AI and Web3 Investment Opportunities: Prospects and Opportunities from an Investor's Perspective
TechFlow Selected TechFlow Selected
AI and Web3 Investment Opportunities: Prospects and Opportunities from an Investor's Perspective
AI and Web3 are entirely different fields—AI is highly centralized, while Web3 emphasizes decentralization, making them not easy to combine.
Navigating Web3 tides with focused insightsAuthor: Lao Bai, Research Partner at ABCDE
After finishing my last piece on the BTC ecosystem, I was supposed to follow up with one on NFTs and NFTFi. However, NFTs have been unusually cold lately—not just in the secondary market, but even in the primary market, where I haven’t come across any NFT or NFTFi-related projects for about two months. Meanwhile, AI projects are emerging like a geyser. So I’ll keep delaying the NFT article and bring forward this piece on the convergence trend between AI and Web3.
I. First, Let’s Talk About AI Itself
The AI industry was actually close to dying out. You know Illia Lapp, founder of Near, right? He originally worked in AI and was a major code contributor to TensorFlow—the most popular machine learning framework. People speculate he switched to Web3 because he saw no future in AI (specifically pre-large-model machine learning).
But by the end of last year, ChatGPT-3.5 arrived and suddenly revived the entire field. This time it truly represented a qualitative leap, not just another wave of hype or incremental improvement. A few months later, the AI startup wave finally reached our Web3 space. In Silicon Valley's Web2 world, competition has become extremely fierce—capital is FOMOing in, homogeneous solutions are engaging in price wars, and big tech companies are pitting their large models against each other...
However, after more than half a year of explosive growth, AI has now entered a relative bottleneck period. For example, Google search interest in AI has plummeted, ChatGPT user growth has significantly slowed, and the inherent randomness of AI outputs limits many practical applications... In short, we are still very, very far from achieving so-called "AGI—Artificial General Intelligence."
Currently, the Silicon Valley venture capital community holds several views on AI’s next stage of development:
1. There are no vertical-specific models—only general large models combined with vertical applications (we'll revisit this when discussing Web3 + AI);
2. Data from edge devices such as smartphones may become a competitive moat, making edge-based AI an opportunity;
3. Future increases in context length could trigger qualitative changes (currently vector databases serve as AI memory, but context length remains insufficient).
II. Web3 + AI
AI and Web3 are fundamentally different domains. AI requires centralized computing power and massive datasets for training—something highly centralized—while Web3 emphasizes decentralization. Therefore, integrating them isn't easy. Yet the narrative that "AI transforms productivity while blockchain transforms production relations" is so compelling that countless people continue striving to find points of convergence. Over the past two months alone, I've reviewed no fewer than 10 AI projects.
Before diving into new integration trends, let’s first revisit older AI+Web3 projects, which were mostly platform-based, represented by FET and AGIX. Here’s what a professional AI expert friend in China told me: “Most of these earlier AI efforts are largely obsolete today, whether in Web2 or Web3—they’re baggage rather than experience. The real direction and future lie with transformer-based large models like OpenAI’s. Large models saved AI.” Think about that.
Therefore, general-purpose platforms aren’t seen as the ideal model for Web3+AI. Indeed, among the 10+ projects I’ve discussed recently, none fall into this category. Instead, current developments cluster around the following areas:
1. Assetization of Bots/Agents/Assistants
2. Computing Power Platforms
3. Data Platforms
4. Generative AI
5. DeFi Trading/Audit/Risk Control
6. ZKML
Today, I’ll focus in detail on the first area—assetization of Bots/Agents/Assistants—which is both the most frequently discussed and the most saturated segment. Simply put, these projects typically use OpenAI as a base layer, supplemented by other open-source or proprietary technologies like TTS (Text-to-Speech), along with specific datasets, to fine-tune bots that perform better than ChatGPT in certain domains.
For instance, you could train a female English teacher bot, choosing whether she speaks with an American or London accent, adjusting her personality and conversational style. Compared to ChatGPT’s mechanical and formal responses, this offers a significantly improved interactive experience. Recently, there was a DApp called HIM—a Web3 female-oriented game featuring virtual boyfriends—that exemplifies this type.
From this perspective, you could theoretically have numerous specialized bots serving you. If you want to cook Sichuan-style boiled fish, a cooking bot fine-tuned for culinary guidance could assist you with more professional advice than ChatGPT. Planning a trip? A travel assistant bot can offer recommendations and itinerary planning. Or if you're a project team, you might deploy a Discord customer service bot to answer community questions.
Beyond building "GPT-based vertical application" bots, derivative projects are also emerging. Since bots represent "model assetization"—akin to how NFTs turned "small images" into assets—could prompts used in AI also be assetized? In tools like MidJourney, different prompts generate different images; similarly, varying prompts during bot training yield different results. Thus, prompts themselves carry value and can be assetized.
Other projects aim to index and search such bots. When we eventually have thousands or millions of bots, how will users find the right one? We may then need a Web2-style portal like Hao123 or a search engine like Google to help "locate" the appropriate bot.
In my personal view, bot (model) assetization currently faces two drawbacks and points toward two directions:
Drawback 1 – Extreme homogenization. This is the easiest-to-understand AI+Web3 niche, somewhat resembling NFTs with added utility features. As a result, the primary market is turning red-ocean fast, becoming highly competitive. But since most rely on OpenAI underneath, there’s little technical differentiation—teams compete only on design and operations.
Drawback 2 – Sometimes it feels like putting Starbucks membership cards on-chain as NFTs: while a good attempt at mainstream exposure, for most users a physical or digital card remains more convenient. The same issue applies to Web3-based bots. If I want to learn English from a bot or chat with simulated versions of Musk or Socrates, why wouldn’t I just use the Web2 platform http://Character.AI? Isn’t that better?
Two potential directions: One near- to mid-term possibility is on-chain models. Currently, these models resemble Ethereum NFTs—tiny pictures—whose metadata mostly points to off-chain servers or IPFS, not purely on-chain storage. Given that models range from tens to hundreds of megabytes, they’re naturally stored on servers.
But with rapidly declining storage costs (e.g., 2TB SSDs for RMB 500) and progress in storage-focused projects like Filecoin FVM and ETHStorage, it should become feasible within the next two to three years to store hundred-megabyte-level models fully on-chain.
You might ask: What’s the benefit of going on-chain? Once on-chain, models can be directly invoked by smart contracts—making them more crypto-native and enabling more complex interactions. It’s reminiscent of Fully On-Chain Games, where all data is natively on-chain. Some teams are already exploring this path, though it remains very early stage.
Another mid- to long-term direction: If you think carefully about smart contracts, they’re actually less suited for human-machine interaction and more ideal for "machine-to-machine" communication. On the AI side, concepts like AutoGPT are emerging—a "digital twin" or "virtual agent" that doesn’t just chat with you but executes tasks based on your instructions, such as booking flights, reserving hotels, buying domains, or setting up websites...
Now, would it be easier for an AI assistant to operate your bank accounts or Alipay, or to manage a blockchain address and conduct transactions? The answer is obvious. Could we envision a future where numerous AI assistants—integrated with AutoGPT-like capabilities—automatically handle C2C, B2C, or even B2B payments and settlements via blockchain and smart contracts across various task scenarios? At that point, the boundary between Web2 and Web3 would become extremely blurred.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
Lao Bai
