Daily: Trump's new chip controls; Tariffs cost chip toolmakers; New FeRAM
5.5 min read.
Highlights
Trump’s new chip controls. In Trump’s first major chip control move, the Trump administration restricts the sale of more AI chips to China. Nvidia now requires a licence to sell even their deliberately scaled-down H20 chips to China. This will cost Nvidia US$5.5b of wasted H20 inventory in this quarter alone. Nvidia had followed the make-Trump-happy playbook, including a $1m dinner with Trump a couple weeks ago and a $500b flashy investment announcement of US-made AI supercomputers on US-soil yesterday. NPR had last reported that Trump would not go after Nvidia’s H20 chips; who knows how to deal with Trump anymore? We’re less than 100 days in and companies might need to strategise another playbook for negotiating with Trump, if that is possible at all.
Tariffs cost chip toolmakers. Trump’s tariffs will cost US chipmaking equipment companies more than US$1b a year. These are the American manufacturing companies that Trump is allegedly trying to protect.
Elsewhere, TSMC and ASML will release earnings this week and we can get a better sense of how key companies are dealing with the tariffs. But I suspect the fallout from the tariffs take at least a few months to really manifest as companies churn through inventory, and also because the foundry business model is usually built around longer-term contracts.
New FeRAM. Germany’s Neumonda and Ferroelectric Memory Company develop new FeRAMs, non-volatile memory chips, for AI applications.
Thanks for reading.
1. Policy and Geopolitics
1.1
NYT (04/16): Nvidia Says U.S. Will Restrict Sales of More of Its A.I. Chips to China
Nvidia said on Tuesday that the U.S. government had blocked the sale of some of its artificial intelligence chips to China without a license and would begin requiring a license for future sales.
The restrictions are the first major limits that President Trump’s administration has put on semiconductor sales abroad. It raises the possibility that Nvidia’s sales to China will evaporate in the coming months, bringing an end to a business that has contracted as the United States has curbed chip exports to its geopolitical rival.
Nvidia has fought hard to maintain sales to China in the face of rising U.S. government restrictions. In 2022, the Biden administration imposed rules to curb the export of Nvidia’s best A.I. chips to China. Nvidia responded by modifying one of its leading A.I. chips, the H100, so that its abilities fell below U.S. government thresholds. The resulting H20 chip became a China-specific product.
Nvidia will take a $5.5 billion charge against its revenue in the current quarter because of H20 inventory, purchase commitments and related reserves, which it won’t be able to sell or fulfill in the wake of the government’s new rule, the company said.
The write-down is a bigger strategic blow than a financial one. Nvidia, which dominates the market for semiconductors used in building artificial intelligence systems, considered selling chips to China vital to its future. If it withdrew from the market, it feared that it would surrender sales to China’s leading A.I. chipmaker, Huawei, and that Huawei would begin to challenge it for sales around the world.
1.2
Reuters (04/16): US tariffs may cost chip equipment makers more than $1 billion, industry estimates
U.S. President Donald Trump's new tariffs could cost U.S. semiconductor equipment makers more than $1 billion a year, according to industry calculations discussed with officials and lawmakers in Washington last week, two sources familiar with the matter said.
Each of the three largest U.S. chip equipment makers - Applied Materials, Lam Research and KLA - may suffer a loss of roughly $350 million over a year related to the tariffs, the sources said. Smaller rivals such as Onto Innovation may also face tens of millions of dollars in extra spending.
1.3
Reuters (04/15): Trump orders tariff probe on all US critical mineral imports
U.S. President Donald Trump on Tuesday ordered a probe into potential new tariffs on all U.S. critical minerals imports, a major escalation in his dispute with global trade partners and an attempt to push back on industry leader China.
The order lays bare what manufacturers, industry consultants, academics and others have long warned Washington about: that the U.S. is overly reliant on Beijing and others for processed versions of the minerals that power its entire economy.
China is a top global producer of 30 of the 50 minerals considered critical by the U.S. Geological Survey, for example, and has been curtailing exports in recent months.
2. Economy, Finance, and Business
2.1
Bloomberg (04/15): TSMC, ASML Outlooks to Reveal Depth of Tariff Pain, AI Angst
Earnings from two chip-industry giants this week are poised to provide an early insight into issues that have punctured investor confidence and sent valuations to multiyear lows.
Taiwan Semiconductor Manufacturing Co. and ASML Holding NV have born the brunt of a broader market selloff, weighed down by both US tariff threats and doubts over future artificial intelligence demand. Chipmaker TSMC is down 18% this year and chip-equipment maker ASML has fallen 12%.
Concern over a potential slowdown in AI demand has intensified after a string of analyst warnings, while the tariff saga has thrown global businesses’ decision making into a tailspin. All that has heightened investor focus on whether the appetite for AI chips can hold up.
2.2
Nikkei (04/16): Japan automakers make joint chip push to rival China's smart cars
Japan's automakers and suppliers are undertaking a rare industrywide effort to boost their competitiveness against Chinese rivals that have pulled ahead in the smart vehicle race.
ASRA, a consortium that includes Toyota Motor, Nissan Motor and Honda Motor, is leading discussions to create a standardized design for next-generation automotive chips that it hopes to have ready by March 2029. Standardizing chip development would help car companies reduce costs and accelerate development schedules -- potentially making up for lost time.
3. Technology
3.1
FT (04/15): TDK claims optical breakthrough to tackle generative AI’s biggest bottleneck
Japan’s TDK is claiming a breakthrough in optical technology that would process data 10 times faster than current electronics and solve a key bottleneck holding back the growth of generative artificial intelligence.
The Apple supplier said it has demonstrated the world’s first “spin photo detector”, combining optical, electronic and magnetic elements to create response times of 20 picoseconds, or 20 trillionths of a second, potentially replacing existing semiconductor-based photo detectors that transfer data between chips.
3.2
TrendForce (04/16): TSMC Reportedly to Bring FOPLP Packaging Tech to U.S., Trial Production May Start in 2027
Amid Trump’s tariffs, TSMC has been stepping up its investment in the U.S. According to Economic Daily News, the company is accelerating the timeline for its second Arizona fab. In addition, TSMC is reportedly set to introduce its latest fan-out panel-level packaging (FOPLP) technology in the U.S. to meet rising demand from customers seeking chips manufactured in the U.S.
The report highlights that, following major clients such as Apple, NVIDIA, and AMD announcing plans to expand their investments in the U.S., TSMC has accordingly adjusted its strategies.
3.3
TrendForce (04/16): Novel Memory “DRAM+”May Emerge as Germany Restarts Memory Chip Production
Recently, German semiconductor company Neumonda announced a strategic partnership with Ferroelectric Memory Company (FMC) to establish a new production line for non-volatile memory chips (FeRAM) in Dresden, Germany.
At the core of this collaboration is FMC’s innovative “DRAM+” technology. It’s reported that this technology overcomes the storage limitations of traditional FeRAM by adopting hafnium oxide (HfO₂)—compatible with sub-10nm manufacturing processes—as the ferroelectric layer, replacing the conventional lead zirconate titanate (PZT) material. As a result, the memory capacity can be increased from the traditional 4–8MB of FeRAM to the gigabit (Gb) or even gigabyte (GB) level, while retaining non-volatility, i.e., the ability to preserve data without power.
Under the agreement, FMC will develop non-volatile DRAM memory products suitable for applications in artificial intelligence, healthcare, industrial automation, automotive systems, and consumer electronics. Neumonda, which holds multiple patents in DRAM design and testing, will provide memory consulting services and offer FMC access to its Rhinoe, Octopus, and Raptor testing platforms for their DRAM+ products.