Latest news with #HBM3
Korea Herald
22-07-2025
- Business
- Korea Herald
Diverging prospects: Samsung eyes rebound, SK braces for HBM shakeup
As fears of HBM oversupply mount, intensifying price competition puts pressure on chipmakers, reshaping AI memory landscape As concerns over a potential supply glut in the high-bandwidth memory market for artificial intelligence semiconductors intensify, South Korea's two memory chip giants — Samsung Electronics and SK hynix — are seeing sharply divergent stock performances. According to the Korea Exchange on Tuesday, so far in July, Samsung Electronics' shares have surged 13.38 percent on the country's main bourse, Kospi, outpacing the broader index's 4.53 percent rise. On the contrary, SK hynix's shares have dropped 6.68 percent during the same period. Analysts say growing fears of oversupply and falling prices in the HBM segment, a key component in AI chipsets, are driving foreign investors to offload shares. This is particularly the case for those of SK, since it holds the No. 1 position in the global HBM market. Samsung's entry into the market of sixth-generation HBM chips, known as HBM4, is expected to erode the price premium previously enjoyed by SK hynix with its fifth-generation HBM3E products. 'There are rising concerns in the market about potential oversupply of HBM from this year and onward,' said Ryu Young-ho, a senior analyst at NH Investment & Securities. 'It has triggered increased selling pressure on SK hynix shares.' At the same time, Samsung Electronics' stock — long overshadowed by SK hynix — has begun to recover as the company secured supply deals for both HBM3 and HBM3E with AMD and Broadcom, signaling renewed confidence in its technological competitiveness. 'If Samsung succeeds in entering Nvidia's HBM supply chain, the oversupply issue could become even more pronounced,' Ryu said. Currently, SK dominates the global HBM market with a 53 percent share, followed by Samsung with 38 percent. With Micron joining the fray, competition is expected to intensify. Samsung shares are also benefiting from expectations of increased HBM revenue, further buoyed by the recent lifting of export restrictions on Nvidia's China-specific AI chip, the H20. With new market avenues opening, Samsung's momentum in the HBM space appears to be strengthening. As the market dynamic shifts, industry insiders are watching whether the HBM sector will transition from a technology-driven race to a price-driven war. Until recently, tight supply allowed suppliers to dictate prices. But as supply loosens, big tech clients like Nvidia and AMD are likely to gain leverage in price negotiations. 'The balance of power is shifting from suppliers to buyers,' said Ryu. 'We may see weaker-than-expected price hikes for the upcoming HBM4 chips.' Investment bank Goldman Sachs also forecast that HBM prices could fall by around 10 percent next year, citing the likely impact of increased competition and capacity. 'HBM has entered a maturing phase,' said an industry source who requested anonymity. 'As supply constraints ease, chip makers will need not only technical excellence but also strong price competitiveness to survive. The real price war is likely to begin in earnest next year.' Against this backdrop, both Samsung and SK are racing to mass-produce HBM4 by the second half of this year. SK was the first chip maker to supply 12-layer HBM4 samples to Nvidia in March. Samsung is expected to follow suit within this month, shipping its samples to both Nvidia and AMD. In an effort to regain lost ground after delays with HBM3E chips, Samsung is utilizing its advanced 1c DRAM technology, which employs a 10-nanometer process, in HBM4, aiming for high precision and yield stability. It is also actively investing in its Pyeongtaek Campus Line 4 (P4) to support volume production. 'Samsung appears to be rebuilding its strategy with a long-term perspective to win back customer trust, rather than focusing on short-term results,' said an anonymous source who is familiar with the matter. 'Though Samsung may be behind SK in HBM development timing, if it can deliver on performance, it still has a strong chance to catch up to its rivals.' SK, meanwhile, is prioritizing stable yields by using the same 1b, the fifth generation of the 10nm process and mass reflow-molded underfill packaging method from HBM3E. It is also deepening collaboration with Taiwan's TSMC to enhance both HBM4 production and advanced packaging capabilities. 'We are accelerating open collaboration with global customers as well as TSMC, not only to develop the highest-performance HBM4 but to reinforce our position as a total AI memory provider,' said Kim Joo-sun, SK hynix president in charge of AI infrastructure.
Time of India
16-07-2025
- Business
- Time of India
US chip policy shift boosts outlook for Samsung, SK Hynix
Seoul: The recent decision by the United States to lift restrictions on Nvidia's AI chip exports to China is expected to significantly benefit South Korean chipmakers Samsung Electronics and SK Hynix , particularly in the high-bandwidth memory (HBM) segment, as reported by The Korea Herald. On Tuesday, US Commerce Secretary Howard Lutnick confirmed that the government has cleared Nvidia to sell its H20 graphics processing units (GPUs) to China. The move is part of broader trade talks with Beijing involving rare earth elements and is aimed at increasing China's dependence on American technology. "We don't sell them our best stuff, not our second-best stuff, not even our third-best". Lutnick added, noting that the H20 is Nvidia's "fourth-best" chip. In April, President Donald Trump had blocked H20 exports to China to limit China's access to advanced US technology. The resulting export freeze left Nvidia with unsold stock, leading to a USD 4.5 billion charge and a USD 2.5 billion drop in quarterly revenue. Now that exports are back on track, demand for HBM chips--which power AI accelerators and high-performance servers by delivering faster data processing than traditional DRAM--is expected to rebound. Samsung and SK Hynix, key HBM suppliers to Nvidia, are likely to see a positive impact. The H20 chip includes both HBM3 and the newer HBM3E memory. While it may not excel at training AI models, it is optimized for inference tasks, an area of AI experiencing rapid growth. Samsung has reportedly been supplying HBM3 for the H20 since last year. However, the earlier export ban had taken a toll, contributing to a 55.9 per cent drop in Samsung's operating profit year-over-year for the second quarter. Meanwhile, SK Hynix, which currently holds the top market share in HBM and was the first to supply HBM3E to Nvidia, stands to gain in the short term from the restored H20 exports. "It's a short-term positive development for the AI GPU and HBM sectors that Nvidia is now able to resume exports to China," said Ryu Hyung-keun of Daishin Securities. "SK Hynix is expected to benefit in the near term from the resumption of H20 shipments."
Time of India
16-07-2025
- Business
- Time of India
US chip policy shift boosts outlook for Samsung, SK hynix
Academy Empower your mind, elevate your skills The recent decision by the United States to lift restrictions on Nvidia 's AI chip exports to China is expected to significantly benefit South Korean chipmakers Samsung Electronics and SK hynix , particularly in the high-bandwidth memory (HBM) segment, as reported by The Korea Tuesday, US Commerce Secretary Howard Lutnick confirmed that the government has cleared Nvidia to sell its H20 graphics processing units (GPUs) to move is part of broader trade talks with Beijing involving rare earth elements and is aimed at increasing China's dependence on American technology."We don't sell them our best stuff, not our second-best stuff, not even our third-best". Lutnick added, noting that the H20 is Nvidia's "fourth-best" April, President Donald Trump had blocked H20 exports to China to limit China's access to advanced US technology. The resulting export freeze left Nvidia with unsold stock, leading to a $4.5 billion charge and a $2.5 billion drop in quarterly that exports are back on track, demand for HBM chips—which power AI accelerators and high-performance servers by delivering faster data processing than traditional DRAM—is expected to rebound. Samsung and SK hynix, key HBM suppliers to Nvidia, are likely to see a positive H20 chip includes both HBM3 and the newer HBM3E memory. While it may not excel at training AI models, it is optimized for inference tasks, an area of AI experiencing rapid has reportedly been supplying HBM3 for the H20 since last year. However, the earlier export ban had taken a toll, contributing to a 55.9% drop in Samsung's operating profit year-over-year for the second SK hynix, which currently holds the top market share in HBM and was the first to supply HBM3E to Nvidia, stands to gain in the short term from the restored H20 exports."It's a short-term positive development for the AI GPU and HBM sectors that Nvidia is now able to resume exports to China," said Ryu Hyung-keun of Daishin Securities. "SK hynix is expected to benefit in the near term from the resumption of H20 shipments."
Yahoo
24-03-2025
- Business
- Yahoo
Meta Offered $800M. This Startup Said No.
FuriosaAI just made a gutsy calland investors should take note. The Seoul-based AI chip startup turned down an $800 million acquisition offer from Meta (NASDAQ:META), choosing to chart its own course instead of getting absorbed by Big Tech. Talks reportedly collapsed not over money, but over how the company would be run post-acquisition. FuriosaAI, founded by June Paik (ex-Samsung, ex-AMD (NASDAQ:AMD)), has been quietly building serious firepower: its second-gen chip, RNGD, uses TSMCs 5nm process and SK Hynixs HBM3 memory to take on Nvidia (NASDAQ:NVDA), Groq, and Cerebras in AI inferencing. The chips already in trials with LG AI Research and Aramcoand it's set to launch later this year. Warning! GuruFocus has detected 3 Warning Signs with NVDA. Rather than cash out, FuriosaAI is now raising around $48 million to fuel its next phase of growthlikely a lead-up to an IPO. It's one of the few Asian startups to catch Metas eye, and it's betting big on staying independent. That decision has real consequences: shares of DSC Investment, a key early backer, dropped over 16% after the Meta deal fell through. Meanwhile, Meta is moving full steam ahead with its $65 billion AI infrastructure push, building its own chips to lessen reliance on Nvidia. But with FuriosaAI staying privatefor nowthe chip wars just got a lot more interesting in Asia. The bigger picture? This is a signal shift. Emerging chipmakers arent just waiting to be boughttheyre gunning for market share. FuriosaAIs refusal could inspire a new wave of homegrown challengers in the AI hardware race, especially across Asia. And for investors watching the next generation of semiconductor disruption unfold, this might be the one to keep on your radar. This article first appeared on GuruFocus.
Yahoo
31-01-2025
- Business
- Yahoo
High Bandwidth Memory Market to Worth Over US$ 5,810.5 Million By 2033
The high bandwidth memory market thrives on HPC expansion, demanding stacked solutions, advanced interposers, and seamless integration, enabling faster data flows, lowered latency, and elevated performance gains in complex workloads. New Delhi, Jan. 31, 2025 (GLOBE NEWSWIRE) -- The global high bandwidth memory market was valued at US$ 501.0 million in 2024 and is expected to reach US$ 5,810.5 million by 2033, at a CAGR of 31.3% during the forecast period 2025-2033. The field of high-performance computing increasingly relies on innovative memory solutions to meet the demands of artificial intelligence, data analytics, and other computationally intensive processes. Among these, high bandwidth memory market stands out for its capacity to deliver rapid data transfer speeds and improved efficiency over traditional counterparts. Recent developments emphasize significant expansions in production lines, technological advancements in chip stacking, and resource-intensive improvements in voltage and architecture. Download Sample Copy @ As production ramps up, performance gains become all the more critical. HBM2E, a key iteration in this lineage, achieves a data transfer rate of up to 3.6 Gbps, translating into a bandwidth of up to 460 GB/s per stack. This iteration employs a 12-Hi stack architecture that provides a total density of 24GB, striking a balance between speed and capacity. However, the performance story goes well beyond HBM2E, as HBM3 offers a notable leap with data transfer rates reaching up to 6.4 Gbps, resulting in an impressive 819 GB/s of bandwidth per stack. This performance surge is particularly influential for tasks requiring near-instantaneous data retrieval. Additionally, HBM3 supports a 16-Hi stack architecture, effectively doubling total density to 64GB compared to previous generations. Alongside improvements in capacity of the high bandwidth memory market, HBM3 facilitates lower power consumption by reducing the core voltage to 1.1 volts from HBM2E's 1.2 volts. This change not only results in better power efficiency but also positions HBM3 as a more attractive choice for energy-conscious data centers and high-performance computing systems. Technical enhancements, such as stacking innovations and refined interface protocols, collectively usher in a new realm of possibility. These breakthroughs directly influence how AI training, machine learning, and complex simulations are undertaken, helping industries push the boundaries of what is computationally feasible. With each performance milestone, stakeholders—from chip designers to software developers—stand to benefit from memory solutions that are increasingly tailored to support robust, real-time processing. Key Findings in High Bandwidth Memory Market Market Forecast (2033) US$ 5,810.5 million CAGR 31.3% Largest Region (2024) North America (36.7%) By Product Type CPU (35.4%) By Application Data Center (38.40%) Top Drivers Growing HPC codes requiring multi-tier memory bandwidth for complex simulations AI accelerators demanding faster data interchange for insight generation tasks Advanced HPC workflows emerging around quantum research with memory requirements Top Trends Increased interposer innovations linking compute dies and stacked HBM together Novel wafer-scale designs unifying CI processes with integrated HBM builds Software optimization emerging to maximize concurrency with HBM-based accelerator platforms Top Challenges Efficient heat dissipation in stacked layouts maintaining stability under stress Ensuring error-free performance across memory dies interconnected through packaging methods Balancing cost trade-offs when integrating specialized substrates for HBM arrays Cost Structures and Strategic Patents to Play Key Role in the High Bandwidth Memory Market Developing and producing advanced memory solutions requires substantial financial investment, reflected in current cost structures. HBM2E's price tag hovers around US$25 to US$35 per GB, highlighting both its specialized manufacturing process and premium performance. In line with accelerating demand and finite production capabilities, HBM prices overall are projected to increase by 5 to 10 additional units (currency not specified) in 2025. This upward trajectory is not unexpected, given the extensive resources and technical expertise required to develop these solutions. Additionally, the average selling price of HBM is at least three times that of standard DRAM, underscoring its value proposition in meeting specialized computational demands. Moreover, HBM2E costs can be approximately three to five times higher than competing technologies such as GDDR6, primarily due to more sophisticated stacking and interface technologies. Another factor shaping the landscape of the high bandwidth memory market growth is patent activity, which protects key innovations and allows companies to maintain technological edges. A noteworthy example is U.S. Patent No. 10,740,262, granted for integrated circuit I/O integrity and degradation monitoring—an essential technique relevant to high-end memory systems. This patent underscores the critical role intellectual property plays in sustaining advancement and ensuring that new performance benchmarks can be defended against competitive pressures. For businesses entering or expanding in this domain, patent portfolios serve as both shields and bargaining chips in negotiations, partnerships, and potential licensing agreements. Over time, these patents incentivize deeper research into the underlying technology, which in turn leads to fresh breakthroughs and enhancements. By carefully managing costs and investing in key intellectual properties, companies are better positioned to innovate quickly while still meeting marketplace demands. Demand Shifts and Infrastructure Growth is Shaping the Demand Outlook The growing need for sophisticated computing in data centers and AI-driven applications has fueled substantial interest in advanced memory architectures in the high bandwidth memory market. One notable projection indicates that the global market dedicated to AI servers incorporating HBM solutions is expected to reach US$15.0 billion by 2032. This figure signals a massive shift as businesses and organizations seek reliable, high-speed memory to handle increasingly complex calculations and real-time data processing. Meanwhile, the consumer segment linked to this domain is projected to grow by around 27 units (a figure indicating an incremental rise) during the forecast period from 2024 to 2029, reflecting the broadening appetite for improved performance in personal devices and gaming platforms. Key to meeting these evolving demands in the high bandwidth memory market is the underlying fabrication technology. Samsung's foundry division, for instance, is zeroing in on Gate-All-Around (GAA) transistors at the 2-nanometer node, ensuring that new memory stacks not only deliver higher performance but also maintain strong energy efficiency. Alongside this push toward more refined lithography, HBM3 harnesses a pseudo-channel architecture that effectively doubles the number of independent sub-channels, smoothing data flow and reducing bottlenecks. Beyond current benchmarks, HBM3E is expected to embrace even higher data transfer rates and better energy efficiency. These architecture-level improvements help data centers and AI platforms handle surging traffic demands and more robust algorithms. The interplay of steady improvements in lithography, coupled with strategic alignment toward AI and data-intensive applications, is reorienting the entire ecosystem toward a future defined by accelerated, streamlined processing. Ask For Customization @ Advanced Architectural Shifts and Upcoming Releases in High Bandwidth Memory Market Cutting-edge architectural innovations are shaping the roadmap for next-generation releases, with significant attention on how upcoming iterations will surpass the lofty benchmarks set by previous HBM generations. One emerging milestone is HBM3E, whose shipments are anticipated to be concentrated in the second half of 2024. Early indicators point to boosted data transfer speeds and refined power management, aimed at catering to advanced AI models and real-time analytics. Alongside these hardware improvements, ecosystem players are exploring ways to streamline the transition process for system integrators and enterprise clients, reducing adoption barriers and ensuring readiness for the next wave of high-performance workloads. As the memory sector marches forward, numerous challenges remain on the path to large-scale deployment in the high bandwidth memory market. However, major manufacturers have openly shared their intentions to fortify output and meet projected demand for next-generation memory. Samsung, for instance, has not only enhanced current production lines but also revealed plans to double its HBM output again in the year following 2024, showcasing confidence in long-term demand. SK Hynix also plans to more than double its HBM production capacity year over year, reflecting a response to the intensifying AI market that requires quick access to aggregated, high-speed data. These expansions signal real momentum toward providing the market with robust, streamlined solutions that fill critical performance gaps in enterprise and consumer-oriented environments. Going forward, the intersection of advanced architectures and methodical expansions in production will become an engine driving the entire high-performance computing ecosystem. Future Roadmaps and Industry Transformations Over the next several years, manufacturers aim to position themselves at the forefront of technological progress, further enabling data-intensive operations in AI and beyond. Micron, for example, intends to introduce Extreme Ultraviolet (EUV) processes and construct a new cleanroom facility to facilitate higher yields and more advanced chip production. By integrating EUV, the company expects to slash defects and boost the overall performance of upcoming HBM solutions. Beyond Micron, Samsung's overarching strategy to continually grow its high-performance memory portfolio remains consistent, particularly as the company leverages advanced lithography and packaging techniques to remain competitive amid rapid market shifts. The increased focus on specialized memory manufacturing inevitably impacts other product lines—most notably DRAM—by causing a reduction in overall production capacity devoted to mainstream solutions. This shift in the high bandwidth memory market can influence the availability of DDR5 modules, potentially shaping the market for consumer and enterprise applications that rely on standard memory solutions. However, the impetus behind the refocusing is clear: HBM stands to unlock unprecedented computational throughput, especially for workloads that require sustained high-speed data access. With trends pointing toward more complex AI models and a rising number of data-hungry applications, committing substantial resources to HBM manufacturing has emerged as a strategic priority for the industry. As these roadmaps unfold, the landscape is poised for transformations that will likely redefine how organizations and end-users perceive data management and processing power in the years to come. Global High Bandwidth Memory Market Major Players: Advanced Micro Devices, Inc. Samsung Electronics Co., Ltd. SK Hynix Inc. Micron Technology, Inc. Intel Corporation Xilinx Inc. Open-Silicon (SiFive) NEC Corporation Cadence Design Systems, Inc. Other Prominent Players Key Segmentation: By Product: Central Processing Unit Field-Programmable Gate Array Graphics Processing Unit Application-Specific Integrated Circuit Others By Application: High-Performance Computing (HPC) Networking and Client Space Data Centers Others By Region: North America Europe Asia Pacific Middle East & Africa (MEA) South America Inquire about this report before purchasing: About Astute Analytica Astute Analytica is a global analytics and advisory company which has built a solid reputation in a short period, thanks to the tangible outcomes we have delivered to our clients. We pride ourselves in generating unparalleled, in depth and uncannily accurate estimates and projections for our very demanding clients spread across different verticals. We have a long list of satisfied and repeat clients from a wide spectrum including technology, healthcare, chemicals, semiconductors, FMCG, and many more. These happy customers come to us from all across the Globe. They are able to make well calibrated decisions and leverage highly lucrative opportunities while surmounting the fierce challenges all because we analyze for them the complex business environment, segment wise existing and emerging possibilities, technology formations, growth estimates, and even the strategic choices available. In short, a complete package. All this is possible because we have a highly qualified, competent, and experienced team of professionals comprising of business analysts, economists, consultants, and technology experts. In our list of priorities, you-our patron-come at the top. You can be sure of best cost-effective, value-added package from us, should you decide to engage with us. Contact Us:Astute AnalyticaPhone: +1-888 429 6757 (US Toll Free); +91-0120- 4483891 (Rest of the World)For Sales Enquiries: sales@ LinkedIn | Twitter | YouTube CONTACT: Contact Us: Astute Analytica Phone: +1-888 429 6757 (US Toll Free); +91-0120- 4483891 (Rest of the World) For Sales Enquiries: sales@ Website: in to access your portfolio
