Zuckerberg promises "personal superintelligence" after spending billions poaching AI talent from competitors. His vision sounds remarkably similar to what OpenAI already offers. Is this innovation or desperation?
Google will sign EU's AI rules while Meta refuses, creating a strategic split in Big Tech. The divide reveals two different approaches to Europe's regulatory push as new AI laws take effect August 2.
Israeli startup Teramount raised $50M to solve AI's hidden bottleneck: connecting processors. While everyone builds faster chips, these physicists found the real problem in the wires between them. Their optical solution promises 100x speed gains.
Nvidia Defies Policy Chaos With Massive China Chip Bet
Nvidia ordered 300,000 H20 chips from TSMC after Trump lifted China ban, but export licenses still haven't arrived. The regulatory limbo reveals deep splits in US policy and shows how quickly billion-dollar tech deals can hang in bureaucratic balance.
👉 Nvidia ordered 300,000 H20 chips from TSMC after surging Chinese demand convinced the company to restart production beyond its existing stockpile.
📊 The order adds to Nvidia's 600,000-700,000 existing H20 inventory after Trump reversed an April ban that threatened $5.5 billion in write-offs.
🏭 Export licenses still haven't arrived despite July assurances, leaving the entire 300,000-unit order stuck in bureaucratic limbo.
🌍 The H20 decision connects to broader US-China trade talks over rare earth magnets, with critics calling it a security risk.
🚀 Policy whiplash shows how quickly billion-dollar tech deals can shift, revealing deep US government splits on China strategy.
Nvidia ordered 300,000 H20 chips from TSMC, ditching its plan to just sell through existing inventory. The order came weeks after Trump lifted the April ban on H20 sales to China, unleashing pent-up demand from Chinese tech companies.
The move shows Nvidia believes Chinese buyers want way more chips than expected. Before the reversal, Nvidia had stockpiled 600,000 to 700,000 H20 units and planned to sell through that stash without making more. Chinese buyers had different ideas.
Sources said strong demand convinced Nvidia to restart the supply chain. The company sold around 1 million H20 chips in 2024, according to research firm SemiAnalysis. Now it's betting that number will climb much higher.
The Policy Whiplash That Started It All
The H20 story shows how confused U.S. policy can get. Nvidia built these chips for China after the U.S. blocked sales of its more powerful H100 and Blackwell processors in late 2023. The H20 delivers less computing power but still runs Nvidia's software tools.
Then in April, Trump hit the brakes. New rules banned H20 sales to China over national security concerns. Nvidia warned investors it faced a $5.5 billion inventory write-off. CEO Jensen Huang told the Stratechery podcast the company also stood to lose $15 billion in potential sales.
The ban lasted three months. In mid-July, Washington changed course and said it would approve export licenses for H20 sales. The reason: preventing Chinese rival Huawei from taking the entire market while Nvidia sat on the sidelines.
Chinese Tech Giants Queue Up
Chinese companies moved fast once the ban lifted. Tech heavyweights including Tencent, ByteDance, and Alibaba had ramped up H20 orders before the April restrictions hit. They were deploying AI models from companies like DeepSeek and building their own systems.
The sudden cutoff created severe shortages. Some Chinese firms turned to black market channels, with reports suggesting at least $1 billion worth of Nvidia GPUs were smuggled into the country during the ban. Repair services for banned chips also boomed as companies tried to keep aging hardware running.
Now that official sales can resume, the order backlog appears large. Nvidia has asked Chinese customers to submit fresh paperwork including demand forecasts to support the licensing process. The company needs those export approvals from the Commerce Department before any chips can ship.
Regulatory Limbo Continues
Those licenses never showed up. Washington said back in July they'd come through fast, but the Commerce Department is still sitting on them. Nvidia won't say where things stand with their paperwork.
The whole 300,000-unit order now sits in bureaucratic purgatory. TSMC can build the chips, but they won't ship anywhere without Washington's say-so. Not exactly what Nvidia expected after getting those July assurances.
The holdup shows how split the U.S. government remains on China policy. Lawmakers from both parties slammed the H20 decision, arguing that any advanced chip access helps China's AI ambitions.
The Huawei Competition Angle
U.S. officials justify the policy reversal as smart competition. They argue that blocking Nvidia completely would hand the Chinese market to Huawei on a silver platter. AI Czar David Sacks explained the logic: "We're not selling our latest greatest chips to China, but we can deprive Huawei of basically having this giant market share."
The concern isn't theoretical. During the ban, Huawei positioned its Ascend AI accelerators as alternatives to Nvidia hardware. Analyst Paul Triolo warned that "Huawei's Ascend 910C GPU will now become the hardware of choice for Chinese AI model developers" if the restrictions continued.
Nvidia and other U.S. companies make a similar argument. They want to keep Chinese developers using Nvidia chips and software tools rather than switching entirely to domestic alternatives. Once developers build expertise with Huawei's ecosystem, they're less likely to return.
The Broader Trade Context
The H20 decision connects to wider U.S.-China trade negotiations. Commerce Secretary Howard Lutnick linked the chip sales to talks over China's rare earth magnet exports, telling Reuters: "We put that in the trade deal with the magnets."
China had cut rare earth exports as trade tensions grew. American companies need these materials for wind turbines, fighter jets, and plenty more. The H20 deal looks like Washington trading chips for minerals.
But critics argue this approach hurts U.S. technology leadership. House Select Committee on China Chair John Moolenaar condemned the decision, noting that "the H20 is a powerful chip that, according to our bipartisan investigation, played a significant role in the rise of PRC AI companies like DeepSeek."
Production Timeline and Scale
Nvidia CEO Jensen Huang said during a Beijing visit this month that restarting the supply chain would take nine months. The 300,000-unit TSMC order suggests the company expects sustained Chinese demand well into 2026.
The order matters, but it won't flood the market. Those 300,000 new chips would roughly double what Nvidia already has sitting in warehouses. Still, it's barely enough to cover what the company sold in China last year - around 1 million units total.
Still, it shows a clear bet that Chinese appetite for AI chips will remain strong despite ongoing trade tensions. TSMC also benefits from the renewed orders, adding to the Taiwanese company's already strong AI chip manufacturing business.
Why this matters:
• Nvidia's production restart reveals just how quickly U.S.-China tech policy can shift, creating billion-dollar swings in business prospects for American companies caught in the middle.
• The episode shows China's growing leverage in trade negotiations - Washington needed Chinese cooperation on rare earth exports badly enough to reverse a national security restriction.
❓ Frequently Asked Questions
Q: What exactly is an H20 chip and how does it differ from Nvidia's other processors?
A: The H20 is a downgraded AI chip Nvidia designed specifically for China after export restrictions blocked sales of its more powerful H100 and Blackwell processors. It has less computing power but still runs Nvidia's software tools, letting Chinese developers stay in Nvidia's ecosystem.
Q: How much revenue does Nvidia make from China?
A: China and Hong Kong represent over 13% of Nvidia's total yearly revenue. The company sold around 1 million H20 chips in 2024 before the April ban, and Huang said they faced $15 billion in lost potential sales during the restriction period.
Q: Why does it take 9 months to restart chip production?
A: Semiconductor supply chains are complex, involving multiple stages from wafer fabrication to packaging and testing. TSMC needs to schedule production slots, source materials, and coordinate with Nvidia's specifications. The 9-month timeline reflects the industry-standard lead time for high-end chip manufacturing.
Q: What are rare earth magnets and why does China control the trade?
A: Rare earth magnets use elements like neodymium and dysprosium, essential for wind turbines, fighter jets, and electronics. China controls about 80% of global rare earth processing and has restricted exports as trade tensions escalated, creating supply vulnerabilities for U.S. manufacturers.
Q: How long do export licenses typically take to approve?
A: Standard export licenses can take 30-60 days, but politically sensitive applications like AI chips to China often face longer delays. The Commerce Department reviews national security implications, which can extend the process indefinitely. Nvidia's applications have been pending since mid-July.
Q: What happens to the 300,000 chips if licenses never get approved?
A: TSMC would likely halt production or redirect the chips to other markets. Nvidia could face another inventory write-off similar to the $5.5 billion charge it took after the April ban. The chips can't leave Taiwan without proper U.S. export authorization.
Q: How does Huawei's Ascend chip compare to Nvidia's H20?
A: Huawei's Ascend 910C offers competitive performance but lacks Nvidia's mature software ecosystem and developer tools. Chinese AI companies prefer Nvidia's chips when available because their existing code and workflows are built around Nvidia's CUDA platform, making switching costly and time-consuming.
Tech journalist. Lives in Marin County, north of San Francisco. Got his start writing for his high school newspaper. When not covering tech trends, he's swimming laps, gaming on PS4, or vibe coding through the night.
Zuckerberg promises "personal superintelligence" after spending billions poaching AI talent from competitors. His vision sounds remarkably similar to what OpenAI already offers. Is this innovation or desperation?
Meta copied a Chinese AI startup's techniques for Llama 4, but developers hated the result. Now Zuckerberg is spending $14 billion to fix his strategy while growth slows and investors wonder if the AI bet will pay off.
OpenAI's Study Mode tries to fix ChatGPT's academic cheating crisis by making students work for answers. The catch: kids can toggle back to regular mode anytime for instant solutions. Student willpower is the only real guardrail here.
Anthropic caps Claude Code usage while hiding usage data from paying subscribers. Users get limits they can't see coming, creating an information gap that benefits only the company. A transparency problem that could cost them developers.