Trump is set to announce the "Hormuz Alliance" this week, while also considering whether to send ground forces to seize Kharg Island. Brent crude oil doesn't believe him, surging over $3 per barrel on Sunday, breaking through $105. On another front today, Alibaba has consolidated all its AI assets into one business group, directly managed by the CEO, while China's second chip factory prepares for 7-nanometer production on the same day.
1| Hormuz Alliance + Kharg Island Option: Trump Falls into His Own Trap for the First Time
Four sources told Axios that Trump is contacting multiple countries to form the "Hormuz Alliance," with an announcement expected this week, aiming to forcibly reopen the Strait. Meanwhile, the White House is evaluating another option: deploying U.S. ground forces to directly occupy Kharg Island, through which 90% of Iran's oil exports transit. The market isn't buying it, with Brent crude jumping over $3 per barrel on Sunday, already surpassing $105. The WSJ also confirmed that the alliance announcement itself failed to suppress oil prices.
The fundamental difference between the two options is that the alliance is a diplomatic display and can still be reversed, while deploying ground forces to the island is a substantial commitment with no "one-click withdrawal." This is the first time in the third week of the Iran war that Trump has been forced into an extreme position. His usual style is intuition and improvisation, playing and retracting cards equally casually. But Axios' behind-the-scenes analysis says the war is the first situation where he cannot unilaterally control the outcome; Iran's drones don't care if he tweets in the afternoon, and the Strait doesn't automatically reopen because he held a press conference. The same day, he accused Iran of using AI to spread disinformation, officially bringing AI into the information warfare framework of the Iran war.
The morning report discussed the cabinet's撕裂 (split), with Sacks wanting to withdraw and Hegseth shouting "no survivors." The evening signal is that the cabinet isn't unified yet, but Trump is simultaneously pushing for more aggressive military options externally. It's not a consistent strategy; it's two opposite directions moving at the same time.
2| Alibaba ATH + China's Second 7nm Factory: The "Waist" of the Sanction System is Widening
Alibaba today established the Alibaba Token Hub (ATH) business group, directly managed by CEO Wu Yongming, consolidating Tongyi Lab, the MaaS business line, Qianwen, Wukong, and the AI Innovation Department—packaging everything from basic model R&D to consumer applications into one chain. U.S. pre-market stock rose nearly 3%. This is the first clear action among Chinese internet giants to organizationally structure the entire AI chain—not "setting up an AI department" but "restructuring into an AI company."
On the same day, Reuters exclusively reported that China's second-largest chip manufacturer is preparing for 7nm mass production on a newly acquired plot in Taiwan. The previous market consensus was that only SMIC could handle advanced processes; now there are at least two. Reuters Breakingviews published a concurrent article stating that the AI influence of Chinese tech veterans like Baidu and 360 is being eroded by new stars like DeepSeek and Kimi.
The logic of U.S. sanctions is to choke the head—restricting TSMC supply and blocking GPU exports. But China's response is taking a different path: at the application layer, integrating AI into the largest commercial entity; at the manufacturing layer, getting a second manufacturer to run 7nm. Each additional node reduces the precision strike effectiveness of the sanctions. This isn't bypassing sanctions; it's turning a single point into a surface.
3| On the Eve of GTC, Thompson Says It's Not a Bubble, Foxconn Uses Profits to Say "But Demand Isn't There Yet"
Ben Thompson today published the most timely article, with the title "Agents Over Bubbles" itself being a thesis. AI is neither a bubble about to burst nor just hype; the business model of the Agent era is the real issue. Nvidia GTC opens tomorrow, with The Information reporting that Groq chips will be unveiled.
But the same day, Bloomberg's headline was that Foxconn's profits missed expectations, raising AI demand concerns. Hon Hai is one of Nvidia's largest hardware OEM partners, and the profit miss made the market start to wonder if AI hardware shipments have already hit a ceiling. Meanwhile, Meta announced an investment of up to $27 billion in AI infrastructure with Nebius, following the morning report's "20% layoffs"—Meta continues to double down on computing power, cutting jobs and spending money simultaneously.
Three signals stacked together: Thompson is saying you're asking the wrong question, Foxconn is saying there's still a gap on the demand side, and Meta is saying I don't care, I'll keep building. GTC is the moment the test papers are handed out; Nvidia needs to clarify how computing power is monetized in the Agent era.
4| Two Gray Markets Diverge: Prediction Markets Whitening, AI Deepfake Fraud Scaling
Wired reports that while the legal battle over prediction markets isn't over, Wall Street financial institutions have already started placing bets. The legal gray area hasn't stopped capital, almost the same script as the prelude to cryptocurrency.
Simultaneously, another Wired report states that dozens of Telegram channels are publicly recruiting "AI facial models," mainly recruiting women, with job content involving配合 (cooperating with) AI video call tools to complete up to 100 scam video calls per day, using real faces to package AI-generated scripts to cheat money. This is no longer "AI tools being used for fraud" but a complete gray labor market with recruitment, division of labor, and薪酬结构 (compensation structure).
The structure of both stories is the same: after AI is widely deployed, where regulation hasn't arrived, the market runs first. Prediction markets are moving towards legalization, with institutional capital entry often a precursor to regulation catching up; AI deepfake fraud is moving towards scale, and industrialization makes it harder to combat. They are not the same path, but both start from the same point: the proliferation speed of AI capabilities is faster than the governance speed.
Also Worth Knowing ↓
On the day Alibaba ATH was announced, U.S. large-cap tech stocks rose broadly in pre-market trading, with Chinese concept stocks collectively strengthening. Meta and Intel rose over 3%, Alibaba rose nearly 3%, Bilibili rose nearly 3%, NIO and Li Auto rose over 2%—the strongest performance for Chinese concept stocks in nearly a week. (Source: 36Kr)
Italy's UniCredit launched a $28 billion offer to acquire the remaining shares of Germany's Commerzbank. After UniCredit acquired a minority stake in Commerzbank last year, it raised the offer to a controlling level today. The attitude of German regulators is the key variable, as European banking integration enters a real M&A phase. (Source: WSJ)
Leapmotor achieved annual profitability for the first time; the "Half-Price Li Auto" learned Li Auto's way of making money. After two years of price wars in China's EV track, the first to achieve a profitable model isn't the leader but Leapmotor, priced at half of Li Auto. The next step for survivors isn't to keep cutting prices but to build moats. (Source: QbitAI)
Google and Accel screened 4000 Indian AI startup pitches; 70% were wrappers, ultimately selecting only 5. None of the truly selected were shell companies. In the hottest AI startup market, what caught the eye of both Google and Accel were those not following the trend. (Source: TechCrunch)
LinkedIn Editor-in-Chief Daniel Roth used Claude Code to launch two iOS apps without writing a single line of code. A journalist by background with zero programming experience, he used a "builder + reviewer" dual-Agent mechanism to push the App to the App Store. This is something that actually happened, not a demo. The line where the programming barrier drops to zero moved forward another notch today. (Source: Lenny's Newsletter)
