Author: Little Pancake, Deep Tide TechFlow

1. WAIC should be held earlier, so that after the visit, tech can be cleared out earlier.

There are too many peak signals: various side events becoming crypto-related, an abundance of beautiful women, products on-site being highly homogeneous… a track has already transformed from a geek's game to a feast for the masses.

Whether in the primary or secondary market, when everyone starts believing the same story, consensus becomes too concentrated, and risks begin to accumulate.

In a bull market, the bulls standing by your side will all become sellers in the scramble during a downturn.

More killing more!

2. Large models have no real moat

The foundation of AGI is still Scaling Law.

The relationship between model capability, parameter scale, training computation, and data quality still follows a power-law relationship.

Who can obtain GPU computing power cheaper, acquire higher quality data, hire better researchers, and efficiently spend money.

Therefore, the moat of large model companies is weaker than many people imagine.

Model capabilities will continue to converge, leading advantages will be chased, and prices will keep dropping.

The real money-making opportunity will be those selling water to all model companies, since everyone is capitalizing on the expenditures of a few big firms.

GPU, HBM, high-speed interconnections, data centers, electricity, data services…

Standing at the upper reaches of the arms race and claiming a slice of the pie, or even just being a middleman connecting resources, are the most profitable businesses today.

3. Data Bottleneck is the potential opportunity seen at this conference

After discussing with some friends, I found that the companies making money in this wave of AI are very low-key and do not even come to exhibit; one direction is data.

The success or failure of unified multi-modal representation essentially depends on high-quality multi-modal training data. There is a severe shortage of data assets that have been properly cleaned, accurately labeled, cross-modal aligned, and can be iterated continuously.

Models are becoming cheaper, GPUs will eventually increase, but truly quality data is becoming increasingly rare.

The multi-modal data industry chain is becoming a lucrative money-making track.

Data cleaning, data labeling, data synthesis, vertical data assets, robot data collection… as well as those algorithm companies specifically solving cross-modal unified representation and multi-modal pre-training encoders.

GPUs are a one-time capital expenditure, while data is a continuous capital expenditure.

Currently, I categorize the data direction into three types: The first type is Expert Data; the second type is RL Environment/Agent Data. In the future, training agents will not just involve collecting question-answer pairs, but constructing Environment → Task → Trajectory → Reward → Verifier, which could become one of the largest incremental markets.

The third type is Embodied/Robotics Data, which is even scarcer than LLM data and incredibly difficult to collect.

4. AI application dilemma, once a microcosm of the crypto world

The current AI resembles past Crypto, fat protocols, and L1 capturing most of the value, which is why Crypto VCs throw money into public chains while applications? Nobody is investing in those.

Currently, it is the same for AI.

Large models are like a POW Layer 1, where even the model equals the application.

In the past few years, most AI applications have been doing something dangerous: packaging capabilities that the model temporarily cannot do as products.

However, the boundaries of model capabilities are continuously expanding. Every time a model upgrades, it's like L1 writing functionalities that originally belonged to the application layer directly into the protocol, such as search, deep research, coding, image generation, video generation, computer use, agent...

If every time a model upgrades the value of your product decreases by one layer, then you essentially become a feature of AI, not an AI application.

Thus, you must do things that the model can accomplish but are difficult to take away, and upon reflection, it's nothing but data, context, workflow, permission, distribution.

The moat of AI applications is becoming a customer of the model, not a competitor of the model.

5. Embodied intelligence, a huge bubble

In the H4 hall on the second floor, watching numerous robots produced from similar supply chains performing similar actions slowly, then looking at the valuations of various companies, it made my scalp tingle and I felt sorry for the investors' money.

The current problem with embodied intelligence is that capital is pricing "software Scaling Law" against "soft-hard complex systems."

The miracle of large models lies in the fact that when a company trains GPT-5, theoretically it can serve hundreds of millions of people globally at nearly zero marginal cost, but robots cannot.

For every additional user a robot serves, a new machine must be built, involving BOM, manufacturing, supply chain, delivery, maintenance, depreciation.

The intelligence of AI can advance exponentially; for instance, when a model upgrades, all global users become smarter at the same time, but the cost in the physical world does not decline exponentially.

If intelligence follows Moore's Law, embodied intelligence still adheres to manufacturing industry principles; ultimately, embodied intelligence may become a vast industry but may not possess the profit margins akin to large models.

Additionally, it's too early.

Autonomous driving is already a highly constrained problem of embodied intelligence, with clear goals, limited action space, standardized road rules, and vast amounts of real data along with a mature automotive industry system. Even so, this industry has burned through hundreds of billions of dollars and still hasn't wholly resolved some long-tail issues on the road. The real difficulty faced by embodied intelligence is several times that of autonomous driving.

6. The best era for pimps

A harsh truth is that most AI startups are not making much money, not even earning as much as those who sell their skills.

Thus, I discovered something interesting; many companies appear to be one business, but upon closer conversation, "Oh, you're also selling tokens," "Oh, you're also selling computing power," and major model companies are also reselling computing power for profit.

The ones truly making money are still the pimps.

Financial advisory, brokering old stocks, flipping B300, selling tokens, selling datasets, selling computing power, even selling an opportunity to meet a certain founder...

AI meets all the conditions for a thriving intermediary market: technology changes fast enough, information disparity is large enough, capital is abundant, and stories are enticing enough.

VCs earn money by predicting the eventual outcome, while financial advisors earn money by gauging consensus without needing to prove a trend is ultimately correct.

On the contrary, the more ambiguous and grand the narrative, the more room for FA to operate.

For example, embodied intelligence, world models... these tracks share a common feature: the future narrative is large enough, the technology is complex enough, and short-term, it is difficult to falsify.

The big companies are burning money to train models, investors are betting on AGI in ten years, while intermediaries are making money.

Either siphoning off the capital expenditures of big firms

or capitalizing on the FOMO spending of LPs.

7. Money is everything

The AI race has a long way to go; people often ask, what is the most scarce resource in the AI era?

Many would respond: talent, computing power, data… but in the end, they all share one name: money.

The greatest competitive advantage in the AI industry is not technology, but the ability to raise funds.

What entrepreneurs need to do now is to keep raising funds, even if the money in their accounts is enough, they still need to raise.

The real war lies in the upcoming industry downturn cycle.

As long as you have money to stay alive, you can wait until the competition runs out of money, and then accept their talent, technology, and clients at low prices.

The secondary market is the same; within the foreseeable 2-3 years, a significant bubble collapse is likely to occur, and as long as there is enough money to pick the bottom, you can outperform 99% of people.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。