Huang Renxun took the stage in Taipei today. The N1X specifications have been fully leaked, and NVIDIA is looking to bring CUDA to laptops. In the same week, the U.S. expanded its chip ban from the border to corporate nationality.

1｜NVIDIA Unveils N1X Chip, Bringing CUDA Ecosystem from Data Centers to Personal Computers

NVIDIA, Microsoft, ARM, and MediaTek released a joint teaser last Friday with the caption “A New Era for PCs,” and Huang Renxun took the stage in Taipei this morning. The N1X chip specifications were leaked in advance, featuring a 20-core ARM v9.2 CPU, TSMC 3nm process, a GPU with 6,144 CUDA cores, on par with the desktop RTX 5070, with Geekbench scores 15% higher than Qualcomm Snapdragon X Elite in single-core performance.

The core of this chip lies not in performance figures, but in CUDA. Mainstream AI frameworks such as PyTorch and TensorRT can run natively on the N1X laptops without the need for code modification, a feat Qualcomm cannot achieve. Dell, Lenovo, ASUS, and MSI have confirmed the first wave of Windows ARM devices. The laptop processor market will shift from a binary x86 landscape to a three-way competition: x86 retaining market share, Qualcomm occupying the thin-and-light segment, and NVIDIA securing the high-performance ARM sector.

2｜U.S. Extends Chip Ban to Overseas Chinese Companies, Shifting Control Logic from Geography to Nationality

Reuters and CNBC reported that the U.S. is taking steps to prevent NVIDIA’s AI chips from flowing to Chinese companies overseas. Previous restrictions were based on geography, prohibiting the export of advanced chips to entities within China. The new measures will extend restrictions to companies’ nationalities, covering Chinese-affiliated companies operating in third countries such as Southeast Asia and the Middle East. The focal point of control is no longer “where the chips are sold” but “who is buying them.”

Enforcement actions have been accelerating. In February, Applied Materials was fined $252 million for illegally exporting ion implantation equipment to China, marking the Commerce Department’s second-largest fine in history. In March, the FBI arrested three individuals for allegedly purchasing 750 servers for China (worth $170 million) and falsely certifying the end-user. Congress approved a 23% increase in the Industrial Security Bureau’s fiscal year 2026 budget, with special appropriations aimed at semiconductor enforcement. The entire chain from legislation to enforcement is tightening in synchronization.

3｜SoftBank Invests €750 Billion in France to Build AI Data Center, Nuclear Grid Becomes Infrastructure Moat

SoftBank has announced a maximum investment of €750 billion (approximately $870 billion) in an AI data center in France to build a 5-gigawatt AI computing capacity, marking SoftBank’s largest AI infrastructure bet in Europe. The first phase of €450 billion will involve the construction of 3.1 gigawatts of data centers in Dunkirk, Bourbourg, and Boulogne-Calais in the Hauts-de-France region, with delivery expected by 2031. Partners include French utility EDF and Schneider Electric.

Speaking to French media, Masayoshi Son stated that France is an “energy-producing and exporting country,” a factor that is “absolutely decisive.” With 70% of its electricity coming from nuclear power, France is the world’s largest net exporter of electricity, with industrial electricity prices at less than half of those in the UK. This is precisely what has been lacking in the site selection for American data centers. The key variable in the AI computing power race is shifting from “who has the most GPUs” to “who has the most stable power supply.” France’s nuclear grid represents an irreplicable advantage in this race.

4｜AI Electricity Consumption Doubles in One Year, Triggering Three-Way Tug of War on Same Infrastructure

Axios reports that from tech giants to automakers, the U.S. economy is collectively diving into the energy business. Electricity has transitioned from a “cheap and abundant commodity” to the most valuable strategic asset. IEA data confirms this assessment: global data center electricity consumption is projected to grow by 17% in 2025, with a 50% surge in AI-specific facility energy consumption. By 2030, data center electricity usage is set to double from 485 terawatt hours to 950 terawatt hours, comprising 3% of global electricity demand. Tech companies are forecasted to exceed $400 billion in capital expenditures by 2025, with an additional 75% increase projected by 2026.

This same fact has triggered three vastly different responses. Retired Air Force Lieutenant General David Deptula warned in The Washington Post that a scarcity of computing power poses a “disastrous” threat to national security, as data centers will determine who wins the next war. Environmental activist Erin Brockovich is targeting data centers for transparency, demanding disclosure of water and energy usage data. Community residents are resisting the siting of data centers and accompanying power plants. The dual identity of the same infrastructure as a national security asset and an environmental burden remains irreconcilable.

5 | The Stock Market Walking a Tightrope Without a Safety Net, AI Simultaneously Creating Market Value and Destroying Jobs

Allianz’s Chief Economic Advisor Mohamed El-Erian warned in the Financial Times that the “put option of policy” that has supported global markets for decades is disappearing. In the past, every time the stock market experienced a sharp decline, central banks would cut interest rates, and governments would provide stimulus to support the market. This time is different: the global oil shock continues to weigh on the market, the U.S. government’s debt level restricts fiscal maneuvering, and the Federal Reserve cannot cut rates until inflation subsides. The safety net is no longer there, yet the stock market continues to reach new highs.

The historic rebound of chip stocks is a concentrated reflection of AI optimism. Bloomberg reports that the debate on the AI bubble is becoming serious, and the Financial Times notes that Wall Street’s bullish bets on the rebound will ignore bubble concerns. Adding tension to this situation is the impact of AI on employment. Wix’s CEO announced a 20% workforce reduction (about 1,000 people), directly attributed to the advancement of AI capabilities. An MIT professor pointed out that the pattern of companies “using AI as an excuse for layoffs” has been ongoing for 20 years, with the tech industry expected to have cut a cumulative 134,000 jobs by 2026. The speed at which AI creates market value and destroys jobs is equally astounding.

Also Worth Knowing

A large investor sold $1.26 billion worth of BlackRock’s IBIT shares in a dark pool in a single trade, accepting a 2.3% discount ($29.5 million loss). Over the past two weeks, Bitcoin ETFs have seen a cumulative outflow of $2.26 billion, reducing total assets from $107.8 billion to $94.2 billion. Speed precedes price, signaling a shift in confidence among large institutional investors.

DTCC subsidiary DTC plans to start live trading of tokenized securities in July and fully launch in October. Wall Street’s traditional settlement infrastructure is officially integrating with blockchain technology. Concurrently, Congress is advancing the Tokenization Modernization Act and the CLARITY Act, with both legislative and infrastructure aspects progressing simultaneously to facilitate the implementation of tokenization.

Quietly, China’s approximately 1.3 billion barrels of strategic petroleum reserves have become a hidden buffer in the global oil market. Consulting firm FGE expects China’s oil stocks to increase by another 266 million barrels (730,000 barrels per day) by 2026. Amid a situation where one-fifth of the Gulf’s supply is blocked, China’s strategic low-cost purchases and stock releases have helped stabilize oil prices in extreme scenarios.

The SEC has sued Privvy founder Nathan Fuller, accusing him of operating a $12.3 million Ponzi scheme under the guise of an “AI trading bot.” He promised a 40-50% return in 30-45 days, but only 3% of the funds were used for crypto trading, with the rest used for personal expenses and to pay early investors. The 150 victims are spread across nine states and two countries. The AI label is becoming a new packaging for financial fraud.

Both Fenghua High-Tech and Baoding Technology issued clarifications on the same day, denying entry into Nvidia’s supply chain certification. Baoding Technology reported a loss of $1.85 million in its copper foil business by 2025, with ultra-thin copper foil revenue of only $100,000. The gap between the A-share market’s “Nvidia concept” hype and the actual business of listed companies is being illuminated by regulatory and announcement scrutiny.

Song Jiaming, the lead author of the DDIM paper, has announced his resignation. The continued mobility of key researchers in the diffusion model field often foreshadows the next wave of technology investment.

[BlockBeats]