Yazılar

Nvidia and SK Group Launch $500 Billion AI Infrastructure and Memory Partnership

/in /tarafından

Nvidia and South Korea’s SK Group have announced a multi-year AI initiative valued at more than $500 billion, marking one of the largest collaborations in artificial intelligence infrastructure and semiconductor development to date.

The partnership combines investments in next-generation AI data centers, advanced memory technologies, and long-term semiconductor supply agreements aimed at supporting the rapidly growing demand for generative AI, AI agents, and future intelligent computing platforms.

A central component of the initiative is a long-term strategic partnership between Nvidia and SK Hynix. Under the agreement, SK Hynix will provide next-generation memory solutions for Nvidia’s AI processors while the two companies collaborate on developing future generations of High Bandwidth Memory (HBM) technologies.

HBM has become one of the most critical components of modern AI systems, providing the ultra-fast data transfer speeds required by advanced graphics processors during AI model training and inference. As AI models continue to increase in size and complexity, demand for high-performance memory is expected to remain one of the fastest-growing segments of the semiconductor industry.

The infrastructure portion of the initiative will be led by SK Telecom, which plans to construct a 2-gigawatt AI data center powered by Nvidia’s upcoming Vera Rubin AI architecture together with SK Hynix’s HBM4 memory. The first facility is expected to become operational in 2027, representing a major expansion of South Korea’s AI computing capacity.

A 2-gigawatt campus would rank among the world’s largest AI-focused data center developments, reflecting the enormous computing resources required to support next-generation artificial intelligence applications.

Beyond the new infrastructure project, Nvidia also announced plans to expand AI computing capacity in South Korea through a separate collaboration with Naver and Brookfield. The companies intend to increase the scale of Naver’s AI data center operations, further strengthening the country’s position as an emerging regional AI infrastructure hub.

The announcement highlights how AI investment is increasingly extending beyond processors alone. Modern AI systems depend on an integrated ecosystem of advanced GPUs, high-bandwidth memory, networking technologies, and hyperscale data centers working together to deliver the computational performance required for enterprise AI workloads.

For Nvidia, the partnership helps secure long-term access to one of the world’s most important suppliers of advanced AI memory. For SK Group and SK Hynix, it reinforces their strategic role within the global AI supply chain as demand for HBM continues to accelerate.

The initiative also underscores a broader industry trend in which technology companies are forming long-term strategic alliances to secure semiconductor supply, expand AI infrastructure, and reduce supply chain uncertainty as global competition for AI computing resources intensifies.

China’s CXMT eyes $4.2 billion Shanghai listing to fund DRAM expansion

/in /tarafından

China’s leading dynamic random access memory (DRAM) chipmaker ChangXin Memory Technologies (CXMT) said on Tuesday it plans to raise 29.5 billion yuan ($4.22 billion) through an initial public offering in Shanghai, as it seeks to expand production and narrow the gap with global rivals.

According to its prospectus, CXMT will issue 10.6 billion shares, with proceeds earmarked primarily for upgrading production lines, improving manufacturing technologies and boosting research and development of advanced DRAM products. The listing follows the company’s unveiling last month of its latest DDR5 DRAM chips, directly challenging established competitors in South Korea and the United States.

Founded in 2016 with strong state backing, CXMT has become a cornerstone of China’s ambition to build a domestic memory chip industry. The global DRAM market is currently dominated by Samsung Electronics, SK Hynix and Micron Technology, which together control more than 90% of the market. CXMT held around a 4% global market share in the second quarter, according to data from Omdia cited in the prospectus.

Picture background

The company operates three 12-inch DRAM fabrication plants in Beijing and at its headquarters in Hefei, Anhui province. After nine funding rounds, CXMT counts major Chinese investors such as Alibaba and Xiaomi, and has developed four generations of DRAM technology.

CXMT is also investing heavily in high-bandwidth memory (HBM), a specialised form of DRAM essential for advanced processors such as Nvidia’s graphics processing units used in generative AI. The company plans to begin HBM production by the end of 2026 at a back-end packaging facility under construction in Shanghai.

Financially, CXMT expects strong growth. It projects revenue could rise by as much as 140% year-on-year in 2025, driven by higher memory prices and increased sales volumes since July. While the company has posted heavy losses in recent years, it said it could turn profitable as early as 2026, depending on wafer shipments and average selling prices. CXMT reported losses of 8.32 billion yuan in 2022, 16.3 billion yuan in 2023 and 7.1 billion yuan in 2024, and recorded a 2.3-billion-yuan loss in the first half of this year.

Micron tops forecasts with AI-fueled HBM demand, sees strong Q1 revenue

/in /tarafından

Micron Technology projected first-quarter revenue of $12.5 billion ± $300 million, well above Wall Street’s estimate of $11.94 billion, as booming demand for its high-bandwidth memory (HBM) chips drives growth amid the AI race.

AI demand supercharges Micron

  • Q4 HBM revenue hit nearly $2 billion, putting Micron on pace for ~$8B annually, CEO Sanjay Mehrotra said.

  • HBM chips, built by stacking DRAM vertically, reduce power use while enabling massive data processing — making them indispensable for training and running advanced AI models.

  • Micron is a key HBM supplier to Nvidia, whose dominance in AI accelerators makes HBM supply one of the most competitive battlegrounds in semiconductors.

2026 outlook already sold out

  • Micron expects to lock in deals for all 2026 HBM capacity in the coming months.

  • HBM3E pricing agreements are nearly complete; HBM4 pricing talks are ongoing.

  • “The pricing on HBM4 is actually significantly higher than the pricing on HBM3E,” said Chief Business Officer Sumit Sadana, citing tight supply and strong ROI expectations.

  • TSMC will partner with Micron to manufacture the base logic die for its HBM4E chips.

Financial performance

  • Adjusted Q4 EPS: $3.03, topping forecasts.

  • Adjusted gross margin forecast (Q1): 51.5%, far above expectations of 45.9%.

  • Analysts said stronger-than-expected pricing drove the margin boost.

U.S. policy and subsidies

  • Micron has received $6.2B under the CHIPS and Science Act, passed under former President Joe Biden.

  • Current Commerce Secretary Howard Lutnick is exploring converting subsidies into equity stakes in chipmakers, but Sadana said Micron does not expect its grant terms to change.

  • Micron recently received a disbursement after completing a milestone at its Idaho fab, Mehrotra confirmed.

Big picture

Micron is riding the wave of AI-driven chip demand, securing long-term contracts at higher prices while boosting profitability. With HBM4 set to command premium pricing, Micron is positioning itself as a critical player alongside Nvidia, Samsung, and SK Hynix in the global AI supply chain.