Friday, July 10, 2026

Stop watching Nvidia's market value evaporate by US$1 trillion. To put it bluntly, it's not that AI is no longer working, but that the market has shifted from "craze for selling shovels" to anxiety about "whether digging for gold can pay back the cost." Nvidia's share price fell back before the AI boom, which just proved a cruel reality: computing power is infrastructure, but with infrastructure alone, we have not seen many application-level companies that can make money on a large scale. Investors are not fools. They start asking,"How much real revenue can a model built by GPUs generate?" This is the sword of Damocles that really hangs over the heads of all AI companies. Those who are still blowing the "AI omnipotent theory" suggest looking at their own financial reports first.

But on the other hand, the actions of domestic manufacturers are quite practical. Tencent, Alibaba, and Byte are all redoing Office. This is not a simple shell. Codex's product manager said it very clearly: AI does not allow you to do faster, it allows you to do things you dared not do before. From managing scripts for the AI knowledge base, to using Codex to automate 80% of PM work, to opening the full-process agent of the office suite, this is the real implementation. Don't talk about the correct nonsense of "AGI saves the world". Being able to automate the brute link of writing a weekly newspaper is a real productivity revolution for migrant workers. However, I have to pour cold water on this: Do you expect these products to directly turn into your super assistants? I think too much. The project is full of pits, state management, data islands, and the unexplainability of Agent decisions. This is basically another version of the "shit mountain shell", but the shell looks smarter.

Finally, a more "nonsense" but alarming signal-the EU has passed Chat Control 1.0. On the surface, this thing is said to be for child protection, but in essence it is large-scale chat content monitoring. For developers, this means that any AI application you make with messaging functions will require an additional level of compliance review if it wants to be launched in Europe in the future. Don't think this is a distant political game. It is directly related to whether your Agent can freely read context in user chats. The risk of privacy disclosure will increase sharply, or the death penalty will be imposed on the entire application model. No matter how good your skills are, you cannot be blocked by a decree.

Summarize today in one sentence: The decline of Nvidia is a rational correction of the computing power bubble. Redoing Office by domestic manufacturers is a signal that AI is really starting to work. However, the Chat Control thing reminds us that no matter how good the technology is, it still depends on the regulatory look. As for those Nilai University No. 5 buildings that "make 10,000 yuan a month" for posting bills and rent 270,000 yuan in electricity, just listen and watch as lively, don't really believe it.

Nvidia's market value has evaporated by US$1 trillion in two months, which may be the most dramatic signal today. But don't be intimidated by this number, put it in a larger picture: On the same day, OpenAI released GPT-5.6, SpaceXAI launched Grok 4.5, Mistral came up with the robot navigation model Robostral Navigation, and Baidu, Tencent, and Google each released OCR, text generation and table prediction models on Hugging Face. The computing power layer of AI is undergoing a revaluation of value, but the application layer and model layer have entered an unprecedented period of intensive innovation. Behind this is the market's logical switch from "selling shovels" to "digging for gold"-investors are no longer willing to pay a premium for the scarcity of GPUs, but are beginning to ask what these computing power can produce. Nvidia's fall is not because AI is ebbing, but because AI has entered the "practical" stage, and everyone has begun to pay attention to cabbage and pork, not the kitchen knife itself.

Another signal worthy of attention is the release of the Xiaomi Pengcheng series, as well as the pricing of the NIO ES8 Big Five and the Tengshi Z Super 1.3 million. Lei Jun proposed that "the interior space should not be fixed", which is ostensibly a product definition, but is actually a paradigm shift taking place in the entire automotive industry. In the past five years, electrification has solved power problems and intelligence has solved interaction problems. Now, spatial variability is becoming the next battlefield. When a car changes from a "driving tool" to a complex of "living space","working space", and "entertainment space", seat layout, modular interiors, and even movable seat tracks will become core selling points. At the same time, Volkswagen announced that it would lay off 120,000 jobs and streamline 50% of its models. This is not only a pain for traditional car companies, but also a redefinition of the "essence of cars": if space can be customized, what brand needs so many fixed forms? Model? Volkswagen's contraction and Xiaomi's expansion just heralded the transition from a "model matrix" to a "space platform".

Connect these signals with the trends of AI office and you will see a bigger story. Tencent, Alibaba, and Byte are redoing Office at the same time, and the AI knowledge base has evolved from "storing articles" to "operating cognitive assets." OpenAI Codex product manager said that AI does not allow you to do faster, but allows you to do something you dared not do before. These are all pointing in the same direction: AI is reshaping the relationship between "work" and "space". When cars can become mobile offices, when AI Agents can automatically complete the closed loop from requirements to code, and when the knowledge base can be dynamically adjusted based on your status, people no longer need fixed workstations, fixed software, or even fixed travel methods. Goldman Sachs prohibits employees from participating in predictive market transactions, essentially guarding against the impact of "everything can be calculated"-AI is blurring the boundaries between prediction and decision-making.

The risk today is that infrastructure investors may be overly pessimistic, and application-level entrepreneurs may underestimate the complexity of "spatial customization". But the opportunities are clearer: Companies that can access physical spaces (cars), digital spaces (offices), and agents (AI Agents) will define the user experience for the next decade. The evaporation of Nvidia's market value is only a temporary cleanup. The layoffs in the automotive industry are the pain of transformation, and the real winners are quietly laying out amid these seemingly contradictory signals.

Nvidia's market value evaporated by US$1 trillion in two months. This figure itself deserves serious consideration. It's not that AI is no longer working, but that the market is finally beginning to doubt the narrative of "unlimited demand for computing power." In 2026, GPT-5.6 and Grok 4.5 will be released one after another, each claiming to be "Opus", but the trend of chip stocks illustrates one problem: the speed of technology iteration is exceeding the speed of cashing commercial returns. Goldman Sachs 'ban on employees from participating in predictive market transactions at this time is even more significant-even Wall Street itself is restricting internal speculative bets on financial markets, which is a vigilance against systemic risks. Historically, before the Internet bubble burst in 2000, the stock prices of Cisco and Lucent were firm on the logic of "everyone is using our devices" for a long time, until they found that the growth of users 'bandwidth demand did not catch up with equipment shipments. In today's AI industry, are GPU shipments already ahead of actual application scenarios? No one wants to think about this question, but Nvidia's share price answers it for us.

Looking at the AI office tools line, the product manager of OpenAI Codex said,"AI allows you to do things you dare not do before." Tencent, Ali, and Byte are redoing Office and stuffing AI Agents into the workflow. Sounds great, but take a closer look at the core logic of these products: They are essentially replacing routine operations with automation, rather than creating new productivity gains. Codex automatically handles Slack feedback and AI writes documents, functions that five years ago would have been called upgraded versions of "macros and scripts." Back then, Microsoft's Office assistant Clippy was ridiculed as useless. Today, if these AI Agents are not careful, they will end up in the same situation-users will feel new when using it once, but they still have to manually adjust the parameters when using it three times. What is even more alarming is that these products are emphasizing a "complete closed loop from coding to release," but the real software engineering problem has never been "writing code too slowly", but "wrong requirements" and "system complexity out of control". AI can help you write a thousand lines of code, but it cannot help you decide whether these thousand lines should be written. This cognitive gap cannot be filled by just a few Agents.

In the automotive industry, the Xiaomi Pengcheng series focuses on "spatial customization", while NIO removes the third row of the ES8 and makes it a big five. Behind these actions is a common dilemma: the space for hardware differentiation for new energy vehicles is getting smaller and smaller, and everyone is making a fuss about software and space layout. Lei Jun said that "refrigerators, color TVs and large sofas are common configurations." This is true, but the problem is that when all car companies can provide these functions, this is no longer a selling point, but a standard feature. NIO's removal of the third row seems to be accurate positioning, but in fact it is admitting that "I can't build a pure electric SUV that satisfies everyone"-no car company has been able to truly solve the comfort problem of the third row of seats. Volkswagen has simply streamlined its model lineup by 50%, and German workers are preparing to protest against the 120,000 job cuts. This is the real response of traditional car companies to the electrification transformation: it is not that they don't want to change, but that if they change, people will die. The reason why "new forces" such as Xiaomi and Nilai can remain calm is largely because they have no historical baggage, but it also means that they lack awe of the complexity of manufacturing. The battlefield of electric vehicles has never been a pile of functions, but boring but fatal things such as supply chain management, manufacturing yield, and after-sales network. Whether today's "space customization" gimmicks can be transformed into maintenance interchangeability three years later is the real question.