Wednesday, July 8, 2026

The signals are mixed today, but I only see two core stories: the implementation of AI is more painful than imagined, and the "robot face" is being forced into the capital market.

Let's talk about AI first. Don't be fooled by those open source models and SOTA news. ACE-Brain-0.5 is good to be open source, but if you look at what the supporting StableVLA article says-the robot will fail if it wears oily glasses, and it has to be equipped with an IB-Adapter to reduce noise. What does this mean? The biggest pit of embodied intelligence now is not the model parameters at all, but the code of the shit mountain of the physical world. Oil pollution, light, and obscuration, things that do not exist in academic datasets, are all exposed as soon as they go to the production line. Ordinary developers? Don't dream, if you go to develop embodied intelligence now, you will be the first to stare at sensor noise for about three months. But in turn, StableVLA used 500 million parameters to do 7 billion more work, which is the true signal: engineering optimization is far more cost-effective than heap parameters.

Looking at the developer discussion that day, someone used Claude Code to burn 600 yuan a day. This is no longer just a complaint. This is the cost structure sounding the alarm. AI coding tools do improve efficiency, but the ceiling for pricing by token is there. Once your codebase is big enough to require frequent context switching, you will scold DeepSeek for irregular interviews while looking at Claude's bill and crying. LangChain and his gang were smart and launched OpenWiki to help write documents-this is a typical information flow business, which is much more stable than directly selling the underlying model. Remember, the most profitable thing now is not AI itself, but the tools and processes that help people manage AI.

Finally, talk about the hurried face on the capital face. xAI changed its name to SpaceXAI, and Ruiwei Technology fell 17% in the dark market-these two things are particularly interesting to see together. Musk stuffed AI directly into SpaceX because the capital market only recognizes the big pie of "aerospace +AI". It is too difficult for independent AI companies to raise funds now and have to find a hard technology shell to cover them. Where's Ruiwei? There were no cornerstones, no green shoes, and they were put on without even installing them. As a result, the amplitude was 41% on the opening day, and the leeks were directly swallowed by the fluctuations. This tells us that the first strand of visual embodied intelligence? Good story, but if the financial model doesn't work, it's wealth on paper.

Don't believe the headline party of "Unified Base Model" and "Million Robot Sales". The real world has never been end-to-end. You haven't even built a robot that can screw and is not afraid of oil pollution. What can you talk about replacing pesticides? At this stage, the survival logic of AI companies is very simple: either cut hard needs that others cannot chew, or make screws in the ecology. Faults are everywhere, but for those who understand the engineering pits, opportunities are also there.

Today's signal allows me to see two accelerating trends. They may seem parallel, but in fact they are mutual cause and effect.

The first trend is the large-scale penetration of AI from the digital world to the physical world, and this time it is not a laboratory demonstration, but a true industrial integration. xAI officially changed its name to SpaceXAI, which is not just a brand name change. SpaceX has invested $12.7 billion in AI, more than three times its spending on aerospace and connectivity businesses, and AI has become the company's core strategy. This means that AI is no longer just software-level intelligence, it is deeply integrated with the most hardcore physical infrastructure-the manufacturing and operations of rockets, satellites, and nuclear submarines. At almost the same time, Daxiao Robot open-source the unified body-based model ACE-Brain-0.5, aiming directly at Physical Aseptic AI; Gefei released air and ground agricultural robots, completely handing over pesticides to the machine; There are also color-changing sensors that allow the robot to directly read tactile information. These signals point in the same direction: AI is gaining full physical perception and execution capabilities. However, the capital market gave a split response-the first stock of visual embodied intelligence technology broke 17% in the dark market. This reminds us that there is a time lag between technological maturity and commercial implementation. Only physical AI products that can be truly scaled and have a clear commercial closed loop will be recognized in the secondary market. Half a year later, we will see a large number of startups sprinting on this track, but only those teams that are bound to specific industry scenarios (such as agriculture, logistics, aerospace maintenance) will survive.

The second trend is the systematic collapse and reconstruction of the information trust system. The China Securities Regulatory Commission opened an investigation into individuals who fabricated false information on public education, but the reason was actually a vulgar bet-indicating that in the AI era, creating a piece of information that can affect billions of market value has become extremely cheap and fast, and traditional regulatory methods can only be held accountable after the fact. TikTok's global layoffs focus on trust and security teams, while the EU Chat Control bill has just passed its first round, which provides a comprehensive authorization for monitoring communications content. Platforms are weakening self-purification capabilities, while governments are strengthening surveillance, creating a dangerous tension: the weapons for spreading false information are empowered by AI, while defense mechanisms are retiring. On the other hand, the developer community is actively saving itself-Strix has open-source AI hacking tools to proactively discover application vulnerabilities, OpenWiki allows AI to automatically maintain code documents, and Facebook has open-source Astrox, a design system for agents. These open source projects herald the emergence of a decentralized trust infrastructure that does not rely on platforms and government rulings, but uses technology to achieve verifiable transparency. This is a larger battlefield than the capital market. In the coming year, the game around "who defines the truth" will determine the underlying rules of the next Internet.

These two trends actually share the same core: power is shifting from humans to systems. AI materialization allows the rules of the physical world to be reshaped by algorithms, and AI information warfare allows social consensus to be reshaped by computing power. We are standing at a decision-making crossroads, and every day's technological signals are laying bricks for this new road.

In today's signal, I most want to talk about two things: AI's hot money and bubble in the capital market, and the foundation of digital trust that is being eaten away. These two lines may seem independent, but they mirror each other in this news list.

Let's first look at the line of AI commercialization. Ruiwei Technology, as the "first stock of visually embodied intelligence", plunged 17% in the dark market, with an amplitude of as high as 41%. There were no cornerstone investors and no green shoes, which was basically equivalent to nude swimming. At the same time, SK Hynix was going to list on NASDAQ and raise US$29 billion, claiming to be the largest foreign investment in history. Anthropic rented a building in New York and expanded its staff to 1000 people. xAI has just changed its name to SpaceXAI. IPO documents show that AI invested US$12.7 billion last year-three times that of aerospace. On the one hand, capital is frantically piling up valuations, and on the other hand, benchmark companies are going bankrupt when they go public. It's a lot like Internet stocks in 2000 and ICO in 2017: all stories are perfect until you see the actual revenue. ACE-Brain-0.5 reached the top of SOTA. StableVLA claims to crack visual interference, but which product has become a product with sustainable revenue? Ruiwei's break tells us that the market has begun to vote with real money and is no longer moved by the title of "Unified Personalized Model". History has repeatedly proved that technological leadership does not equal commercial success, and when "SOTA" and "break" appear at the same time on a track, the cracks in the bubble are clearly visible.

Another more subtle but more deadly signal is a crisis of trust. The China Securities Regulatory Commission opened an investigation into An's fabrication of false information on public education. A vulgar bet could make stocks go up and down. This shows that the information environment has reached a terrible level. On the same day, TikTok laid off employees and cut its trust and security teams around the world-guess who will intercept the next time fake news spreads? Xiaohongshu's back-end protocol controversy has been called by developers to "destroy the entire VIE industry", implying that the platform can change the rules at any time. Taken together, these events outline a digital society where the infrastructure is being reshaped at will: there are no firewalls for the authenticity of information, review teams on social platforms are laying off workers, and rumors in the capital market can be easily priced. Contact the complaints from the developer community-some people use Claude Code to burn 600 yuan a day, some people complain that DeepSeek interviews are not standardized-and you will find that the entire chain is overdrawing trust. There have been similar scenes in history: spam flooding caused the credit collapse of the mail system in the 1990s, and the collapse of trust in financial derivatives triggered a crisis in 2008. Today, from rumors of individual stocks to layoffs on platforms, loopholes in trust are expanding, and the speed of repairs is far behind.

The intersection of the two themes is that AI companies are getting bigger and bigger, while society's trust in information and systems is getting thinner and thinner. When Ruiwei's share price and a fake news piece appear on the screen at the same time, you will understand that what makes money in this ecosystem may not be technology, but people who manipulate expectations. Those who pay the bill are those institutions that take over the break and individual investors who chase after the gains amid rumors.