DELIVERCHIP（HK）LIMITED — 31 March: This Week’s News

Highlights of the Week

1. SK Hynix accelerates equipment installation at its M15X DRAM plant to meet surging demand for Broadcom’s HBM

2. Samsung vows to reclaim leadership in the AI memory market: HBM4 chips to enter production in the second half of the year

3. Sharp sells assets related to its SDP plant to SoftBank for 4.8 billion yuan

4. General Motors to use NVIDIA AI chips for vehicle and factory automation

SK Hynix Accelerates Equipment Roll-out at M15X DRAM Plant to Meet Surge in Broadcom HBM Demand

According to South Korean media outlet TheElec, SK Hynix plans to bring equipment into its new M15X wafer fab two months ahead of schedule.

Sources indicate this is due to a surge in orders for High Bandwidth Memory (HBM) from customers, particularly Broadcom.

SK Hynix also plans to expand the production capacity of M15X beyond the originally planned 32,000 wafers per month. The new capacity target is to be finalised next month. Sources indicate that capacity could nearly double.

The M15X plant in Cheongju was originally scheduled to begin equipment installation in December, but suppliers have been asked to deliver the equipment in October.

The wafer fab is expected to produce 1b DRAM, which SK Hynix is using as the core chip for HBM3E. The South Korean memory chip manufacturer has already expanded its 1b DRAM production capacity at its existing plants. Monthly production capacity for 1b DRAM is expected to reach 178,000 wafers by the end of the year. When the M15X facility comes on stream at the end of 2026, SK Hynix will secure a production capacity of 240,000 wafers per month. The primary reason for bringing forward production is the exceptionally large order volume from Broadcom. SK Hynix plans to begin producing HBM for its latest major client in the third quarter.

Sources indicate that Broadcom is expected to account for 30 per cent of SK Hynix’s total HBM production capacity by the end of this year. Last year, SK Hynix announced plans to invest 20 trillion won in the construction of M15X.

Meanwhile, the company also announced on 19 March that it had provided HBM4 12H samples to customers. SK Hynix stated that HBM4 can process 2TB of data per second, has a capacity of 36GB, and is manufactured using its advanced large-scale reflow moulding underfill technology.

Samsung vows to reclaim leadership in the AI memory market: HBM4 chips to go into production in the second half of the year

Jun Younghyun, head of Samsung’s chip business, stated that Samsung plans to supply enhanced 12-layer HBM3E as early as the second quarter of this year, and intends to produce cutting-edge HBM4 chips in the second half of the year. He was formally appointed as Samsung’s co-CEO following the company’s annual general meeting.

Jun Young-hyun acknowledged that Samsung’s failure to secure an early lead in the HBM market had caused it to fall behind its rival SK Hynix. He emphasised that Samsung would not repeat the same mistakes with the next-generation HBM4 and custom chips. HBM4 memory is expected to be integrated into NVIDIA’s upcoming Rubin GPU architecture. SK Hynix is actively positioning itself as NVIDIA’s primary HBM4 supplier and has announced that it has delivered the world’s first 12-layer HBM4 samples to key customers ahead of schedule.

Samsung has decided to modify the design of its HBM3E chips to secure NVIDIA’s approval. This approval has been a long time coming, and Samsung is working hard to catch up with SK Hynix. At this year’s CES in Las Vegas, NVIDIA CEO Jensen Huang stated that Samsung would have to come up with a new design, “but they can do it. They work very quickly. They are very committed to making this happen.”

Jeon Young-hyun also predicted that the memory market would recover in the coming quarters, driven by strong demand for AI and mobile devices, and he expects this to boost Samsung’s profits in the second half of the year.

Sharp Sells Assets Related to SDP Plant to SoftBank for 4.8 Billion Yuan

Sharp recently announced that it would sell fixed assets—including land and factory buildings—related to its liquid crystal display (LCD) panel plant (SDP) in Sakai City, Osaka Prefecture, Japan, to SoftBank for 100 billion yen (approximately 4.868 billion yuan). It is reported that SoftBank has already signed a sales agreement with Sharp, with the primary purpose of the acquisition being to build a large-scale AI data centre.

At Sharp’s financial results briefing last May, the company decided to cease production at the SDP Sakai plant’s G10 line in the first half of 2024 (by the end of September). According to Foxconn Group, Sharp’s largest shareholder, the Sakai plant will be converted into an AI data centre in the future to facilitate Sharp’s transition towards a light-asset business model.

SoftBank has recently confirmed that it will invest nearly 5 billion yuan to acquire the relevant assets of the SDP Sakai plant. It plans to utilise the approximately 450,000 square metres of land and the factory buildings with a total floor area of approximately 840,000 square metres at Sharp’s Sakai plant to build an AI data centre with a power capacity of approximately 150 MW. The centre is scheduled to commence operations by the end of 2026, with plans to expand its power capacity to 250 MW in the future.

SoftBank has stated that, in addition to being used for the research and development of generative AI and other AI-related ventures, this AI data centre will also be made widely available to universities, research institutions and enterprises to meet various external usage requirements.

General Motors to Use NVIDIA AI Chips for Vehicle and Factory Automation

General Motors recently announced that it will utilise NVIDIA’s artificial intelligence chips and software to develop autonomous driving technology for its vehicles and to improve factory workflows.

Traditional car manufacturers have long been striving to commercialise autonomous driving technology, which has proved more challenging and costly than anticipated, but has become a means of boosting sales and generating subscription revenue from drivers.

GM and NVIDIA plan to collaborate on building AI systems using NVIDIA’s platform, with the aim of training AI manufacturing models for factory planning. GM also plans to utilise NVIDIA’s autonomous driving technology for future advanced driver-assistance systems.

When asked about the financial terms, a General Motors spokesperson stated that the two companies have a strategic partnership regarding the use of artificial intelligence in manufacturing, and that General Motors will purchase chips from NVIDIA for driver-assistance technology.

This year, many car manufacturers and suppliers, including Toyota and Hyundai, have partnered with NVIDIA to develop their own autonomous driving capabilities in response to competition from Tesla. Tesla uses proprietary technology to operate its Full Self-Driving system.

GM uses NVIDIA’s chips to power its autonomous Cruise robotaxis. However, before shutting down its driverless taxi business last year to focus on autonomous technology for private vehicles, the company had begun developing custom chips in-house to reduce costs and dependence on external suppliers. GM forecasts that its Super Cruise driver-assistance technology will generate total revenue of approximately US$2 billion within five years.

All news in this article is sourced from the internet.