In recent years, the AI concept has been everywhere, with large models and decentralized technology intertwining, creating a dizzying array of developments. As a VC firm, we have reviewed hundreds of crossover projects and want to share some hard truths — including the real cycle of the AI industry, why AI cannot do without Crypto, and what we value most and are most wary of when evaluating projects.

Note: The content of this article is excerpted from the roundtable discussion "AI+Web3-The Real Revolution" at the "AI + Bitcoin, the Next-Gen Revolution - BTC Vegas Side Event" held on April 28th.

I. The "Seven-Layer Matrix" and Investment Cycle of the AI Industry

Although giants like OpenAI, Anthropic, and Google are currently riding high and seem poised to dominate everything, our research indicates that the entire AI industry is still in a very early stage.

Being in an early stage doesn't mean one can invest blindly now, especially in overheated areas like AI hardware manufacturing, which require cooler heads. In our view, the AI industry chain can be broken down from bottom to top into seven different layers, and how to invest money in each layer is completely different:

• Layer Seven: AI Agents

• Layer Six: Token Optimization

• Layer Five: Token Distribution

• Layer Four: Large Language Models

• Layer Three: GPUs

• Layer Two: AI Data Centers

• Layer One: Power Infrastructure

Take Layer Two, AI Data Centers, as an example. This is a typical cyclical industry. Over the past two years, global capital has been frantically buying land, building facilities, and expanding with heavy assets. However, there is a time lag between when this capacity is built, powered up, and actually brought to market. Based on our observations, this massive wave of capacity is likely to come online around 2028.

This means that if people are still single-mindedly pouring money into traditional data centers next year or even now, they will likely run headfirst into severe oversupply when everyone opens for business around 2028. Rents and returns will plummet, and capital that jumps in late will have a tough time. Therefore, aligning with the cycle and timing correctly is key.

Back in 2024, we believed there was a significant opportunity in the AI computing power layer, particularly in the crossover area of using Web3 methods for computing power scheduling. So, we focused a lot of effort and capital there. Currently, an AI computing power infrastructure company in which we are a major shareholder is in the final stages of preparing for a Nasdaq listing.

II. Why Are AI and Crypto Inseparable?

Many people often ask: Why does AI need Crypto when Web2's large models are already so powerful? In our view, this is not a forced concept mashup. Rather, there are two things that traditional Web2 cannot solve as AI advances:

1. In the Near Term: Payment — AI Agents Need Their Own "Financial Sovereignty"

Traditional bank cards, PayPal, or gateway payments are all designed for "humans," with an extremely cumbersome identity verification and centralized settlement process. But the transaction model for future AI Agents collaborating with each other is completely different.

Their interactions often involve extremely micro-amounts (e.g., spending $0.0001 to buy a few seconds of computing power), ultra-high frequency, and continuous transactions. If they go through traditional channels like Visa, the fees could exceed the transaction amount. More importantly is the issue of "payment sovereignty": AI Agents have autonomous execution capabilities. As they evolve to a certain level, they will inevitably need to own and have absolute control over their own wallet accounts, rather than forever relying on human physical bank cards or worrying about having their API accounts shut down by centralized giants at any moment. Cryptocurrency payments on the blockchain are currently the only technological means that allows machines to settle autonomously with each other.

2. In the Long Term: Trust — Guarding Against Hallucinations and Credible Audits

Everyone using large models now knows they sometimes fabricate things, the so-called "hallucination rate." In the traditional Web2 black box, when an AI provides a wrong or biased answer, it's difficult for humans to investigate: Is this a random technical hallucination, or is the centralized giant behind it intentionally manipulating the algorithm, poisoning data, or deceiving people?

This creates a hard need for blockchain. Pushing key AI operational data, prompt records, and call paths onto the chain in real-time can create an immutable, traceable, and auditable evidence storage system. This isn't just about defining data property rights; it's a reliable path for humans to supervise and verify whether AI behavior is compliant in the future.

III. Investing in Projects: Using "Honesty" as the Measuring Stick

Every year, we see thousands of business plans for AI + Web3. After reviewing so many stories, the core screening principle we've settled on is actually just two words: Honesty. This sounds like simple common sense, but in the current浮躁 industry, it might be the rarest quality. We break down honesty into two levels:

1. The Team Must Be Honest — Reject Packaging and Patchwork

The resumes of founders and core teams must have no falsification or excessive boasting. In the industry today, we often see hastily assembled teams. To catch a bull market or chase a trend, they forcefully slap together labels like "big company scientist" or "prestigious school background," but there's zero internal team chemistry. Such projects often fall apart at the first sign of a technical bottleneck or market adjustment. A team that can truly succeed needs complementary skills, and the founder's background must be genuine — someone who is earnest about building something.

2. The Product Must Be Honest — Reject PPT-Driven Hype and False Metrics

What a product can actually achieve, how its user data looks — these must be supported by tangible underlying metrics like code and node counts. We've seen too many projects that simply wrap an OpenAI API on the frontend, change the interface, and then dare to claim in their PPT that they are an "autonomous native large model," using a fake demo to fool investors. On the product level, one must be realistic, able to genuinely solve problems.

In the capital markets, dishonest projects driven by hype and fabrication might, in the short term, inflate their valuations highly in secondary markets or exchanges through fancy financial engineering. But because there's no real underlying business supporting them, once the bubble bursts, their ultimate fate is inevitably to go to zero.

Conversely, projects that work diligently and honestly might seem slower initially because they disdain hype and fraud, perhaps even appearing a bit笨 in the eyes of some speculative capital. But because they have a solid foundation, they can often go the distance. In investing, slow is fast. These long-lasting projects are the ones we are willing to invest in and support strategically.

IV. The Most Underestimated Opportunity in 2026: The Ultimate Fusion of AI + Blockchain + Entertainment

Finally, regarding future opportunities that everyone hasn't yet noticed or that are severely underestimated. Based on our research, what we're most excited about is actually the deep integration of AI + Blockchain + Entertainment.

Currently, most market capital is focused on relatively hardcore, dry B2B infrastructure like AI payments and decentralized computing (DePIN). These tracks are important, but the competition is too fierce. People are overlooking the consumer end, which is best at attracting mass users and retaining capital.

Our observation and logic are as follows: As large models and Agents mature, AI, as an efficient labor force, will inevitably replace the vast majority of repetitive white-collar and blue-collar human jobs in the future. When productivity becomes extremely abundant and material costs are extremely low, human society will undergo a fundamental transformation — most people won't need to work forced jobs just to make ends meet. At that time, the demand for吃喝玩乐 will explode. Where will human time and energy go? High-quality entertainment is an inevitable destination.

Future entertainment should fully integrate AI.

Take gaming as an example. Every NPC in the game could have autonomous consciousness, with their own memories, personalities, and social relationships. When players enter the game, they no longer face a repeater bot, but a "living person" who develops emotions based on your actions and words, and might even spontaneously conduct transactions with you on-chain. The integration of AI will lead to a qualitative leap in the freedom and enjoyment of games.

And blockchain's role here is "property rights": it defines land ownership, the uniqueness of rare items in this virtual world, and uses Tokens to establish economic order.

When people no longer need to work in reality and pour a large amount of time, energy, and even assets into such a fun, AI-driven virtual world where assets belong to the players themselves, the commercial value unleashed within will be terrifying. We are currently actively looking for pioneers in this direction. This could be the next narrative that ignites the entire industry.

Disclaimer: The content of this report represents only the author's and the institution's industry research views. It is intended for industry discussion only and does not constitute any investment advice.