--- title: "The \"super windfall in storage\" after HBM - NAND! AMD decisively moves to acquire MEXT flash memory, leading the \"AI inference economics\"" type: "News" locale: "en" url: "https://longbridge.com/en/news/289902646.md" description: "AMD announced the acquisition of MEXT, which focuses on AI storage optimization, aiming to address memory bottlenecks in data centers. This move promotes NAND flash memory's upgrade from traditional capacity storage to an \"almost memory layer\" for AI inference, seen as a new trend following HBM. As a result, AMD's stock price surged nearly 7%, bringing its market value close to $890 billion; Sandisk's stock price also rose approximately 7%" datetime: "2026-06-16T10:32:59.000Z" locales: - [zh-CN](https://longbridge.com/zh-CN/news/289902646.md) - [en](https://longbridge.com/en/news/289902646.md) - [zh-HK](https://longbridge.com/zh-HK/news/289902646.md) --- # The "super windfall in storage" after HBM - NAND! AMD decisively moves to acquire MEXT flash memory, leading the "AI inference economics" According to Zhitong Finance APP, American PC and data center high-performance chip leader AMD (AMD.US) once again received strong bullish sentiment from investors on Monday, after the company announced the acquisition of MEXT—a technology company that has focused on AI-driven data center DRAM/NAND storage optimization technology for many years. AMD's stock price surged nearly 7% to $547.26 by the close of U.S. markets on Monday, with a market capitalization hovering around $890 billion. The stock price of American NAND flash leader Sandisk (SNDK.US) also rose nearly 7%, highlighting that as AMD acquires MEXT, the market increasingly recognizes that NAND flash technology is upgrading from a "storage capacity asset" to a "quasi-memory layer" in the AI computing system, likely becoming the next super trend in the storage chip field after HBM. It is understood that AMD, which rarely participates in the M&A market, completed this latest acquisition to address one of the biggest pain points in modern data centers—memory limitations are becoming increasingly prominent as AI and data-intensive workloads continue to grow exponentially. Enterprises are facing significant constraints with very limited HBM/DRAM memory access capabilities, which can slow down performance and significantly increase overall storage chip costs. NAND flash is transitioning from being traditionally classified by industry insiders as a "cold data/capacity storage" asset to a fully upgraded "extended quasi-memory layer" in the AI inference era. It is likely to become one of the most important cutting-edge technologies in the storage chip industry after the HBM super storage system. The signals conveyed by AMD's acquisition of MEXT are crucial—AMD officially emphasizes that MEXT's AI-driven storage optimization technology aims to make flash memory perform closer to DRAM, while maintaining performance and efficiency, expanding usable memory capacity, reducing infrastructure costs, and significantly enhancing the scaling efficiency of AI and general workloads. According to AMD's latest acquisition announcement, MEXT's exclusive technology helps make the performance and energy efficiency of much cheaper NAND flash type storage chips more akin to DRAM or even HBM systems. AMD stated that, simply put, this will significantly strengthen AMD's product lineup dominated by CPU+GPU+high-performance network infrastructure in the AI data center field, and the significant cost advantage of NAND flash over DRAM will determine that NAND's market share will increasingly expand in the AI inference era. AMD stated that after acquiring MEXT, it will help customers deploy massive AI training/inference workloads with higher efficiency and lower total cost of ownership (TCO). The addition of MEXT's engineering talent is also expected to support the company's future accelerated expansion needs in enterprise AI application implementation and cloud computing. In its acquisition statement, AMD noted that as AI large models are updated and iterated, and as data analysis, virtualization, and high-performance computing workloads continue to grow in scale and complexity, memory has become the most critical constraint in cloud and enterprise AI workload environments. For customers, addressing these bottlenecks is crucial for improving performance per dollar, enhancing data center resource allocation efficiency, and accelerating large-scale AI deployment; AMD is addressing this challenge through the acquisition of MEXT, a pioneer in AI technology-driven storage optimization technology The market reacted positively to this transaction. AMD's stock price surged nearly 7% by the close, continuing its strong upward trend this year; the market also gave positive bullish feedback to SanDisk, the leader in NAND flash technology based in the United States. **Why are AMD, SanDisk, SK Hynix, and Samsung Electronics accelerating the iteration of NAND flash?** The signal revealed by AMD's acquisition of MEXT is crucial, indicating that MEXT's AI-driven storage optimization technology aims to make flash memory perform closer to DRAM, while maintaining performance and efficiency, expanding usable memory capacity, reducing infrastructure costs, and significantly enhancing the scalability of AI and general workloads. MEXT positions itself as a technology path that "allows low-cost, high-capacity system flash to present to the operating system at speeds close to DRAM," claiming to provide an expansion of usable memory capacity by about 2 to 4 times. This also indicates that AMD is not solely focused on the NAND-led data center eSSD supercycle, but rather on using software, controllers, predictive scheduling, and memory/storage chip-level management to collectively integrate NAND into the AI server cluster ecosystem dominated by AMD. The underlying driving force behind this shift in data center construction from DRAM/HBM memory increasingly towards NAND flash is the rapidly expanding demand for memory capacity and cost driven by AI inference. As AI inference enters the explosive phase of long context, RAG, vector databases, multimodal caching, and KV caching, the most scarce resources are not just bandwidth and the AI GPU clusters dominated by NVIDIA/AMD or the TPU AI computing power clusters led by Google, but rather high-capacity, low-cost, low-power scalable memory pools. However, the supply elasticity, cost, and power consumption of DRAM/HBM cannot solely meet all demands. According to TrendForce data, traditional DRAM contract prices are expected to rise sharply by about 93% to 98% quarter-on-quarter in Q1 2026, driving a significant quarter-on-quarter revenue increase of 81% to $97 billion for the DRAM industry; TrendForce also indicates that NAND flash prices are on an upward trajectory, but the overall increase and price base since the storage supercycle began in 2025 are still far behind DRAM/HBM memory systems. TrendForce predicts that NAND Flash prices will rise by 55% to 60% quarter-on-quarter, mainly driven by enterprise SSD order demand from North American cloud service providers. More significantly, Citrini Research estimates that the cost per bit of NAND flash is about 1/55 of DRAM—QLC NAND is approximately $0.05 per GB, DDR5 DRAM is about $2.75, and HBM3E is as high as $15. This price difference is exploitable in AI inference, where the largest single memory consumption—KV caching (which records all previously tagged contexts in each generation step of the model, potentially growing to hundreds of GB in long conversations)—has much lower speed requirements for reading than the decoding path of model weights. For such sequential read data, the speed advantage of DRAM is significantly narrowed, while the capacity advantage of flash is fully realized The significant emergence of the HBF technology route built on NAND flash memory (i.e., "High Bandwidth Flash") further strengthens this judgment. Sandisk, SK Hynix, and Samsung clearly define High Bandwidth Flash (HBF) as a new type of NAND form aimed at addressing the "memory wall" in AI, with the goal of providing larger capacity in AI inference and claiming that HBF can achieve performance close to "infinite capacity HBM" in relevant inference tests, while significantly enhancing usable memory capacity. SK Hynix and Sandisk have already accelerated the standardization of HBF together, positioning it as a new high-speed flash standard for AI inference servers in storage industry reports. The academic community is also quickly following suit; for example, the HAVEN paper proposes using HBF as a supplement to HBM within GPU packaging for large-scale vector databases and RAG retrieval scenarios. Modeling results show that compared to GPU-DRAM and GPU-SSD systems, reordering throughput can be improved by up to 20 times, and latency can be improved by as much as 40 times. This means that NAND is no longer just "storage outside the server," but may enter GPU packaging, near-memory computing, vector retrieval, long-context inference, and other key AI paths. However, it is worth noting that NAND/HBF will not immediately replace HBM, nor will it replace DRAM; instead, it will become a new third-tier core asset in the AI memory system. That is, NAND will not immediately replace HBM but will serve as a new supplementary layer of "high capacity, low cost, and relatively high bandwidth," handling massive data access scenarios such as AI inference, RAG, vector databases, long context, and multimodal caching. HBM will still be responsible for the highest bandwidth and lowest latency in training and high-end inference main paths; DRAM will still handle general memory and hot data caching; NAND/HBF/enterprise SSDs will be more suitable for accommodating large-capacity model weights, KV caches, low-frequency access data, edge model parameters, and RAG vector libraries. Apple's "LLM in a Flash" test has already demonstrated that through methods such as windowing and row-column binding, models exceeding the available DRAM capacity can run on DRAM-constrained devices, achieving inference speed improvements of 4-5 times on CPUs and 20-25 times on GPUs compared to naive loading methods. **x86 chip giant AMD rushes into the wild bull market** Undoubtedly, HBM remains the crown jewel of storage systems in the AI training era, while NAND/HBF may become the "capacity + cost + energy efficiency" crown in the AI inference era. Meanwhile, Sandisk, SK Hynix, Samsung, Micron, Kioxia, and a host of enterprise SSD controller, CXL/PCIe interconnect storage companies, as well as AI system platform companies like AMD and NVIDIA, are initiating a new round of value reassessment in the storage industry chain around the "AI memory wall" and "NAND/HBF." As mentioned above, the core of AMD's acquisition of MEXT is not to enter the NAND manufacturing field but to fill the capability gap of "making flash memory behave more like DRAM" at the software and AI computing system layer. AMD officially states that MEXT's AI-driven DRAM/NAND storage optimization technology can make NAND flash perform closer to DRAM, expanding usable memory capacity, reducing infrastructure costs, and improving the scalability efficiency of AI and general workloads On Wall Street, top financial institution Barclays has set a target price of $665 for AMD (a significant increase from the previous $500), corresponding to about 21.5% potential upside from the current record-high stock price, implying a market value of approximately $1.10 trillion for AMD. ![image.png](https://imageproxy.pbkrs.com/https://img.zhitongcaijing.com/image/20260616/1781605329206052.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg) Barclays' core bullish logic is not simply about "AI GPU market share catching up to NVIDIA," but rather that the wave of agent-based AI (i.e., the wave of AI agents) will drive a comprehensive explosion in CPU demand, narrowing the ratio of CPUs to GPUs. With its data center server-level CPUs, AI accelerators, and cloud customer base, AMD is expected to become one of the most important beneficiaries of AI computing infrastructure, aside from NVIDIA. Barclays' analyst team also believes that scheduling, tool invocation, request routing, and multi-step workflows in agent-based AI tasks will significantly elevate the market space for AMD server CPUs, which hold an important position in the x86 ecosystem. Another positive outlook comes from Citigroup, which has a bullish target price of $575, primarily betting on AMD becoming a more reliable second core supply source for AI GPU architecture beyond NVIDIA, potentially gaining a larger share from major clients like Meta ### Related Stocks - [SNDK.US](https://longbridge.com/en/quote/SNDK.US.md) - [AMD.US](https://longbridge.com/en/quote/AMD.US.md) - [AMDD.US](https://longbridge.com/en/quote/AMDD.US.md) - [AMDL.US](https://longbridge.com/en/quote/AMDL.US.md) - [AMUU.US](https://longbridge.com/en/quote/AMUU.US.md) - [SNDG.US](https://longbridge.com/en/quote/SNDG.US.md) - [SNDQ.US](https://longbridge.com/en/quote/SNDQ.US.md) - [SNDU.US](https://longbridge.com/en/quote/SNDU.US.md) - [SNXX.US](https://longbridge.com/en/quote/SNXX.US.md) - [AMDG.US](https://longbridge.com/en/quote/AMDG.US.md) - [DAMD.US](https://longbridge.com/en/quote/DAMD.US.md) - [AMDY.US](https://longbridge.com/en/quote/AMDY.US.md) - [AMDW.US](https://longbridge.com/en/quote/AMDW.US.md) - [SMH.US](https://longbridge.com/en/quote/SMH.US.md) - [SOXX.US](https://longbridge.com/en/quote/SOXX.US.md) - [SOXL.US](https://longbridge.com/en/quote/SOXL.US.md) - [SOXQ.US](https://longbridge.com/en/quote/SOXQ.US.md) - [XSD.US](https://longbridge.com/en/quote/XSD.US.md) - [PSI.US](https://longbridge.com/en/quote/PSI.US.md) - [FTXL.US](https://longbridge.com/en/quote/FTXL.US.md) - [SSNGY.US](https://longbridge.com/en/quote/SSNGY.US.md) - [MU.US](https://longbridge.com/en/quote/MU.US.md) - [NVDA.US](https://longbridge.com/en/quote/NVDA.US.md) - [GOOGL.US](https://longbridge.com/en/quote/GOOGL.US.md) - [GOOG.US](https://longbridge.com/en/quote/GOOG.US.md) - [AAPL.US](https://longbridge.com/en/quote/AAPL.US.md) - [BCS.US](https://longbridge.com/en/quote/BCS.US.md) - [C.US](https://longbridge.com/en/quote/C.US.md) - [META.US](https://longbridge.com/en/quote/META.US.md) - [SNDKV.US](https://longbridge.com/en/quote/SNDKV.US.md) - [SMSN.UK](https://longbridge.com/en/quote/SMSN.UK.md) - [NVD.DE](https://longbridge.com/en/quote/NVD.DE.md) - [BARC.UK](https://longbridge.com/en/quote/BARC.UK.md) - [C-R.US](https://longbridge.com/en/quote/C-R.US.md) ## Related News & Research - [Alchemi Wealth LLC Makes New Investment in Advanced Micro Devices, Inc. $AMD](https://longbridge.com/en/news/289657754.md) - [BUZZ-Rackspace up, cuts workforce and signs data center deal with AMD](https://longbridge.com/en/news/289967854.md) - [Zinnia Wealth Advisory LLC Buys Shares of 3,283 Advanced Micro Devices, Inc. $AMD](https://longbridge.com/en/news/289772638.md) - [United Capital Management of KS Inc. Buys Shares of 1,413 Advanced Micro Devices, Inc. $AMD](https://longbridge.com/en/news/289660566.md) - [AMD Commits up to £2 Billion to Accelerate AI Innovation and Research in the United Kingdom | AMD Stock News](https://longbridge.com/en/news/289024339.md)