# Memory Related Articles

HTX News Center provides the latest articles and in-depth analysis on "Memory", covering market trends, project updates, tech developments, and regulatory policies in the crypto industry.

The King of Blind Date Attire in Korea: How SK Hynix Made a Comeback Against Samsung?

In South Korea's dating scene, SK Hynix employees are now highly sought after, a status shift fueled by the company's astronomical profits and employee bonuses, projected to reach up to 6.1 million RMB per person by 2027. This marks a dramatic reversal for the long-time second-place player in memory semiconductors, which has now surpassed its rival Samsung in annual operating profit. The turnaround story began in 2008 when a struggling Hynix, emerging from bankruptcy restructuring, took a risky bet by agreeing to develop High Bandwidth Memory (HBM) with AMD. At the time, HBM had no clear market beyond high-end graphics cards and was a costly, complex technology. Major players like Samsung, pursuing its own HMC technology, declined. For Hynix, with only memory as its core business, it was a gamble born of necessity. The pivotal moment came in 2012 when SK Group Chairman Chey Tae-won acquired Hynix. Defying industry downturns, he invested heavily in R&D and fabrication, sustaining the HBM project through over a decade of commercial uncertainty and internal challenges. A key break occurred around 2016-2017 when Samsung faced production issues supplying HBM2 for Google's TPU, allowing SK Hynix to gain a crucial foothold in the data center market. The AI explosion post-ChatGPT in 2022 was the catalyst, turning HBM into a critical bottleneck for AI accelerators like NVIDIA's GPUs. By 2025, SK Hynix captured 62% of the global HBM market, leaving Samsung at 17%. For the first time, its annual operating profit exceeded Samsung's. Analysts point to the "innovator's dilemma" to explain Samsung's miss: its vast, successful business portfolio made it risk-averse, preventing an all-in bet on the initially niche HBM technology. In contrast, SK Hynix, as a challenger with its back against the wall, had no choice but to commit fully. The story highlights how Korea's chaebol system allows for ultra-long-term bets beyond quarterly pressures. However, SK Hynix's lead isn't guaranteed. Samsung is aggressively catching up on HBM4, and challenges like customer concentration (heavy reliance on NVIDIA) and technical hurdles in advanced packaging remain. The narrative underscores a market truth: the greatest alpha often comes from betting on uncertain, long-term directions others dismiss, much like HBM in 2008.

marsbit05/11 11:08

The King of Blind Date Attire in Korea: How SK Hynix Made a Comeback Against Samsung?

marsbit05/11 11:08

An In-Depth Analysis of the Reasons Behind the Recent Surge in the Memory Storage Industry

"Understanding the Recent Boom in the Memory Industry" summarizes the key structural shift in the memory sector driven by AI. Traditionally, the memory industry (DRAM, NAND) has been highly cyclical, with prices driven by inventory cycles and consumer electronics demand, which can be deferred. However, AI has fundamentally changed the demand structure. AI workloads create continuous, non-deferrable demand for high-bandwidth memory (HBM) due to GPU memory bottlenecks. This high-margin HBM production is prioritized by major manufacturers like Samsung, SK hynix, and Micron, structurally reducing the supply of traditional DRAM and NAND. Consequently, the market is transitioning from a spot-based model to a contract-based allocation system. HBM supply is reportedly booked through 2026, leading to spot shortages and sharp price increases for other memory types. Delivery lead times have also extended dramatically. Pricing power is shifting from the open market to those with long-term contracts. Primary beneficiaries are HBM-leading suppliers and large hyperscalers (Microsoft, AWS, Google) that can lock in future supply. Key risks to this "AI supercycle" include a potential slowdown in AI capital expenditure, disruptive new memory architectures, or a return to aggressive industry capacity expansion, which could reintroduce cyclicality.

marsbit05/11 10:30

An In-Depth Analysis of the Reasons Behind the Recent Surge in the Memory Storage Industry

marsbit05/11 10:30

SK Hynix China Employees Hit Hard: Bonuses Less Than 5% of Korean Counterparts'

"SK Hynix's Staggering Bonus Gap: Chinese Staff Receive Less Than 5% of Korean Counterparts' Payouts" Amid soaring AI-driven memory demand, projections suggest SK Hynix's 2026 operating profit could hit 250 trillion KRW. Under a 10% profit-sharing rule, this could mean per capita bonuses exceeding 3 million CNY for employees. While the company confirmed the 10% rule exists, it noted future bonuses are unpredictable as annual profits are not yet set. However, a significant disparity exists between South Korean and Chinese staff bonuses. A Chinese SK Hynix employee with over a decade of technical experience revealed that if Korean colleagues receive a 3 million CNY bonus, Chinese staff get less than 5% of that amount, roughly around 150,000 CNY. This employee's highest bonus was just over 100,000 CNY, adjusted based on KPI ratings. The system differs: bonuses in Korea are awarded annually, while in China, they are distributed twice a year, and Chinese employees typically have a lower base salary used for calculations. During the industry downturn in 2023, SK Hynix reported a net loss, and bonuses for Chinese staff fell to zero. Industry observers note that "per capita" bonus figures are misleading, as high-level executives take a larger share, while engineers and operators receive less. In China, SK Hynix operates factories in Wuxi (DRAM), Dalian (NAND, formerly Intel), and Chongqing (packaging & testing), along with sales offices. Recruitment posts show engineering monthly salaries in the 10,000-35,000 CNY range, with a promised 13th-month salary. Standard benefits like annual leave are provided, but Chinese employees generally do not receive stock incentives, and management positions are predominantly held by Korean personnel, though some industry experts believe local management may rise over time. Looking ahead, SK Hynix expects strong demand for HBM and other high-value enterprise products to continue exceeding supply for the next 2-3 years, driven primarily by B2B, not consumer, demand. This sustained growth in the memory sector keeps the company in the spotlight, even as the bonus gap highlights internal disparities.

marsbit05/11 05:52

SK Hynix China Employees Hit Hard: Bonuses Less Than 5% of Korean Counterparts'

marsbit05/11 05:52

Your Claude Will Dream Tonight, Don't Disturb It

This article explores the recent phenomenon of AI companies increasingly using anthropomorphic language—like "thinking," "memory," "hallucination," and now "dreaming"—to describe machine learning processes. Focusing on Anthropic's newly announced "Dreaming" feature for its Claude Agent platform, the piece explains that this function is essentially an automated, offline batch processing of an agent's operational logs. It analyzes past task sessions to identify patterns, optimize future actions, and consolidate learnings into a persistent memory system, akin to a form of reinforcement learning and self-correction. The article draws parallels to similar features in other AI agent systems like Hermes Agent and OpenClaw, which also implement mechanisms for reviewing historical data, extracting reusable "skills," and strengthening long-term memory. It notes a key difference from human dreaming: these AI "dreams" still consume computational resources and user tokens. Further context is provided by discussing the technical challenges of managing AI "memory" or context, highlighting the computational expense of large context windows and innovations like Subquadratic's new model claiming drastically longer contexts. The core critique argues that this strategic use of human-centric vocabulary does more than market products; it subtly reshapes user perception. By framing algorithms with terms associated with consciousness, companies blur the line between tool and autonomous entity. This linguistic shift can influence user expectations, tolerance for errors, and even perceptions of responsibility when systems fail, potentially diverting scrutiny from the companies and engineers behind the technology. The article concludes by speculating that terms like "daydreaming" for predictive task simulation might be next, continuing this trend of embedding the idea of an "inner life" into computational processes.

marsbit05/11 00:15

Your Claude Will Dream Tonight, Don't Disturb It

marsbit05/11 00:15

The US Stock Market in 2026, It's Almost Too Easy, and That Makes Me Nervous

The U.S. stock market's performance in 2026, particularly in the semiconductor memory sector, has generated significant returns that make some investors uneasy. A popular sentiment contrasts the perceived skill required for success in China's A-shares with the apparent ease of profiting from simply holding U.S. stocks. The primary driver is a global memory chip boom. Stocks like Micron, Seagate, Western Digital, and especially SanDisk (spinning off from WDC in 2025) have skyrocketed, with some gains exceeding 500% or even 2200%. Korean giants Samsung and SK Hynix, dominating their domestic index, have also surged. This rally is fueled by an AI-driven demand surge for memory like HBM (High-Bandwidth Memory), critical for AI chips. Tech giants like Google and Microsoft are placing massive, "unpriced" orders, while analysts continuously upgrade forecasts. SK Hynix reported its 2026 HBM capacity is already sold out. Despite record profits and sky-high margins (e.g., SK Hynix's 72% operating margin), major memory manufacturers are deliberately restricting capital expenditure and capacity expansion, controlling over 90% of DRAM supply. This supply discipline sustains high prices but draws parallels to cartel behavior. The situation presents two narratives. The bullish case sees AI demand as a structural, long-term shift with a prolonged supply gap. The bearish case, exemplified by short-seller Citron's failed bet against SanDisk, warns of a classic commodity cycle where prices eventually crash rapidly, as seen historically. The irony is noted: while retail investors marvel at easy gains, insiders like Western Digital are selling SanDisk shares at a 25% discount. Ultimately, the high cost of memory in consumer devices feeds into the record profits of memory companies and the soaring stock prices, leading many to question the sustainability of a market where making money seems "as easy as breathing."

marsbit05/08 02:57

The US Stock Market in 2026, It's Almost Too Easy, and That Makes Me Nervous

marsbit05/08 02:57

a16z: AI's 'Amnesia', Can Continuous Learning Cure It?

The article "a16z: AI's 'Amnesia' – Can Continual Learning Cure It?" explores the limitations of current large language models (LLMs), which, like the protagonist in the film *Memento*, are trapped in a perpetual present—unable to form new memories after training. While methods like in-context learning (ICL), retrieval-augmented generation (RAG), and external scaffolding (e.g., chat history, prompts) provide temporary solutions, they fail to enable true internalization of new knowledge. The authors argue that compression—the core of learning during training—is halted at deployment, preventing models from generalizing, discovering novel solutions (e.g., mathematical proofs), or handling adversarial scenarios. The piece introduces *continual learning* as a critical research direction to address this, categorizing approaches into three paths: 1. **Context**: Scaling external memory via longer context windows, multi-agent systems, and smarter retrieval. 2. **Modules**: Using pluggable adapters or external memory layers for specialization without full retraining. 3. **Weights**: Enabling parameter updates through sparse training, test-time training, meta-learning, distillation, and reinforcement learning from feedback. Challenges include catastrophic forgetting, safety risks, and auditability, but overcoming these could unlock models that learn iteratively from experience. The conclusion emphasizes that while context-based methods are effective, true breakthroughs require models to compress new information into weights post-deployment, moving from mere retrieval to genuine learning.

marsbit04/25 04:23

a16z: AI's 'Amnesia', Can Continuous Learning Cure It?

marsbit04/25 04:23

A 120,000 Yuan Tombstone or 399 Yuan AI Immortality: Which Would You Choose?

"The 'Deathcare Moutai' Fushouyuan, once a highly profitable cemetery operator, has halted trading amid a severe crisis, with its net profit plummeting by 52.8% in 2024. This reflects a broader trend of people rejecting expensive traditional burials, as average grave prices in China have soared to over ¥120,000. In response, the industry is pivoting to digital alternatives, with companies like Fushouyuan offering AI-powered memorial services, such as virtual farewell halls and AI-generated recreations of the deceased. Simultaneously, a low-cost, unregulated AI 'resurrection' industry has emerged online, with services priced as low as ¥399. These often use open-source tools to create crude digital avatars from photos and voice clips, exploiting vulnerable individuals, particularly bereaved parents who have lost their only child. However, these services raise significant ethical and legal concerns, including data privacy risks and potential use in scams. Academic studies warn that such AI companions may exacerbate grief, leading to prolonged mourning disorders and emotional dependency, rather than providing genuine comfort. While regulations are being drafted to manage digital human services, the deep emotional drive to 'reconnect' with loved ones often overshadows rational concerns. Ultimately, the article questions whether digital immortality truly preserves memory or merely offers a commercialized illusion, emphasizing that no technology can replace the real, irreplaceable loss of a human life."

marsbit04/22 08:34

A 120,000 Yuan Tombstone or 399 Yuan AI Immortality: Which Would You Choose?

marsbit04/22 08:34

Open-Source Plugin Ignites Underlying AI Model Warfare: Behind Claude-mem's Explosive Popularity Lies Big AI Companies' Best-Kept Profit Secret

The open-source plugin "Claude-mem" has ignited a hidden war in the AI industry by tackling a critical weakness in large language models: their lack of memory. This tool, which exploded in popularity on GitHub, works by locally storing and compressing conversation history, slashing redundant token usage by up to 95%. This directly undercuts the "context tax"—the costly practice of repeatedly sending historical data to the cloud with each new interaction. Its integration with another tool, OpenClaw, enabled users to exploit a pricing loophole, using low-cost personal subscriptions to run high-frequency automated tasks meant for expensive enterprise API plans. In response, Anthropic banned third-party OAuth access, triggering a backlash and even a major service outage. Despite the crackdown, Claude-mem’s founder circumvented traditional monetization by launching a cryptocurrency, $CMEM, on the Solana network. The episode highlights key tensions in the AI industry: the fight over pricing models, the value of local memory control, and the risks of building on proprietary platforms. The battle over AI’s future is being waged in the code.

marsbit04/20 00:20

Open-Source Plugin Ignites Underlying AI Model Warfare: Behind Claude-mem's Explosive Popularity Lies Big AI Companies' Best-Kept Profit Secret

marsbit04/20 00:20

OpenAI Completely Overhauls Codex, Grows Its Own Mouse, Schedules Itself to Outwork Humans

OpenAI has fundamentally transformed Codex from a coding assistant into a versatile AI agent capable of autonomous computer control. The updated Codex can now operate in the background with its own independent cursor, allowing it to run applications like Xcode, test software, and fix bugs without interrupting the user’s workflow. It features a built-in browser for real-time visual editing of web pages, enabling developers to make changes through direct UI interactions. The release also introduces over 90 plugins for tools like JIRA, CircleCI, and SharePoint, integrating end-to-end development and productivity workflows. A new "heartbeat" feature allows Codex to self-schedule tasks, maintain context across sessions, and work autonomously over extended periods. Additional enhancements include memory capabilities, support for handling GitHub code reviews, multi-tab terminals, and built-in file generation for Excel, PDF, and PowerPoint. Currently, advanced computer control is available on macOS (excluding EU/UK users), with Windows support limited to data retrieval. This expansion positions Codex as a foundational step toward OpenAI’s planned "super-app," combining coding, automation, and multi-tool integration into a single platform.

marsbit04/17 11:20

OpenAI Completely Overhauls Codex, Grows Its Own Mouse, Schedules Itself to Outwork Humans

marsbit04/17 11:20

Hermes Agent Guide: Surpassing OpenClaw, Boosting Productivity by 100x

A guide to Hermes Agent, an open-source AI agent framework by Nous Research, positioned as a powerful alternative to OpenClaw. It is described as a self-evolving agent with a built-in learning loop that autonomously creates skills from experience, continuously improves them, and solidifies knowledge into reusable assets. Its core features include a memory system (storing environment info and user preferences in MEMORY.md and USER.md) and a skill system that generates structured documentation for complex tasks. The agent boasts over 40 built-in tools for web search, browser automation, vision, image generation, and text-to-speech. It supports scheduling automated tasks and can run on various infrastructures, from a $5 VPS to GPU clusters. Popular tools within its ecosystem include the Hindsight memory plugin, the Anthropic Cybersecurity Skills pack, and the mission-control dashboard for agent orchestration. Key differentiators from OpenClaw are its architecture philosophy—centered on the agent's own execution loop rather than a central controller—and its autonomous skill generation versus OpenClaw's manually written skills. Installation is a one-line command, and setup is guided. It integrates with messaging platforms like Telegram, Discord, and Slack. It's suited for scenarios requiring a persistent, context-aware assistant that improves over time, automates workflows, and operates across various deployment environments.

marsbit04/13 13:11