical applications. Decentralized AI (DeAI) has gained attention due to the demand for data privacy and autonomy, leading to the gradual rise of ecosystems like Bittensor. The future will focus on the combination of agents with real utility and underlying technologies, driving industry maturity.

The following is the original content (for ease of reading and understanding, the original content has been reorganized):

The market value of AI agents has soared to over $20 billion within a few months, only to crash just as quickly. However, this field is gradually maturing: infrastructure, decentralized AI, and practical applications are becoming dominant. The next wave is forming, and why should we pay attention to it? Here are the reasons.

In the fourth quarter of last year, we witnessed explosive growth in the "AI agent" sector, growing from zero to over $20 billion in just a few months—from those funny, charming, and entertaining AIs, to those financial agents that promise to make you rich overnight through trading and investing.

That's it, there are also various "smart agents that make you rich" and a whole bunch of investment DAOs—DAO organizations composed of humans or AI (3,3) used to invest in other AI agents.

From speculation to infrastructure

We have all learned that when a new field emerges (especially against the backdrop of new catalysts like Web2 AI, Trump supporting Crypto, and AI), people simply do not care about the fundamentals. As long as a project looks lively, has a cool presentation, and is filled with topics, its market value can easily reach hundreds of millions of dollars.

@virtuals_io has become a representative of this ecosystem, perfectly executing market promotion, capturing the public's attention, telling the most compelling stories, and building the best narrative. This not only attracts developers to build projects on it but also draws retail traders to participate in speculation.

Then there is @elizaOS, which follows a different path - open-source AI, allowing any developer to use this "gold mining shovel" to create their own intelligent agents. Around this concept, a large community quickly formed, adopting rapid growth, with stars and forks on GitHub increasing rapidly (and they are still rising now).

Some leading projects in Virtuals once had valuations that broke through 5 billion dollars, with Eliza's peak being about half of that. Other standout projects also reached valuations in the tens of millions, such as AIXBT, which had a market cap of a billion at its peak.

Of course, the situation is completely different now. Newly launched and well-performing smart contracts have an average market value of only 3 million to 10 million USD; established and continuously operating ones are between 10 million to 50 million USD. The valuation ceiling has been significantly compressed, and the total market capitalization has fallen from its peak of 20 billion USD to the current range of 4 to 6 billion USD.

Infrastructure Dynamics & Web2 Acceleration

The current market values "pure fundamentals" more - especially infrastructure and decentralized AI, as AI models in the Web2 realm are rapidly evolving: Meta's Llama, OpenAI's GPT, Musk's Grok, DeepSeek, and Alibaba's Qwen are releasing new generations of more efficient models every month. For example, as soon as the latest image generation feature of ChatGPT was released, it immediately sparked a viral craze for Ghibli-style images.

What's more, the consumer layer of Web2 is evolving at a faster pace – as AI models become more and more capable, many features that would not have been possible are now being implemented. Tools like Lovable, Bolt, Cursor, Windsurf allow developers to iterate on products faster. Agentic workflows and AI agents have become ubiquitous, with minimal barriers to entry and virtually zero user switching costs – what you don't like about the app? Find an alternative with a better UI and a better price in no time.

Data Sovereignty Awakening

At the same time, more and more people are starting to think: "With so many intelligent agent apps using centralized AI everywhere, who actually owns my data? Where is my data going? I shared private conversations with the AI; will it really keep them confidential? Or will it be taken elsewhere?"

This concern has become more significant because @OpenAI recently updated the "memory" feature of ChatGPT – it can now review all your conversations to provide more personalized responses. This means that everything you say to the AI is remembered and can be referenced repeatedly.

Brother... these features are indeed pretty cool, and they are likely to spark a wave of "personalized AI agents", such as dedicated AI assistants, AI co-drivers, personal secretaries, AI psychological counselors, companion AIs, and so on.

But you can imagine that if these AIs use centralized technology, and your data is held or controlled by "others," the consequences could be very serious.

The rise of decentralized AI (DeAI)

Last year, I made several predictions, one of which is: decentralized AI will start to emerge in the second quarter of 2025. By then, AI infrastructure related to privacy protection, data transparency, verifiability, and user ownership will receive more attention and broader usage - because the demand for these will only grow stronger.

The current trends can be divided into three main categories (although there are overlaps between them):

·Web2 AI Venture Capital Trends

Companies incubated by YC (Y Combinator) are crazily launching vertical AI agents; a16z (Andreessen Horowitz) is also positioning itself for the future consumer AI trend through various investment arguments; Perplexity AI recently launched its own AI fund, directly investing in the AI track.

·Web3 AI Venture Capital Trends (DeAI Infrastructure Investment, Distributed Training, Inference Networks, etc.)

·Web3 AI Retail Investor Trends (AI Agent Ecosystem, Consumer-Facing Agents, AI Consumer Applications)

Web2 and Web3 AI: Completely Different Atmosphere

For Web2, the total addressable market (TAM) is much larger than that of Web3 — in other words, many enterprises are looking to leverage AI to transform/optimize their business, enhance workflows, resulting in more potential customers, higher conversion rates and sales, better customer retention, reduced management costs, and more efficient operations — therefore, these enterprises will seek solutions that can address their specific pain points and meet the needs of their vertical segments.

This "optimization demand" has attracted many young entrepreneurs to try using AI agents to improve traditional workflows. Compared to traditional SaaS, AI agents can significantly reduce costs or bring in more leads, allowing these startups to charge higher subscription fees for their services (this is why we see many startups achieving 7-8 figure ARR within a few months).

The trend of Web3 venture capital is very different. This is because blockchain is inherently an ideal infrastructure for decentralized AI, such as: verifiable/immutable transaction records, a trustless environment, decentralized computing power, and AI reasoning and training under minimal trust assumptions (there are quite a few terms, but you get it).

In summary: the future direction is - people know how their data is processed, understand the reasoning of AI, own their data, models, and application scenarios, and are incentivized to share without censorship. At the same time, Web3 VCs are continuously investing in this future.

Why do retail investors like AI agents (even if they haven't understood DeAI yet)

For retail investors in Web3, DeAI is difficult to understand because it requires mastering a large number of terminologies and figuring out what the core values are (sometimes it feels like an alien language). Therefore, they tend to flock to those projects that are the "easiest to understand"—such as starting with chatbots that are articulate, humorous, and interactive.

However, as they continue to stay in this industry, they will realize that these things cannot create sustained value for users (yes, many AI entities are actually useless and lack originality). This awakening, combined with the current sluggish and volatile market environment, has led to a market cleanup—useless entities are being eliminated, while those that are useful, despite a significant drop in valuation, are still surviving.

People are beginning to realize: there must be truly useful AI products and real-world applications. This awakening drives teams to either develop genuine products or integrate with AI projects that have real technology, such as @AlloraNetwork and @opentensor (Bittensor).

This transformation has two benefits:

1. Encourage people to pay attention to those infrastructures that are not easily understood but are truly valuable;

2. Allow AI entities to obtain practical application scenarios that can be presented to the community.

Before this change: the agent would only perform basic functions (chatting, post analysis)

After this change: the agents master more advanced and practical skills (AI-driven betting, trading, market making, mining, etc.)

Agents like @AskBillyBets and @thedkingdao have become representatives showcasing the capabilities of the Bittensor subnet, bringing cool technology into the public eye.

Bittensor ecosystem

I think the particularly interesting aspect of the Bittensor ecosystem is that it is a decentralized AI ecosystem where you can directly participate in investments. Currently, most DeAI projects are still in very early stages, and only VCs or strategic investors can participate; ordinary people can't invest at all because many projects haven't issued tokens yet.

But on Bittensor, anyone can stake their $TAO into the subnet they want to support, thereby exchanging it for the subnet's Alpha Token (which is equivalent to 'going all in' on these DeAI projects).

I have previously complained about the poor user experience of bridging and trading, but in terms of technology, products, and atmosphere, especially with teams like @rayon_labs, they are really strong and have a great vibe.

What I like about Rayon Labs is that they are creating consumer-focused UI/UX optimized products. Given that the dTAO mechanism determines how much incentive emission each subnet can obtain based on the market, as well as the valuation of that subnet, it makes "whether each subnet can build products that are easy to understand and use" increasingly important.

Rayon has many cool subnetworks (the coolest might be Gradients, which is an autoML platform where you can easily train models), and even cooler is their latest flagship product—the Squad AI Agent platform, where you can create an agent by dragging and dropping modules (similar to a node builder style, like @figma).

Final note: I am currently in the early stages of gaining a deeper understanding of Bittensor, and I will write a dedicated article later to share some interesting findings and how to seize opportunities within it. If you're also interested in understanding other trends and changes in the market, you might want to check out the article I recommend below.

"Original link"

：