Amid tightening U.S. chip export controls to China, Shanghai-based AI chip companies, MetaX and Enflame, have rumored to downgrade chip designs to TSMC in late 2023 in order to comply with the U.S. export requirements, according to a report by Reuters on 5 June.

Regarding this rumor, TSMC declined to comment, Reuters stated.

In recent years, the U.S. has continuously introduced measures to limit China’s access to high-end chips, chip manufacturing equipment, and advanced processors.

MetaX and Enflame, which formerly claimed that their chips can rival NVIDIA’s GPUs, are recognized as “Little Giants,” a title given to startups with potential for development in key areas and valued by the Chinese government.

MetaX was founded in 2020 by former senior executives from AMD and has multiple R&D and wafer fab projects in China. Citing sources familiar with the matter, Reuters disclosed that the company reportedly developed a downgraded product, C280, in order to be manufactured by TSMC, while its inventory of its most advanced GPU, C500, was sold out earlier.

Enflame, established in 2018, counts Chinese tech giant Tencent among its backers and raised $2.7 billion last year. The company sells its products to state-owned enterprises and collaborates on projects with various local governments, Reuters reported.

In October last year, the U.S. Commerce Department’s Bureau of Industry and Security (BIS) announced a new package of export control measures, and giving specific details on the chip specifications under restriction. For instance, any chip with a total processing performance of 4,800 or higher, or a performance density of 5.92 or more, is prohibited from being shipped to China.

For context, GPU giant NVIDIA’s A100 and the even more potent H100, were banned from export to China in late 2022. The less powerful A800 and H800 chips, tailored by NVIDIA for the Chinese market, were also subject to bans last October. According to a report by Asia Times, A800’s performance is approximately 70% of the A100’s.

(Photo credit: TSMC)

On June 5th, NVIDIA CEO Jensen Huang gave a congratulatory gift to TSMC’s new chairman C.C. Wei, emphasizing that TSMC’s stock price has been undervalued. He supports Wei’s value theory and will back TSMC in its wafer and CoWoS pricing.

As per a report from Commercial Times citing sources, it’s estimated that both parties will negotiate chip prices for next year, potentially boosting TSMC’s revenue and profit margins further.

Jensen Huang revealed that he is not particularly worried about geopolitical issues because Taiwan has a strong supply chain. To ensure annual advancements in computing power, NVIDIA is building complete systems and creating more value.

Huang further emphasized that TSMC is not just manufacturing wafers but also handling numerous supply chain issues. He agrees that the current pricing is too low and will support TSMC’s price increase actions.

Notably, according to a previous Commercial Times’ report, NVIDIA’s H200 and B100 are said to adopt TSMC’s 4-nanometer and 3-nanometer processes, respectively.

Industry sources cited by Commercial Times also noted that in NVIDIA’s most recent quarterly report, its gross margin reached 78.36%, significantly outperforming AMD’s 46.78% and even TSMC’s first-quarter margin of 53.07%, exceeding it by 25 percentage points. If TSMC raises prices for its advanced processes, NVIDIA’s gross margin will remain unaffected. However, this price increase will dilute the gross margins for other companies using TSMC’s advanced processes, including Apple, AMD, and Qualcomm. 

In response to long-term capacity planning, TSMC held a board meeting on June 5th and approved a capital budget of USD 17.356 billion to expand advanced process capacity, primarily to address long-term capacity planning and the burgeoning demand for AI.

Semiconductors are fundamental to driving AI, with advanced processes and packaging being crucial competitive factors. Last year, TSMC’s advanced packaging capacity was strained, with ODM and OEM companies waiting for supply. Although there has been some relief in the first quarter of this year, the market demand is still unmet.

TSMC has confirmed the strong demand, stating that even tripling the capacity from 5nm to 3nm processes is insufficient, necessitating further capacity expansion. Sources cited by Commercial Times estimate that by the end of this year, TSMC’s CoWoS monthly capacity could reach 45,000 to 50,000 wafers, while SoIC capacity could reach 5,000 to 6,000 wafers.

Read more

• [News] A Quick Overview at TSMC’s Latest Collaboration with Intel, NVIDIA and AMD at COMPUTEX 2024
• [News] End of an Era under Mark Liu’s Leadership, TSMC Discusses Future Development Plans

(Photo credit: TSMC)

Charles Liang, President and CEO of AI server giant Supermicro, delivered a keynote speech at COMPUTEX earlier today. According to a report from CNA, while expressing his optimism in expanding the company’s market share in the Direct Liquid Cooling (DLC) field, Liang targets to capture a 15% market share in the new data center market by next year.

According to the report, Liang mentioned that Supermicro’s DLC solutions had a market share of less than 1% over the past 30 years. However, as DLC solutions now have a cost comparable to air-cooled solutions, yet more energy-efficient, more manufacturers are beginning to adopt them. It is estimated that Supermicro has around 1000 liquid-cooled cabinets shipped each month, and the aim is to capture 15% of the new data center market share within next year.

Supermicro offers comprehensive solutions at both the cabinet and data center levels. Its operations have expanded globally from the headquarter in San Jose, California, to manufacturing and operation centers in Asia and Europe. Liang stated that Supermicro’s Malaysian plant is expected to be operational in the fourth quarter of this year, with a monthly production capacity of up to 5000 cabinets, according to CNA’s report today.

Regarding Supermicro’s expansion plans, Liang said earlier that Supermicro, with over 4000 employees worldwide, its top priority would be expanding in Taiwan, aiming to finalize plans for a new campus in the near future.

In addition, Liang pointed out that AI is rapidly and significantly changing the world. Through its collaboration with GPU giant NVIDIA, Supermicro has become one of the leading suppliers of AI systems. He also emphasized that by adopting more energy-efficient systems, green computing can be cost-free and offer significant bonuses, according to the report.

While reiterating Supermicro’s strong bond with NVIDIA, Liang invited NVIDIA CEO Jensen Huang to join him on stage. When Liang introduced the liquid-cooled server cabinets on stage, Huang praised the liquid cooling solution for significantly reducing data center costs.

Read more

• [COMPUTEX 2024] The Rise of Generative AI Sparks Innovation across Industries, with Taiwan-based Companies Leading as Essential Partners in the Global Supply Chain

(Photo credit: Supermicro)

NVIDIA CEO Jensen Huang revealed that Samsung’s High Bandwidth Memory (HBM) is still striving on the certification process, but is one step away from beginning supply.

According to a report from Bloomberg on June 4th, Huang, during a briefing at the COMPUTEX, told reporters that NVIDIA is evaluating HBM provided by both Samsung and Micron Technology. Huang mentioned that there is still some engineering work needed to be completed, expressing the desire for it to have been finished already.

As per Huang, though Samsung hasn’t failed any qualification tests, its HBM product required additional engineering work. When asked about Reuter’s previous report concerning overheating and power consumption issues with Samsung’s HBM, Huang simply remarked, “there’s no story there.”

Previously, Reuters cited sources on May 24th, reporting that overheating and power consumption issues would affect Samsung’s fourth-generation HBM chip, “HBM3,” as well as the fifth-generation “HBM3e” planned for release by Samsung and its competitors this year.

Per the same report from Reuters, Samsung has been attempting to pass NVIDIA’s tests for HBM3 and HBM3e since last year, while a test for Samsung’s 8-layer and 12-layer HBM3e chips was said to fail in April.

In an official statement, Samsung noted that it is in the process of optimizing products through close collaboration with customers, with testing proceeding smoothly and as planned. The company said that HBM is a customized memory product, which requires optimization processes in tandem with customers’ needs.

Currently, SK Hynix is the primary supplier of NVIDIA’s HBM3 and HBM3e. According to an earlier TrendForce’s analysis, NVIDIA’s upcoming B100 or H200 models will incorporate advanced HBM3e, while the current HBM3 supply for NVIDIA’s H100 solution is primarily met by SK Hynix.

According to a report from the Financial Times in May, SK hynix has successfully reduced the time needed for mass production of HBM3e chips by 50%, while close to achieving the target yield of 80%.

Read more

• [News] Reasons for Samsung’s HBM Chips Failing Nvidia Tests Revealed, Reportedly Due to Heat and Power Consumption Issues
• [News] Samsung Reportedly Establishes New HBM Team, Looking to Improve AI Chip Yield

(Photo credit: Samsung)

Intel CEO Pat Gelsinger delivered a keynote speech at COMPUTEX Taipei earlier today, unveiling the next-generation client architecture set to launch this year. According to a report from CNA, He expressed gratitude to TSMC for collaborating on the development of the Lunar Lake processors, intended for the next generation of AI PCs. Currently, there are over 80 designs from 20 manufacturers.

Previously on the IFS Direct Connect event in San Jose, USA, Gelsinger pointed out in an interview that two generations of CPU Tiles would be manufactured using TSMC’s N3B process,  marking the official arrival of Intel CPU orders for laptop platforms.

Gelsinger’s interview confirms that Intel has indeed expanded its outsourcing orders to TSMC. Currently, TSMC is responsible for producing Intel CPUs, GPUs, and NPUs tiles for the Arrow and Lunar Lake platforms.

As per CNA’s report, Gelsinger announced the launch of the Xeon 6 platform and processor family designed to meet the demands of data centers, as well as the Gaudi AI accelerator. He also unveiled details of the Lunar Lake processor architecture.

As the flagship processor for the next generation of AI PCs, Lunar Lake significantly enhances graphics and AI processing, reducing system-on-chip power consumption by 40% and providing over three times the AI computing capability. Lunar Lake processors are expected to start shipping in the third quarter of this year.

Additionally, Intel plans to ship over 40 million Core Ultra processors this year, further solidifying its position in the AI PC field.

Gelsinger remarked that having 100,000 transistors on a chip would be remarkable in the early days, but now there are already 1 billion transistors on a chip, with the potential to reach even 1 trillion in the future.

Contrary to what NVIDIA CEO Jensen Huang recently described in his speech, Gelsinger indicated further that Moore’s Law is alive and well, and Taiwan continues to play a core role.

According to Gelsinger, Intel has been operating in Taiwan since 1985 and will enter its 40th year of operation next year. The partnership between Intel and Taiwan spans 39 years, and the combined initials of Intel (I) and Taiwan (T) stand for information technology (IT). Together, Intel and its Taiwanese partners can change the world once again.

Read more

• [News] Intel’s Lunar Lake Bundled Memory Reportedly Causes Uproar in the PC Supply Chain
• [News] Battle in AI PCs: Exploring Intel, AMD, and Qualcomm Chips’ AI Computing Power

(Photo credit: Intel)