Phison Ekectronics

The share price of Phison Electronics (Taiwan) surged from 500 to amost 900 in the past six months. Any thoughts why?

Several factors drove the Phison Electronics stock surge in 2025, primarily tied to the massive growth in the artificial intelligence (AI) sector, which created unprecedented demand for NAND storage solutions

. Key drivers included Phison's leading position in the enterprise market with its new high-performance products and its record-breaking financial performance. 

Impact of AI on NAND flash market

A combination of surging AI-driven demand and underinvestment in NAND flash capacity has placed the memory market on a path toward a multi-year "supercycle". Phison's CEO has even warned that the NAND flash shortage could potentially last for a decade. 

• AI inference boom: Large-scale AI inference workloads require vast amounts of high-speed, NAND-based storage.
• Data center storage: A data center AI infrastructure buildout requires massive amounts of storage, which Phison and its partners, like Supermicro, are providing.
• Limited supply and higher costs: Phison's CEO noted limited investment in flash capacity, leading to constrained supply and higher prices, which directly benefits storage component suppliers. 

New products and enterprise dominance

Phison cemented its role as a leader in high-performance storage by releasing new, cutting-edge products, especially for the enterprise sector. 

• PCIe 5.0 SSD controller: In early 2025, Phison introduced its next-gen PCIe 5.0 SSD controllers (like the PS5028-E28).
• Collaborations: Phison is a dominant supplier to major PC manufacturers and continues to strengthen partnerships to address enterprise needs. For example, a collaboration with Supermicro addresses the need for high-density storage in AI and hyperscale workloads.
• 128TB enterprise SSD: In early 2025, Phison showcased the world's first 128TB enterprise SSD, demonstrating its innovation and capability in the data center market. 

Strong financial performance

Phison's revenue and earnings reports throughout 2025 exceeded expectations and reflected its market growth.

• Record-high revenue: Phison reported record-high quarterly revenues in the second quarter of 2025, demonstrating strong demand for its products.
• Controller revenue surge: The company saw a 90% quarter-over-quarter and 49% year-over-year surge in client SSD controller revenue.
• Increased market share: Phison has become the largest supplier in the mobile storage controller market. 

AI integration and innovation

Phison is embedding AI and machine learning into its own technology to drive innovation and support the market.

• aiDAPTIV+ technology: Phison's aiDAPTIV+ integrates AI and machine learning directly into its NAND controllers, enhancing the performance and reliability of its storage solutions for advanced AI applications.
• Addressing the AI market: The company is focused on expanding access to AI by offering cost-effective solutions for smaller organizations. 

Memory

 

I assume personal computing (desktop, laptop) will contine to be an important component in human lives. The question is this: will chatbots drive a decline in the need for "edge" memory (SSD, HDD) or will the need for such memory actually increase?

AI Overview

Chatbots will likely increase, not decrease, the need for "edge" memory (local SSDs and HDDs) in personal computing

. While most of a large language model's (LLM) training is performed in the cloud, running these models locally for real-time inference and fine-tuning has significant advantages for users. 

The shift toward local AI will increase storage demand

1. Privacy and control: When a chatbot runs entirely on your device, your data never leaves your immediate environment. Sensitive information, such as financial records or personal conversations, is processed and stored securely on your hardware. This eliminates the risk of it being transmitted to and kept on third-party servers. 

2. Faster performance: Local processing eliminates the latency associated with communicating with cloud-based servers. For tasks that require immediate, real-time responses—such as voice control or advanced, in-app functionality—having the AI model on your local machine is far more efficient. 

3. Offline functionality: A local AI model continues to function even without an internet connection. This is crucial for users in remote locations or those who experience network outages, ensuring uninterrupted access to AI features. 

4. Data ownership: Running a local AI empowers you, the user, to control what your AI remembers and for how long. The future of AI memory is heading toward user-curated memory, where you can explicitly approve or delete information the AI retains. 

5. Personalization: A local AI assistant can develop a deeper, more private understanding of your work patterns and preferences without sharing this information with external servers. This leads to a more personalized and effective experience that compounds over time. 

6. Edge AI data processing: Increasingly, AI is being deployed at the "edge" (on the user's device) to handle real-time data from IoT devices, wearable tech, and other sensors. This processing requires fast, reliable local storage, with SSDs playing a crucial role. 

Local AI requires significant storage

The core models that power chatbots are substantial in size. As of early 2025, even with advanced compression techniques, large language models can consume many gigabytes of storage. The trend is toward larger, more capable models, which will likely increase this footprint over time. 

For example, hardware provider Puget Systems recommends high-capacity NVMe SSDs (2-8TB) for users running LLMs on personal servers, specifically citing the large storage needs of model parameters and datasets. Manufacturers are already developing specialized high-performance SSDs to meet the demands of these AI workloads. 

The overall trend shows a future where the data generated and used by AI is stored across a hybrid system:

• Local, high-speed storage (SSD): For real-time inference, user context, and frequently accessed data needed by the AI.
• Centralized cloud storage: For massive-scale AI model training and vast, less frequently accessed datasets.
• High-capacity, low-cost storage (HDD): For long-term archiving of training data and other large data volumes. 

Apple And The Cloud

 AI prompt: with whom has Apple partnered for cloud-based AI support?"

What Is "The Cloud"

AI prompt: 

Have large data centers (LDCs) become "the cloud" for all practical purposes?

Answer: 

 

What's The Supply Chain For TPUs

October 25, 2025: 

IA prompt: 

I assume Google designs, markets, sells their TPUs, but who actually does the manufacturing -- the fab?

Answer: 

 

Radio Waves Captured By Rocks / Crystals -- October 20, 2025

AI prompt: why can some rocks / crystals receive radio waves?

Some rocks and crystals can act as simple radio receivers because certain minerals, when brought into contact with a wire, function as semiconductors. This principle was famously used in early "crystal radio" sets, where a piece of crystalline mineral, such as galena, served as the core component for detecting and demodulating radio waves 

 How crystals detect radio waves

For a crystal to detect a radio signal, it must be integrated into a larger radio circuit. Here is a simplified version of the process: 

1. Antenna: A long wire antenna is used to capture radio waves and convert their energy into a weak alternating current (AC).
2. Tuner: A tuning circuit, often a coil and capacitor, is used to select the desired radio frequency from the mixture of signals picked up by the antenna.
3. Detector: The electrical signal is then passed through the mineral crystal, which acts as a diode. A diode is an electronic component that allows current to flow in one direction but blocks it in the other.
4. Demodulation: A radio wave is a high-frequency carrier wave with an audio-frequency signal layered on top of it. The crystal's one-way conductivity rectifies the signal by cutting off half of the AC signal, leaving only the audio component.
5. Headphones: The weak pulsating direct current (DC) that remains contains the audio signal. It is sent to a pair of high-impedance earphones, which are sensitive enough to convert this small amount of electrical energy into audible sound. 

The role of semiconductors

Semiconductors are materials with electrical conductivity between that of a metal conductor and an insulator. This property is crucial to the crystal radio, and certain mineral crystals exhibit this behavior naturally due to trace impurities in their crystalline structure. 

Impurities create junctions: A pure crystal might not be a semiconductor, but trace metal impurities can create an excess or deficiency of electrons in its crystal lattice.

The "cat's whisker": To take advantage of this, early radio operators used a fine, pointed wire known as a "cat's whisker" to touch the crystal's surface. By adjusting the point of contact, they could find a "sweet spot" with the best rectifying properties.

Efficiency: Some natural crystals, like galena (lead sulfide), have a low band gap, making them highly efficient at converting the radio signal into sound. Galena was the most common mineral used in these radios for this reason. 

Beyond crystal radios

While crystal radios are the most common example of this phenomenon, other crystals have distinct electrical properties that interact with radio waves. 

Quartz crystals: Quartz crystals exhibit a property called piezoelectricity, meaning they can generate a voltage when compressed. More importantly, they will vibrate at an extremely precise frequency when an electric field is applied. This property makes them useful for stabilizing frequencies in radio transmitters and receivers, acting as filters to isolate specific radio channels.

Modern diodes: The finicky nature of "cat's whisker" detectors led to the development of sealed, consistent components. Modern semiconductor diodes, which function on the same principle, are now used in crystal radios built by hobbyists. Germanium diodes are often preferred over silicon because their lower voltage drop allows them to detect weaker signals

Reading Suggestions

Below are 7 recommended books — a short guide to what each does best and why you’ll enjoy it if you liked Hodges.

---

Top picks (concise guide)

1) 

Turing’s Cathedral — George Dyson (2012)

Why read it: Dyson writes with literary sweep and deep archival research about the Princeton / von Neumann circle and the birth of the digital-computing paradigm. It situates ENIAC/EDVAC/early mainframe ideas inside the intellectual program that produced them (von Neumann, John Wheeler, Oppenheimer’s circle). For a Hodges-style blend of ideas, people and consequences, this is a top recommendation. 

2) 

The Innovators — Walter Isaacson (2014)

Why read it: Isaacson’s panoramic, vividly readable history links Ada Lovelace → Babbage → Turing → von Neumann → ENIAC → modern computing. It’s excellent for seeing the web of conversations and institutional links (Moore School, Harvard, Bell Labs) that turned thought experiments into machines. Isaacson is superb at crafting human portraits and connective storytelling. 

3) 

ENIAC: The Triumphs and Tragedies of the World’s First Computer — Scott McCartney (2016)

Why read it: Focused on the ENIAC story and the personalities (Eckert, Mauchly, the Moore School), McCartney tells the drama, disputes, and engineering feats in a lively narrative. Good on human conflict, legal battles, and the machine’s wartime origins. (More journalistic than technical — ideal if you want the human drama.) 

4) 

ENIAC in Action: Making and Remaking the Modern Computer — Thomas Haigh, Mark Priestly & Crispin Rope (2019)

Why read it: A scholarly yet readable deep dive into ENIAC’s engineering, its re-use, and how early computing practice evolved. Haigh’s work is excellent when you want technical clarity without dry academic prose — especially useful if you want to understand how the machine actually worked and how it shaped later practice. 

5) 

The Man Who Invented the Computer — Jane Smiley (2017)

Why read it: Focuses on John Atanasoff and the ABC (Atanasoff-Berry Computer), illuminating an often overlooked strand of the pre-ENIAC story and the patent controversies that followed. Good storytelling and useful corrective history. 

6) 

Engines of the Mind: The Evolution of the Computer — Joel Shurkin (1996)

Why read it: Presents a readable history spanning early electromechanical machines through ENIAC and into mainframes. Strong on people and context; a good bridge book if you want steady narration across decades. 

7) 

John von Neumann and the Origins of Modern Computing — William Aspray (1990)

Why read it: If you want a rigorous, well-researched account of von Neumann’s role (architecture, stored-program concept, influence on ENIAC/EDVAC), Aspray is one of the best academic treatments that still reads well for a general reader. Use this to deepen your understanding of the intellectual pivot from machine to architecture. 

---

Reading order I’d recommend

1. The Innovators (Isaacson) — quick panoramic orientation.

2. Turing’s Cathedral (Dyson) — intellectual/Princeton context (von Neumann).

3. ENIAC: The Triumphs and Tragedies (McCartney) — human drama at the Moore School.

4. ENIAC in Action (Haigh et al.) — technical & historiographic depth.

5. Atanasoff / von Neumann books (Smiley, Aspray) — fill in origin disputes and architecture theory.

   This sequence gives the grand sweep first, then the personalities, then the machine mechanics and archival nuance.