China and Nvidia star in the “great technological divorce” of 2026. A bureaucratic hell that is erasing it from the market

Talking about Nvidia is talking about artificial intelligence glue. The GPU giant has invested millions financing cocompanies like OpenAI or Anthropicbut along the way has not forgotten startups or to make purchases for strengthen your position in the market. The problem is that it is missing out on a potential $50 billion market: China. Because Nvidia is eager to enter China, but it is trapped between bureaucracy, the Trump Government, Xi Jinping’s Government, and the smuggling of its graphics cards. The great divorce. In a very short time, Nvidia has gone from dominating the Chinese GPU market for artificial intelligence to losing it completely. The restrictions of the Trump Administration and the intensification of the trade war between the powers left Nvidia out of the game. Either it would adapt its GPUs and create less capable versions of those it sold in the West or it would not be able to sell in China. For a time, Nvidia was selling the H20 to adapt to the new rules, but it is something that has taken its toll. As AI needs demanded more powerful GPUs and own chinese industry with Huawei, Cambricon and Moore Threads was developed, Nvidia was being left out of the game. Official quota. In the middle of last year, Nvidia CEO Jensen Huang pressed Donald Trump to see reason: it was better for Nvidia to be able to enter China both to make money and to slow the accelerated development of the domestic industry, one that Western restrictions had given wings to. In the end, the US gave in previous tariffs of 25% and one condition: all GPU orders from Chinese companies to Nvidia would be reviewed one by one. There is a problem: the US body in charge of reviewing these export licenses has decreased by 20% in recent months, which is causing delays of months when it comes to fulfilling an order. From when a Chinese company asks for Nvidia GPUs until they are given an answer, the ‘chinese dragons‘They have already released some product. The result? Huang points out that Nvidia has gone from being a leader in China to have a 0% quotapainting the situation as a true drama and pointing directly to the strategies of both China and, above all, the United States as the cause of his company falling into the offside of the large Asian market. Furthermore, it is China itself that encourages its companies to, to the extent possible, use Chinese hardware that they is developing at accelerated rates. “Official” fee. But the fact that Huang claims that his market share in China is 0% does not mean that there are no GPUs for AI in China because it seems that there are H100, H200 and even B200 due to something very simple: smuggling. Despite the proprietary technological solutions they are developing, it is evident that a large part of the AI ​​industry is built with Nvidia GPUs and that implies that the tools are very well optimized for them. There are several occasions in which Nvidia AI chip smuggling networks have been reported, with modest seizures on occasions (just tens of millions of dollars) and somewhat larger seizures on others (hundreds of millions in a few months). Chinese companies obtain these chips through indirect routes from Hong Kong and Singapore and, although Nvidia tries to trace the origin, the clandestine flow and opaque chains make the task complex. trapped. Someone is lining their pockets and that someone is not Nvidia. And the problem is that Huang’s pressure had an effect, but the solution they gave him is not as agile as the market needs. Returning to the issue of bureaucracythe United States Office of Industry and Security, which is responsible for reviewing these export licenses, reduced its workforce by 19% in 2024. Specifically, those who develop standards linked to the semiconductor industry and review licenses have decreased by 20%. The result is an average of 76 days to resolve export requests, something that is extending so far this year and which is disastrous news for both Nvidia and others deeply involved in the AI ​​segment, such as AMD. From China, things are not much better, since companies must make it very clear why they need Nvidia AI chips and cannot meet their objectives using national alternatives. Jensen, almost excluded. In any case, it is evident that Huang does not like to be missing the AI ​​party in China, in the same way that he is going to miss the new trip of Donald Trump and other executives to a summit between Trump and Xi Jingping that will be held between the 13th and 15th of this month. Or so it seemed. This is an event in which conversations will focus on agriculture and commercial aviation, so a priori Jensen didn’t have much in mind. But of course, alongside Trump are CEOs like Elon Musk, Cristiano Amon or Tim Cook, among others. And, although it seemed that he was not invited, as we see in South China Morning PostIn a message from Trump on his social network, it was confirmed that Huang will finally accompany him on the trip. In the end, it’s about money. Jensen Huang doesn’t want China to have the best chips because He wants to save those for the United States.but it is a very large market in which Nvidia can offer chips strategically: it makes money while making companies opt for its product instead of that of the Chinese companies themselves. In Xataka | Nvidia’s superpower is not having money, it is making everyone work for it: Foxconn is the latest to join

The AI erasing the entire code of your app

He ‘Vibe Coding‘is all A revolution. One that has even made veteran programmers who “in a year or two code editors will not exist.” However, it is also causing Very large damage in the code. Jason Lemkin, former Vice President in Adobe and the founder of the community Saastr, discovered it for the badaccording to X. Colega, where is my code? It was the eighth day of a series of ‘Vibe Coding’ by Lemkin using the online programming platform Replit (Among the favorites of developers) When he realized a huge disaster. Artificial intelligence had deleted the entire database of the application of commercial contacts that it was developing. The wizard reported what happened Without giving many details: The system worked the last time you started session, but now the database appears empty. This suggests that something happened between that moment and now that it deleted the data The AI did not follow the rules. The assistant even acknowledged that he had violated a Replit directive that tells the assistant not to make more changes without permission and always show all the proposed changes before implementing them. In the process, he deleted contacts from 1,206 executives and 1,196 companies. In the end, the assistant admitted to having made a “catastrophic error of trial”, but gave no option to go back through a backup. Replit did not differentiate that the changes made on the database were made in a production environment (with sensitive data) and not only of development, so it affected Lemkin’s real product. Except for the fact that Lemkin explained after the application was not active. “If it had been 2 to 4 weeks later, it could have been much worse,” he said. The application was almost ready, but not finished. Replit, the company, to the rescue. After becoming known in networks, Amjad Masad, CEO of Replit, confirmed the case and admitted that it was “unacceptable” and that something like that “should never be possible.” And they went to action: they introduced changes so that the development and production database are completely separatedso that the changes were made only in test environments and the Replicit agent cannot access the data. Faced with what the assistant told Lemkin, Masad said they did have backups that allow everything to be restored in a click. The problem is that the agent could not execute it, even if the platform had the option. Replit ended up refunding Lemkin his subscription for the discomfort caused. An example of a whole paradigm. The great person in charge of the case is the development platform, in the sense that it did not comply with the restrictions imposed by its developers. But the case also shows that Attendees are not magicas Massad remembered, and that they have to be used with background and caution if there are sensitive data at stake. Presume to use AI to program without having a programming idea It sounds good, but you have these problems. Software engineers will have a lot of work. Yeah, fixing pifias. “When the code can occur at lightning speed, prevention and intuitive judgment become especially important”: thus Pascal Biese summarized it (Developer and founder of ‘LLM Watch’). Also There are those who joke: “Vibe Coding: an industry of billions of dollars. Fix/maintain written code with AI: Industry of one billion dollars.” In words of Gergely OroszEngineer and influential voice in the technological sector for its Newsletter The Pragmatic Engineer, “the most important role of a software engineer in the era of coding tools with AI is to become a large code reviewer.” Instagram co -founder Mike Kriegerit was even more forceful When he said that, from here to three years, software engineers are no longer writing any code: they will only check the code created by AI. Image | Mohammad Rahmani In Xataka | Openai is raining blows on all sides: the last one signs Google “raising” his commitment to Vibe Coding

Log In

Forgot password?

Forgot password?

Enter your account data and we will send you a link to reset your password.

Your password reset link appears to be invalid or expired.

Log in

Privacy Policy

Add to Collection

No Collections

Here you'll find all collections you've created before.