# 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.

From 'Cash Incinerator' to 'Money Printing Machine': ChangXin Technology's Remarkable Turnaround, Raking in 50 Billion in Half a Year

Changxin Technology: From "Money Incinerator" to "Money Printer" in Six Months Changxin Technology, a Chinese DRAM chipmaker once dubbed a "money incinerator" for years of massive losses, has staged a staggering financial turnaround. Its updated IPO prospectus reveals explosive 2026 first-half results: revenue forecast of 110-120 billion yuan (up 613-677% year-on-year) and net profit of 50-57 billion yuan (up 2244-2544% year-on-year). This half-year profit rivals that of major state-owned energy giants. The reversal stems from a historic memory chip super-cycle fueled by AI. Massive demand from AI servers, consuming 8-10x more DRAM than traditional servers, coupled with a supply crunch as major players shift capacity to premium HBM, has driven DRAM prices to multi-year highs. As China's only large-scale DRAM IDM (integrated design and manufacturing) firm, Changxin was positioned to capitalize. With upgraded product lines (DDR5/LPDDR5) and high capacity utilization, it achieved both volume and price increases, doubling its global market share to 7.67% in just half a year. This follows a decade of heavy investment and losses totaling 36.65 billion yuan, a gamble led by Chairman Zhu Yiming, who famously vowed to take no salary until the company was profitable. The IPO aims to raise 29.5 billion yuan, implying a valuation that some analysts project could reach 1-2 trillion yuan long-term. Debate persists over the sustainability of profits given DRAM's cyclicality, but supporters point to structurally sustained AI demand and Changxin's strategic national importance. The story is a textbook financial comeback, rewarding persistent investment in a critical industry.

marsbit05/18 13:04

From 'Cash Incinerator' to 'Money Printing Machine': ChangXin Technology's Remarkable Turnaround, Raking in 50 Billion in Half a Year

marsbit05/18 13:04

Topping GitHub's Trending, the Essential Guide for Claude Code Users

The CLAUDE.md file, trending on GitHub, is a project-level guide for Claude Code designed to dramatically improve its accuracy and efficiency. It addresses key issues like repetitive context explanations, unauthorized code changes, and forgotten decisions across sessions. By placing this plain-text file in a project root, Claude Code reads it automatically at the start of each session. The guide includes rules to eliminate redundant explanations, enforce strict behavioral constraints (e.g., no modifications outside the requested scope without confirmation), and establish a "memory" system using companion files like MEMORY.md and ERRORS.md to log past decisions and failures. It also locks in the project's specific tech stack to prevent inappropriate tool recommendations. Highlighted are four foundational rules from Andrej Karpathy that reportedly increased coding accuracy from 65% to 94%: always ask for clarity first, implement the simplest solution, never touch unrelated code, and explicitly flag uncertainties. The article quantifies significant weekly cost savings for developers and teams by eliminating wasted time on re-explaining context, rolling back unauthorized edits, and re-evaluating previously rejected solutions. The core message is that a small, upfront investment in creating a CLAUDE.md file leads to a more predictable, controlled, and cost-effective AI programming assistant.

marsbit05/18 09:38

Topping GitHub's Trending, the Essential Guide for Claude Code Users

marsbit05/18 09:38

The AI Investment Landscape Is Being Reshaped: Beyond the 'Magnificent Seven', What Opportunities Lie in the Semiconductor Supply Chain?

AI Investment Map is Reshaping: Opportunities Beyond the 'Magnificent Seven' Since ChatGPT ignited the AI wave, investment initially focused on the "Magnificent Seven" tech giants dominating cloud infrastructure. However, the rise of DeepSeek and debates on AI capital expenditure effectiveness are shifting this dynamic. Investors now recognize opportunities deeper in the supply chain—the companies providing the essential "picks and shovels." Early concerns about an AI investment "arms race" and potential low returns were partly alleviated by strong Q1 earnings from cloud providers, validating robust compute demand. This has highlighted a more certain investment thesis: regardless of which AI applications ultimately win, massive capital expenditure will first fuel demand for semiconductors and related components. This "pick-and-shovel" logic has driven semiconductor ETFs to record highs. Key beneficiaries include: * **Memory Chipmakers (e.g., SK Hynix, Samsung, Micron)**: High Bandwidth Memory (HBM) is a critical bottleneck for AI training. * **Photonics Companies**: Crucial for high-speed data transfer within AI data centers. * **The Broader "AI-11" Semiconductor Ecosystem**: This encompasses foundries & lithography (TSMC, ASML), logic & custom chips (AMD, Broadcom, Intel, Marvell), and enterprise storage (SanDisk, Western Digital). Every dollar of AI infrastructure spending flows through this chain. While the "Magnificent Seven" remain dominant in market size, their earnings growth premium over the rest of the S&P 500 ("S&P 493") is narrowing. Market attention and marginal investment are shifting towards the expanding semiconductor supply chain. The investment narrative is evolving from "betting on the ultimate AI winner" to "investing in the certainty of the infrastructure build-out." Understanding this shift from the demand side to the supply side is key to identifying future AI investment opportunities.

marsbit05/12 08:06

The AI Investment Landscape Is Being Reshaped: Beyond the 'Magnificent Seven', What Opportunities Lie in the Semiconductor Supply Chain?

marsbit05/12 08:06

Morgan Stanley 2026 Semiconductor Report: Buy Packaging, Buy Testing, Buy China Chips, Avoid Traditional Tracks

Morgan Stanley 2026 Semiconductor Report: Buy Packaging, Buy Testing, Buy Chinese Chips; Avoid Traditional Segments. The core theme is the shift in AI compute supply from NVIDIA dominance to a three-track system of GPU + ASIC + China-local chips. The key opportunity is capturing share in this expansion, while non-AI semiconductors face marginalization due to resource reallocation to AI. Key investment conclusions, in order of priority: 1. **Advanced Packaging (CoWoS/SoIC) - Highest Conviction**: TSMC is the primary beneficiary of explosive demand, driven by massive cloud capex. Its pricing power and AI revenue share are rising significantly. 2. **Test Equipment - Undervalued & High-Growth Certainty**: Chip complexity is causing test times to double generationally, structurally driving handler/socket/probe card demand. Companies like Hon Hai Precision (Foxconn), WinWay, and MPI offer compelling value. 3. **China AI Chips (GPU/ASIC) - Long-Term Irreversible Trend**: Export controls are accelerating domestic substitution. Companies like Cambricon, with firm customer orders and SMIC's 7nm capacity support, are positioned to benefit from lower TCO (30-60% vs NVIDIA) and growing local cloud demand. 4. **Avoid Non-AI Semiconductors (Consumer/Auto/Industrial)**: These segments face a weak, structurally hindered recovery due to AI's resource "crowding-out" effect on capacity and supply chains. 5. **Memory - Severe Internal Divergence**: Strongly favor HBM (Hynix primary beneficiary) and NOR Flash (Macronix). Be cautious on interpreting price rises in DDR4/NAND as true demand recovery. The report emphasizes a 2026-2027 time window, stating the AI capital expenditure cycle is far from over. Key macro variables include persistent export controls and AI's systemic "crowding-out" effect on traditional semiconductor supply chains.

marsbit05/12 01:30

Morgan Stanley 2026 Semiconductor Report: Buy Packaging, Buy Testing, Buy China Chips, Avoid Traditional Tracks

marsbit05/12 01:30

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?