China's Success Formula: No Chips In, No Experts Out

Trump's Export Controls Fuel China's CCP and AI

The export controls were intended to slow down China's AI development through the Nvidia boycott. Instead, Washington has given the Chinese chip industry a billion-dollar gift. Huawei is now supplying the Ascend chip, DeepSeek is building the V4-Pro model on it, and prices have dropped by 75 percent. The sanctions haven't slowed down China's AI sector. They've subsidized it. DeepSeek is now selling frontier-level AI at a fraction of US prices—on the very chip Washington wanted to ban. While Anthropic and OpenAI are betting that developers will pay premium prices for the best performance, DeepSeek is targeting the majority: good quality at a bargain price. → AI Secret

Synthszr Take: With its chip sanctions, Washington has achieved the exact opposite: a state-sponsored Chinese AI industry that is now pulverizing Western pricing models. DeepSeek demonstrates what happens when you treat AI as infrastructure instead of a premium product. The 75 percent price reduction isn't dumping—it's the natural marginal cost logic of digital goods when the state covers the initial investments. Anthropic and OpenAI bet on quality differentiation (classic Silicon Valley thinking). DeepSeek is betting on volume and integration. This is fatally reminiscent of the solar industry: Germany developed, China scaled, the price collapsed. The West must learn: with exponential technologies, it's not premium that wins, but speed plus government backing.

The Great Firewall of Talent: Beijing Locks Down Its AI Researchers

China is tightening exit controls for top AI talent at private companies like Alibaba and DeepSeek. Employees on strategically important AI projects now require government approval before they can leave the country. Bloomberg reports this measure, citing insiders, which was already issued as a 'recommendation' to AI executives in March 2025—at that time, focusing on travel to the US. The fears: data leaks, technology theft, talent poaching. In parallel, China's AI industry is systematically sealing itself off. Domestic chip manufacturers now control 41 percent of the Chinese AI accelerator market (according to IDC). Beijing also blocked Meta's planned acquisition of the agent startup Manus AI. The technological system competition between China and the US continues to intensify. → Techpresso

Synthszr Take: China is turning its AI researchers into state property—at least concerning their freedom to travel. This is the logical consequence of viewing technology as a geopolitical resource: the most valuable assets are not the chips or models, but the minds that develop them. What is standard for state defense researchers is now being extended to private tech companies. The 41 percent market share for domestic AI chips shows that the isolation is working in the short term. In the long term, it could become China's Achilles' heel. AI innovation thrives on global exchange, on conferences, collaborations, the informal chat after a keynote. Locking up your talent might gain you control—but you lose touch with the global cutting edge.

Tencent's Flood of Agents Tests Its AI Catch-Up Strategy

Between March and May 2026, Tencent flooded the market with AI agents: WorkBuddy for office work, CodeBuddy for developers, QClaw for controlling desktops, DataBuddy for data analysis, and Ardot for UI/UX design. At the end of May, Marvis followed, an operating system assistant with six specialized agents. Chairman Pony Ma personally endorsed over a dozen variants on WeChat Moments. Tang Daosheng, CEO of the cloud division, announced the paradigm shift from chatbots to agents. In parallel, Tencent equipped Yuanbao, QQ Browser, Sogou Input, and Tencent Meeting with agent functions and is preparing to convert WeChat Mini Programs into directly callable AI skills.

The numbers behind this are less impressive: Yuanbao's monthly active users are at 57.35 million, ByteDance's Doubao has 345 million, and Alibaba's Qwen has 166 million. In the Chinese cloud IaaS market, Tencent Cloud ranks 5th with an 8% market share. In the Model-as-a-Service segment, where the token-based business is growing fastest, ByteDance's Volcano Engine, Alibaba Cloud, and Baidu AI Cloud together hold almost 90%—Tencent doesn't even appear in the rankings. The new AI products are weighing on the operating result with an estimated 8.8 billion yuan ($1.2 billion) per quarter. → Hello China Tech

Synthszr Take: Tencent is throwing two dozen specialized agents onto the market, hoping that quantity will turn into quality. The strategy follows a familiar logic: if you're lagging in foundation models, you have to score points in the application layer. The vertical integration via WeChat Mini Programs as callable skills shows where Tencent's real strength lies: in the ecosystem used daily by 1.3 billion users. Can WorkBuddy and CodeBuddy compete with Doubao? Questionable. But if Marvis is seamlessly integrated with WeChat Pay, Tencent Meeting, and QQ Browser, model quality suddenly takes a back seat. The 8.8 billion yuan quarterly loss is the price of a late entry—or the down payment on a position Tencent can defend with its ecosystem.

Doubao Cashes In: Why Bytedance Now Charges 5088 Yuan

Bytedance's AI assistant Doubao is launching a pricing experiment that is making the Chinese tech scene sit up and take notice. While the free basic version is being retained, the company is introducing three paid tiers: Standard (68 yuan/month, approx. €8.70), Plus (200 yuan/month, approx. €25.60), and Professional (500 yuan/month or approx. €64, or 5088 yuan/year or approx. €650). With 345 million monthly active users, Doubao is already by far the largest AI assistant in China—Alibaba's Qianwen has 166 million, Baidu's Yuanbao 57 million. The pricing clearly targets power users: PPT generation, data analysis, and video production are intended to justify the high inference costs. Bytedance is also testing integration with its in-house e-commerce ecosystem Douyin—users can buy products directly from the chat. → Hello China Tech

Synthszr Take: 345 million users with 120 trillion daily tokens—that's a cost structure that not even Bytedance can subsidize forever. The price points are calculated: 68 yuan (approx. €8.70) scratches the psychological threshold (surveys show 48 yuan is acceptable), while 5088 yuan/year (approx. €650) is almost triple the price of Western competitors. But the real game is being played elsewhere: Doubao is set to become China's super-app for the AI era, with direct shopping integration into the 4.4 trillion yuan Douyin ecosystem. Users complain about 'three different answers to the same question'—but with this head start, Bytedance can afford a few experiments. If you lose 15% of your users, you still have more than all your competitors combined.

Tencent Brings AI Assistant to the Operating System Level

Tencent is integrating its AI assistant Marvis directly into the operating system—a move that could show where the journey is headed for all major tech platforms. The Chinese tech giant is getting serious about the vision of a ubiquitous AI layer that no longer runs as a separate app but functions as a fundamental part of the system. Marvis is intended to automate tasks system-wide, provide context-aware suggestions, and act as a universal translator between apps. The integration is happening on both desktop and mobile levels. Tencent is thus positioning itself directly against Microsoft's Copilot integration and Apple's Intelligence-features. → Hello China Tech

Synthszr Take: OS-level integration is the logical next step after the app chaos of the last two years. Tencent is demonstrating what Microsoft with Copilot and Apple with Intelligence are aiming for: AI disappears as a visible category and becomes infrastructure. This is reminiscent of the evolution of TCP/IP—nobody asks about it anymore, it just works. The interesting question is governance: who controls an assistant that sits deeper in the system than any app? China's approach will likely differ from the West's. For enterprise customers, this means the discussion shifts from 'Which AI assistant do we use?' to 'Which operating system do we trust with our processes?'. That is a more fundamental decision with significantly higher switching costs.

Huawei Announces Chip Breakthrough: Gap with TSMC Is Shrinking

Huawei aims to produce 1.4-nanometer chips without ASML machines by 2031—using its own 'LogicFolding' technology based on the in-house 'Tau Scaling Law'. The gap with TSMC would shrink from five years to three. Semiconductor chief He Tingbo promises 'not a continuation, but a great leap' for the Kirin-chips in the fall. The Chinese Star-50 index hit a record high after the announcement, and SMIC shares rose by 18 percent. Huawei is experimenting with Self-Aligned Quadruple Patterning (SAQP) to increase transistor density without EUV-lithography. The company has developed 381 chips over the past six years based on its Tau principle, which optimizes data transmission instead of transistor shrinking. → Techpresso

Synthszr Take: Huawei isn't rewriting physics; it's circumventing it. Instead of shrinking transistors (Moore's Law), you accelerate their data transfer (Tau Scaling). This is reminiscent of M-Pesa in Kenya: if you don't have banking infrastructure, you invent mobile banking. The 381 chips in six years show engineering velocity under the pressure of sanctions. Whether SAQP-lithography at 1.4 nm actually enables mass production remains questionable—the yield could be prohibitive. The real test comes in the fall with the Kirin-chips. If Huawei pulls this off, we'll have two parallel semiconductor universes: one with EUV, one without.

Memory Chips Become a Trillion-Dollar Business

Memory chip manufacturers Micron and SK Hynix today surpassed a market capitalization of over one trillion dollars for the first time. Micron shares jumped 19 percent after UBS tripled its price target from 535 to 1,525 dollars per share. SK Hynix gained 11 percent. Since the beginning of the year, both stocks have more than tripled: Micron by 214 percent, SK Hynix by an even more impressive 250 percent. The reason: AI companies are buying up every available High-Bandwidth-Memory-chip on the market, and production can't keep up. Samsung had already cracked the trillion-dollar mark a few weeks ago. → siliconangle.com

Synthszr Take: A trillion dollars in market cap for memory chip manufacturers—this is the most brutal redistribution of value we've ever seen in tech history. Micron was worth 60 billion just two years ago. The entire AI revolution is hooked on these memory chips like a junkie on a needle: no large language models without HBM, no inference without DRAM. The pricing power of these three players (Micron, SK Hynix, Samsung) is now greater than that of OPEC in the '70s. Anyone who thinks this is just a temporary shortage underestimates the physical limits of chip production. These companies are currently printing money with memory chips, while Intel is kicking itself for backing the wrong horse.

Cursor Is Having Its Moment

Cursor reaches an annual revenue of 3 billion dollars with over 3,000 enterprise customers. The AI-powered development environment is achieving what takes others years: making the leap from a developer toy to enterprise software. In parallel, Anthropic's Project Glasswing is scanning over 10,000 critical security vulnerabilities in open-source projects with Claude Mythos Preview. Meanwhile, Google is making the biggest deals with its smallest model, Gemini 3.5 Flash—one billion monthly users in AI search mode. DeepSeek is cutting prices for its V4-Pro-model by 75 percent to as low as 0.83 dollars per million tokens. Box-CEO Aaron Levie diagnoses 'AI psychosis' in executives: they see the demos and overlook the tedious integration work. → The Code

Synthszr Take: Cursor demonstrates what happens when AI tools break through the productivity threshold: 3 billion dollars in revenue is proof that coding agents have reached the mass market. Google is playing a different game than OpenAI or Anthropic—while they compete for benchmark victories, Google prefers to scale the second-best model to a billion users. That's platform thinking versus model fetishism. DeepSeek's price drop shows that the commoditization of inference has begun, just as Ryan Dahl predicted. What Levie calls 'AI psychosis' is a precise observation: executives are mistaking demo capability for production readiness. Real progress lies in integration, not in the next benchmark record.

Europe's Tech Awakening: AI is Shifting the Centers of Gravity

Swedish legal-AI Legora cracks 100 million dollars in annual revenue, Lovable jumps 33 percent in one month and is looking for acquisition candidates, Klarna is preparing for its IPO: Europe's tech scene is currently having its iPhone moment. The numbers are impressive: 20 percent of the top 100 law firms in the US are already using Legora's software instead of its Silicon Valley competitor, Harvey. Lovable-CEO Anton Osika speaks of a 'structural shift'—Europe has long produced top technical talent but has traditionally struggled with global scaling. George Robson of Sequoia confirms: this is not a perceptual distortion but a real change that has been building for years. The flywheel effect of established European tech companies like Spotify is now creating a self-reinforcing ecosystem. → Business Insider

Synthszr Take: Europe is currently doing what Asia did with semiconductor production: jumping from the second row to the top. The AI wave is hitting a region with excellent technical universities, sensible data protection regulations, and—this is the key—a certain distance from Silicon Valley groupthink. While US startups are still building their eighth chatbot variant, Europeans are solving concrete industrial problems. Legora is digitizing law firms, ASML builds the only machines for modern chips, and SAP keeps the global economy running. The scaling problem remains real (Douglas Brion of Matta cites a lack of scale-up capital as the main obstacle), but the success stories show that the continent has learned how to turn hidden champions into global players. The question is no longer whether Europe can keep up. The question is in which areas it is leading.