{"slug": "sk-hynix-chipping-away-at-ai-memory-wall-from-several-angles", "title": "SK Hynix chipping away at AI memory wall from several angles", "summary": "SK Hynix is pursuing multiple business development avenues involving fabs in China, Japan, Korea and the USA, Solidigm fund-raising, co-packaged optics, and high-bandwidth flash, as AI demand drives ever-higher memory access speeds. The company plans a memory fab in Miyagi Prefecture, Japan, with investment of tens of trillions of won, and is building an HBM packaging plant in Indiana, USA. Former SK Hynix president Roh Jong-won has set up Techbridge Investment to invest in a potential Solidigm IPO, with plans for up to ₩2 trillion ($1.44 billion) joint investment with Stonebridge, and SK Hynix has valued Solidigm at ₩50 trillion ($36 billion) with a ₩10 trillion ($7.2 billion) fundraising target.", "body_md": "FLASH\n\n# SK Hynix chipping away at AI memory wall from several angles\n\nSK Hynix is pursuing multiple business development avenues involving fabs in China, Japan, Korea and the USA, Solidigm fund-raising, co-packaged optics, and high-bandwidth flash, as AI demand drives ever-higher memory access speeds.\n\nAI Memory profitability is so high that SK Hynix can fund its way to do all this through a combination of its own revenues and private equity investors convinced that memory for AI is a long-term goldrush.\n\n###### SK Hynix and Japan\n\nThere are [reports](https://english.hani.co.kr/arti/english_edition/e_business/1274081.html ) that SK Hynix intends to build a memory fab in Miyagi Prefecture, northeastern Japan. It will involve an investment of tens of trillions of won, which seems relatively small-scale compared to its [₩700 trillion ](https://www.blocksandfiles.com/flash/2026/08/07/sk-hynix-building-new-dram-and-nand-fabs/5284568)($481 billion) Korean fab production capacity increase plans.\n\nIt is also building an HBM packaging plant in Indiana in the USA. The overall picture is that SK Hynix’ main DRAM/HBM and NAND production will be in Korea, with supplementary sites in China (Solidigm NAND fabbing at Dalian), the USA (Indiana), and Japan.\n\nMicron (via acquired Elpida) and TSMC also have semiconductor fabs in Japan.\n\n###### Solidigm funding\n\nFormer SK Hynix president Roh Jong-won (also known as Kevin Jongwon Noh) has set up Techbridge Investment, a US private equity firm, to invest in a future potential Solidigm IPO, according to Korea’s [Seoul Economic Daily](https://en.sedaily.com/signal/article/ex-sk-hynix-president-launches-fund-to-invest-in-solidigm) outlet. Techbridge plans a joint investment in Solidigm with Korean investment partner Stonebridge of up to ₩2 trillion ($1.44 billion). Roh Jong-won was the architect for SK Hynix’ $9 billion acquisition of Intel’s NAND division, which became Solidigm, in 2021. He has also been SK Hynix’s CFO and was involved in its investment in Kioxia via Bain Capital.\n\nApparently institutional investors such as Mubadala, the UAE sovereign wealth fund, have revealed interest in investing in Solidigm through lead manager Goldman Sachs. SK Hynix has, SE Daily says, valued Solidigm at ₩50 trillion $36 billion and has a ₩10 trillion ($7.2 billion) target for Solidigm fund raising.\n\nAfter SK Hynix set up Solidigm as a US business entity the company endured several years of losses but became profitable in 2025 as demand for its high-capacity QLC drives rose and prices increased substantially as well. SK Hynix is setting up a US-based [AI company](https://www.blocksandfiles.com/ai-ml/2026/01/28/sk-hynix-to-spin-solidigm-into-new-us-ai-investment-vehicle/4090437) which will own the Solidigm business and pursue other opportunities.\n\nAlthough SK Hynix has not formally announced it will pursue a Solidigm IPO, the Techbridge and Goldman Sachs activities imply it is deeply involved in fundraising for Solidigm with a Nasdaq IPO as the eventual outcome. We note that Solidigm’s NAND manufacturing operation in Dalian, China, is owned by a separate SK Hynix business entity; SK Semiconductor (Dalian) Co. Production capacity is being [increased](https://www.blocksandfiles.com/flash/2026/08/12/sk-hynix-expanding-solidigm-dalian-nand-plant/5286679) at this Dalian plant.\n\nTechbridge is also raising a private equity fund, using overseas investors’ capital, with a second office in Singapore, to invest in other target companies.\n\n###### High-bandwidth flash\n\nSK Hynix and Sandisk are developing high bandwidth flash (HBF) in which stacked NAND chips provide a higher bandwidth to flash-stored data than existing PCIe bus-based interconnects. It should have the same read bandwidth as HBM with 8 to 16 times the capacity. Assuming a 48 GB HBM3e capacity that implies a 3784 to 768 GB capacity range.\n\nSandisk revealed that it had taped out its first HBM die at its 2026 Investor Day earlier this month with HBF product samples due next year.\n\n###### Going beyond memory\n\nSK Hynix manufactures high-bandwidth Memory (HBM) which uses DRAM chips to place memory close to GPUs so that they can access data in memory faster than through the traditional x86 socket interface. However the demands on memory capacity and speed by GPUs involved in AI training and inference are going beyond what HBM, limited to its GPU shoreline connectivity, with one HBM stack per side of the oblong GPU chip, can provide.\n\nThe company is involved with co-packaged optics as a [Nature Electronics paper ](https://www.nature.com/articles/s41928-026-01681-6)(paywall) illustrates. Its abstract says “Optical compute interconnects, which replace electrical links with co-packaged photonic channels, could provide low propagation loss, high bandwidth and superior signal integrity. …We … provide a technology roadmap from two-dimensional (2D) co-packaged optics, 2.5D interposer-based integration and 3D heterogeneous stacking, identifying critical challenges in thermal management, manufacturability and standardization that will need to be addressed to establish optical compute interconnects as a foundational communication technology for high-performance computing infrastructure.”\n\nThe idea here is to provide a pool of memory connected to and sharable by a group of GPUs. Seunghoon Hong, AI Infra Team Lead at SK Hynix and paper co-author, said: “CPO integrates optical transceivers (TRx) into the same package as the processor, enabling chips to exchange data using light instead of long electrical interconnects. This fundamentally changes how data moves through AI systems, removing one of the biggest barriers to scaling compute.”\n\nThis would, we understand, provide faster memory access than CXL.\n\nProfessor Kyusang Lee, Department of Electrical and Computer Engineering, University of Virginia (UVA), and a co-author of the Nature paper; said: “The technology has already moved beyond the laboratory and entered the early stages of commercialization. Significant challenges remain, from integrating low-power photonic devices to developing coherence protocols and improving system reliability. Ultimately, however, the key lies in co-designing memory devices and controllers, photonic components and packaging as a single integrated system. This is precisely where collaboration with SK hynix is especially meaningful.”", "url": "https://wpnews.pro/news/sk-hynix-chipping-away-at-ai-memory-wall-from-several-angles", "canonical_source": "https://www.blocksandfiles.com/flash/2026/08/25/sk-hynix-chipping-away-at-ai-memory-wall-from-several-angles/5292113", "published_at": "2026-08-25 11:33:00+00:00", "updated_at": "2026-08-25 11:44:53.268664+00:00", "lang": "en", "topics": ["ai-infrastructure", "ai-chips", "ai-products"], "entities": ["SK Hynix", "Solidigm", "Techbridge Investment", "Stonebridge", "Goldman Sachs", "Mubadala", "Sandisk", "Roh Jong-won"], "alternates": {"html": "https://wpnews.pro/news/sk-hynix-chipping-away-at-ai-memory-wall-from-several-angles", "markdown": "https://wpnews.pro/news/sk-hynix-chipping-away-at-ai-memory-wall-from-several-angles.md", "text": "https://wpnews.pro/news/sk-hynix-chipping-away-at-ai-memory-wall-from-several-angles.txt", "jsonld": "https://wpnews.pro/news/sk-hynix-chipping-away-at-ai-memory-wall-from-several-angles.jsonld"}}