Daily: U.S.-China trade talks end; Huawei downplays chip capabilities; Shake-ups in memory landscape
6.5 min read.
Highlights
U.S.-China trade talks end. U.S. and China agree in-principle to a “framework to implement the Geneva Consensus,” according to Commerce Secretary Howard Lutnick. The language is as soft as you can make it, so I would still consider the U.S.-China trade situation very precarious. Nothing is confirmed yet, though it seems there is an understanding that China will back down on the rare earth restrictions (by approving licences), and the U.S. will reciprocate in Lutnick’s words, “our export implementation will come down as well.” This is vague and unclear whether it refers to chip sanctions or trade tariffs, some combination of both, or something else entirely. Stay tuned as both sides return to report back to their respective leaders and more details are hashed out.
Huawei downplays chip capabilities. Huawei’s founder Ren Zhengfei gave an interview to the People’s Daily, saying that Huawei’s chip “still lags behind the U.S. by one generation.” Ren said the West was “exaggerating” Huawei’s capabilities and Western media reported Ren’s interview as “downplaying” Huawei’s capabilities. I feel like the right tone should be “Huawei lags behind the U.S. by only one generation.” That’s still very impressive!
And I don’t think anyone is claiming that Huawei is on par with Nvidia. Western, and now Chinese, reporting all agree that Huawei’s latest Ascend chips are roughly H20 or H100 equivalents, which is one generation behind Nvidia’s latest B100 chips. So, everyone is on the same page and this is a matter of framing and timing, given the ongoing U.S.-China trade talks.
Shake-ups in memory landscape. SK Hynix offered a very ambitious, long-term vision for memory innovation, including 4F Vertical Gate and 3D DRAM technologies.
Meanwhile, Micron beat Samsung and SK Hynix to become the first supplier of SOCAMM (advanced AI memory) chips to Nvidia.
Thanks for reading.
1. Policy and Geopolitics
1.1
Bloomberg (06/11): US, China Officials Say Consensus Reached on Geneva Framework
The US and China have agreed in principle to a framework for deescalating trade tensions by implementing the consensus they reached in Geneva, negotiators for both sides said.
“We have reached a framework to implement the Geneva consensus,” US Commerce Secretary Howard Lutnick told reporters in London.
US Trade Representative Jamieson Greer said there were no other meetings scheduled, but added that the American and Chinese sides talk frequently and are able to do so whenever they need.
“We do absolutely expect that the topic of rare Earth minerals and magnets with respect to the United States of America will be resolved in this framework implementation,” Lutnick said.
“Also, there were a number of measures the United States of America put on when those rare earths were not coming,” Lutnick added. “You should expect those to come off — sort of, as President Trump said, in a balanced way. When they approve the licenses, then you should expect that our export implementation will come down as well.”
1.2
Reuters (06/11): Trump's AI czar downplays risk AI chip exports could be smuggled
White House AI czar David Sacks on Tuesday downplayed the risk that coveted American AI chips could be smuggled to bad actors, and expressed concern that regulating U.S. AI too tightly could stifle growth and cede the critical market to China.
"We talk about these chips like they could be smuggled in the back of a briefcase. That's not what they look like. These are server racks that are eight feet tall and weigh two tons," Sacks said at the AWS summit in Washington.
1.3
SCMP (06/10): Chinese academy launches automated system to speed up chip design amid US software curbs
The Chinese Academy of Sciences (CAS), the nation’s premier science institution, has unveiled a chip design system driven by artificial intelligence (AI) that has the potential to significantly accelerate semiconductor development and replace human programmers.
Developed by the State Key Laboratory of Processor and the Intelligent Software Research Centre, both under CAS, and the University of the Chinese Academy of Sciences, the fully automated integrated circuit design system is called QiMeng, which means “enlightenment” in Chinese.
1.4
Reuters (06/10): Nvidia, HPE to build new supercomputer in Germany
Nvidia and Hewlett Packard Enterprise said on Tuesday they are partnering with the Leibniz Supercomputing Centre to build a new supercomputer using Nvidia's next-generation chips.
The Blue Lion supercomputer, as the project is called, will become available to scientists in early 2027, using Nvidia's "Vera Rubin" chips.
2. Economy, Finance, and Business
2.1
FT (06/10): China’s Huawei plays down its chipmaking capabilities
Huawei’s founder said the US was exaggerating the Chinese chipmaker’s capabilities and played down his company’s technology amid trade talks between Beijing and Washington that include discussions of export controls.
In a rare interview with China’s state-run People’s Daily on Tuesday, Ren Zhengfei said Huawei’s Ascend chip, the main rival to Nvidia’s products in China, “still lags behind the US by one generation”. He added that the “US has exaggerated Huawei’s capabilities — we’re not that strong yet”.
2.2
Bloomberg (06/10): TSMC’s May Revenue Surges 40% in Sign of Resilient AI Demand
Taiwan Semiconductor Manufacturing Co., the main chipmaker for Nvidia Corp. and Apple Inc., reported a 40% jump in May revenue after companies stockpiled chips in response to mounting trade uncertainty.
The Hsinchu, Taiwan-based company’s revenue reached NT$320.5 billion ($10.7 billion). That compares with a 48% gain during April. Analysts on average expect a 39% increase in TSMC’s second-quarter sales. While sales in May appeared to be relatively healthy, they dropped 8.3% from April.
2.3
TrendForce (06/10): TSMC Speeds Up Arizona Expansion, yet U.S. Packaging Plant Sites Reportedly Remain Up in the Air
TSMC is fast-tracking its U.S. expansion with a $100 billion investment covering three fabs, two advanced packaging plants, and an R&D center. However, supply chain insiders suggest the locations for the two U.S.-based packaging plants have yet to be decided, according to Liberty Times.
2.4
Nikkei (06/10): Honda to invest in Japanese chipmaker Rapidus
Honda Motor is preparing to invest in Japanese chipmaker Rapidus to domestically procure semiconductors for next-generation automobiles, Nikkei has learned.
Rapidus already has Toyota Motor as its main shareholder. By backing the company established in August 2022, the two major Japanese automakers will secure sources of chips made in the country, which will also help Rapidus begin mass production of cutting-edge products and find customers.
3. Technology
3.1
TrendForce (06/10): SK hynix Charts DRAM Roadmap, Highlighting 4F VG and 3D DRAM Adoption for Sub-10nm Era
Following its March rollout of 12-layer HBM4 samples, SK hynix unveiled a long-term DRAM roadmap and sustainability vision at the 2025 IEEE VLSI Symposium in Kyoto (June 8–12), highlighting plans for sub-10nm technologies, according to its press release and ZDNet.
Notably, in his keynote on June 10, CTO Cha Seon Yong highlighted the growing challenges of scaling with current technologies, adding that SK hynix will adopt 4F Vertical Gate and 3D DRAM technologies for next-gen nodes at 10nm and below, backed by breakthroughs in structure, materials, and components.
As noted by ZDNet, the 4F VG (Vertical Gate) platform is a next-gen DRAM technology designed to minimize cell area and enable higher density, faster speeds, and lower power consumption through a vertical gate structure.
3.2
Korea Economic Daily (06/10): Micron set to supply SOCAMM chips to Nvidia ahead of Samsung, SK Hynix
In a move that could reshape the global memory industry’s competitive dynamics, Nvidia Corp. has chosen US chipmaker Micron Technology Inc. as its first supplier of the next-generation memory solution known as SOCAMM.
SOCAMM, short for small outline compression attached memory module, is a new type of high-performance, low-power memory for AI servers in data centers.
The technology is closely watched in the industry, with some dubbing it the “second HBM,” or high-bandwidth memory, given its critical role in enabling AI acceleration.
3.3
Bloomberg (06/10): Qualcomm Demonstrates New Processor for AI-Based Smart Glasses
Qualcomm Inc., the top maker of smartphone processors, is making a bigger push into smart glasses with low-power components that can run artificial intelligence software.
The chipmaker demonstrated a pair of glasses at the Augmented World Expo trade show that rely on its new Snapdragon AR1+ Gen 1 processor. The idea is to let glasses run AI without having to be connected to the internet or rely on a phone.
Qualcomm Senior Vice President Ziad Asghar said he was able to chat with a fully functional AI assistant — based on the Llama 1B model — run by the company’s chips contained in the glasses. “No phone. No cloud. Just the processor powering the glasses themselves,” he said in a blog post Tuesday timed to the event in Long Beach, California.
3.4
SCMP (06/10): Chinese institute begins photonic chip production despite US curbs
An institute affiliated with Shanghai Jiao Tong University has commenced production of photonic chips, marking a milestone in China’s semiconductor sector and providing a boost to the country’s quantum computing capabilities in spite of increased US export controls.
3.5
WSJ (06/10): IBM Has a Roadmap to a ‘Fault-Tolerant’ Quantum Computer by 2029
International Business Machines said Tuesday it has a plan for building what it calls the world’s first large-scale, fault-tolerant quantum computer at its New York data center before the end of the decade.
The computer, called IBM Quantum Starling, will be housed in its Poughkeepsie, N.Y., center and have 20,000 times the computational power of today’s quantum computers, the tech giant said.
-