#Shezhen-made# movie #DearYou# is now in overseas theaters. Yet the real "Musheng" tale in Bangkok's #Chinatown# is more gripping. Watch our exclusive video! #OverseasChineseStories#

Madison Huang, Senior Director of Product and Technical Marketing for #NVIDIA#'s Physical AI Platform and daughter of NVIDIA CEO Jensen Huang, experienced a #Shenzhen-made# robot at the World Robot Conference 2026 in Beijing on August 19. The robot was developed by Dobot, a Shenzhen-based robotics company and a leading player in the global collaborative robotics industry. #WRC2026# #Robotics# #HumanoidRobots# #Guangdong# #ChinaTech#

#Manufacture# | 🤓🌏 Shenzhen-based Langzhiyin Electronic has sold more than 1 million smart glasses to over 80 countries and regions, with its products ranking among Amazon’s best sellers. Now, the company is expanding into U.S. retail, with an opening order from Target and other chains covering more than 1,500 stores. Founded in 2005, Langzhiyin is now bringing AI features, including Qwen and ChatGPT integration and translation in 165 languages, to its smart glasses. 🚀 To learn more: #Shenzhen# #Longgang# #SmartGlasses# #WearableTech# #AI# #TechInnovation# #MadeInChina# #GlobalMarkets#

visited a few robotics companies + factories in shenzhen last month. a few very unscientific takeaways after seeing things irl: -humanoids are exciting, but still feel early. maybe 5–10 years before we see real mass adoption. a lot of what’s impressive today is still very demo-driven. -near term, i’m probably more bullish on boring robots. robots that are designed to do one/few job(s) really well or special uses cases will get adopted much faster. hotel delivery robots are already everywhere in china and just…work. -data is still a huge bottleneck. but something i found interesting: most companies i talked to don’t really want to buy third party data. they’d rather collect their own because of privacy, quality, and how specific the data needs to be for each use case. -the biggest “oh shit” moment for me was probably the manufacturing side. in shenzhen, you can prototype something, change a component, find another supplier, test it and manufacture it all within this insanely dense ecosystem. it made me wonder: if robotics really becomes the next massive hardware platform, how does the us compete with china on manufacturing and speed of iteration? no strong conclusion yet. still learning. but seeing the physical supply chain in person definitely changed how i think about the robotics race.

I will be releasing a new interface this month. Solo founder doing hardware for 1st time. 12 months in Shenzhen. Team of 2.5 people. <$400k raised. Since I've operated so lean, made some decisions on our pricing & open source that should make lots of ppl happy. Super excited!

#Weather# | ⛈️🚆 Typhoon Maysak Alert! Typhoon Maysak made landfall in Hainan on July 3, triggering severe weather alerts across #Guangdong# through July 5. Expect widespread torrential rainstorms, particularly in the PRD, #Shaoguan#, and #Qingyuan#. Major transit suspensions are in effect: all Hainan Circular High-Speed trains and Qiongzhou Strait ferries are halted, with service cuts hitting the Guangzhou-Zhanjiang and Shenzhen-Zhanjiang lines. Check the 12306 app before traveling. Be careful: #TyphoonMaysak# #GuangdongWeather# #TravelAlert# #China#

A week is a useful unit. Here is what one looked like: Monday 7 to Friday 11 September, on one desk covering the AI supply chain. What went in first. IQE's half-year results presentation, read as a deck rather than as a press summary. Oracle's FQ1 and Adobe's FQ3 out of the 8-K exhibits the night they filed, with Oracle's own fifteen slides alongside. Korean memory inventory, Taiwan trade data, US CPI and PPI, each from the body that issued it. A broker's two-volume optical networking report, read for what its own charts said rather than for its conclusion. Infineon on the power shortage, in its own materials. Two items on that list did not exist before we made them. Jensen Huang's session at Goldman, 2,256 seconds and 371 timestamped segments, and Lumentum at the Citi conference. We pulled the audio and transcribed both. And three days on the floor at the Shenzhen optoelectronics show, behind an archive of 3,182 exhibitors, 604 agenda rows and 54 conference pages that we counted before the doors opened. What came out: 555 files across the five days. 266 figures, 162 post drafts in two languages, 40 news and data notes, and twelve primary documents pulled in from the issuers. 77 pieces were written in English, each with a Chinese counterpart. Not all were published, because a week produces more drafts than it has slots. None of that is the interesting part. Four of the five days produced roughly the same amount of work, so volume tells you almost nothing. The number I care about is four, which is how many times that week the collection got wider. The news triage learned simplified-Chinese optical vocabulary and picked up the show's own conference feed. A technology-maturity desk went in with thirteen standing sources and nineteen technology nodes, swept weekly and compared word for word against the previous copy. Open-access research became its own leg. And a financial layer that reads straight from SEC XBRL went live, alongside a bank for material that has no slot yet. Those four are still running. The 555 files are not. They were read once. Almost nothing above is exclusive. Every document was published by whoever made it, and anyone could have read it. The only difference is that it was read. $IQE $ORCL $ADBE $NVDA $LITE

$285A KEY READ-THROUGHS FROM KIOXIA HOLDINGS’ FY2026 Q2 EARNINGS CALL Kioxia’s October 30, 2026 call reinforces a widening divergence between upstream memory producers and downstream hardware buyers. Enterprise storage demand remains strong enough to support exceptional NAND profitability and extend shipment visibility into 2027 and 2028, but consumer hardware is already experiencing demand destruction from higher component costs. The most actionable conclusion is that slowing NAND price inflation does not provide immediate relief to customers: prices can stop accelerating while remaining sufficiently elevated to pressure device margins, purchasing budgets, and replacement cycles. For upstream suppliers, the principal risk has shifted from insufficient demand toward expectations that extrapolate extraordinary pricing increases. Kioxia’s approximately 14% December-quarter revenue guidance shortfall versus consensus is therefore important, but shipment timing, currency, and application mix prevent its mechanical transfer to competitors. The highest-conviction implications favor differentiated enterprise-memory and process-equipment exposure over undifferentiated “AI hardware” exposure. Magnitude assessments below refer to operating exposure, not predicted share-price movements. Country labels identify the listing market. NAND MEMORY: SANDISK BENEFITS FROM SHARED TECHNOLOGY AND SCARCITY, BUT NOT IDENTICAL QUARTERLY ECONOMICS (READ-THROUGH 1) Affected company: Sandisk Corporation (SNDK: United States). Directional impact and magnitude: Strongly positive for medium-term earnings durability; negative for expectations of uninterrupted, exceptional quarterly pricing acceleration. Fundamental exposure is high because Sandisk is directly concentrated in flash memory and shares manufacturing and technology development with Kioxia. Kioxia reported mid-20% sequential blended ASP growth, low-30% like-for-like ASP growth, and low-teens bit shipment growth. Management stated, “CY27 is sold out, and CY28, 50% plus is pretty much decided already.” It also emphasized that there were “no signs of a price decrease,” while acknowledging that maintaining the earlier pace of increases would be difficult. These statements support continued scarcity economics but not another automatic step-change in pricing. The direct transmission mechanism is stronger than a generic peer correlation. Sandisk participates in the Japanese manufacturing partnership underlying the BiCS technology platform. Successful BiCS8 production and subsequent technology transitions support its access to competitive NAND, while sustained industry tightness supports realized prices and manufacturing utilization. However, Kioxia’s customer allocations, SSD qualification schedules, product mix, and contracts do not become Sandisk orders or revenue. Shared manufacturing does not imply identical finished-product shipment timing. Sandisk’s October 29 results provide an important counterweight to an indiscriminate negative interpretation of Kioxia’s guidance. Sandisk reported USD 11.031 billion of quarterly revenue and guided the following quarter to USD 12.2 billion to USD 12.6 billion. The midpoint implies approximately 12.4% sequential growth, materially stronger than Kioxia’s approximately 0.8% yen-denominated revenue guidance. Sandisk also reported 61% sequential growth in data-center revenue. These different category definitions and currency bases preclude a direct market-share conclusion, but demonstrate why Kioxia’s flat guide should not simply replace Sandisk’s own forecast. Near-term trading relevance lies in earnings-estimate revisions and the distinction between pricing deceleration and actual price declines. The negative read-through is strongest for valuation assumptions that require repeated large pricing increases. The longer-duration positive is sustained supply-demand tightness and a stronger technology platform. Neither Kioxia’s LTA coverage nor the shared manufacturing relationship establishes a guaranteed Sandisk margin floor. Sandisk had already reported before this call, so the new evidence is corroboration and a test of forecast assumptions, not a preview of its completed quarter. NAND MEMORY: ENTERPRISE SSD DEMAND IS BROADENING BEYOND A SINGLE SUPPLIER, BUT DIVERSIFIED MEMORY EXPOSURE MATTERS (READ-THROUGH 2) Affected companies: Micron Technology (MU: United States), SK hynix (000660: South Korea), and Samsung Electronics (005930: South Korea). Directional impact and magnitude: Positive and potentially substantial for each company’s enterprise NAND franchise. The consolidated impact is less direct than for Sandisk because DRAM, HBM, and, in Samsung’s case, downstream devices and other businesses also determine earnings. The supporting call evidence is specific: agentic AI adoption, increasing KV-cache requirements, expanding data lakes, and growing high-capacity QLC SSD demand. Kioxia expects NAND market bit growth in the low 20% range in calendar 2027 and projects its QLC bit mix to increase from approximately 30% in 2027 to 40% in 2028. Management expects data-center and enterprise demand to offset weaker smartphone and PC demand. The transmission mechanism is a combination of higher enterprise storage content, favorable product allocation, and sustained industry pricing. Enterprise qualifications and firmware integration can also make customer relationships more durable than undifferentiated component transactions. These mechanisms favor suppliers with qualified enterprise portfolios, rather than every company exposed to memory in some form. The peer evidence confirms that this demand is not unique to Kioxia. Micron’s September 30 release stated that its 7600 and 9650 SSDs were shipping for KV-cache applications and that its 6600 ION product had achieved a strong volume ramp. SK hynix’s October 28 release described expanding sales of 321-layer TLC enterprise SSDs and plans to move that process above 50% of domestic NAND production capacity in Q4. Samsung’s October 29 release similarly highlighted server SSD demand and increasing QLC requirements associated with personal AI agents. Kioxia therefore strengthens the duration argument for a broad enterprise-storage expansion rather than proving a company-specific demand anomaly. Ownership is important for portfolio mapping. Solidigm exposure belongs within SK hynix’s investment case; it is not an independent listed beneficiary. SK hynix also announced that it had discontinued its review of a Solidigm IPO. Samsung requires an additional distinction: upstream memory pricing benefits coexist with downstream device cost pressure. Its latest release described weaker device profitability despite resilient flagship demand. The group should not be modeled as a pure beneficiary of component inflation. The near-term catalyst is evidence from subsequent pricing and qualification updates that enterprise strength continues to outweigh consumer weakness. The longer-duration shift is NAND becoming a more strategically important part of AI infrastructure. The principal limitation is competitive allocation: Kioxia’s LC9 progress validates the market but can also increase competition. Its projected QLC mix is not evidence that Micron, SK hynix, or Samsung will gain share. No incremental HBM revenue can be inferred directly from this NAND call. PCS: HIGHER SELLING PRICES CAN MASK DETERIORATING UNIT ECONOMICS (READ-THROUGH 3) Affected companies: HP Inc. (HPQ: United States) and Lenovo Group (0992: Hong Kong). Directional impact and magnitude: Negative, with high materiality to PC operations. The consolidated effect is more concentrated at HP’s Personal Systems business and partially offset at Lenovo by infrastructure and services exposure. Kioxia explicitly stated that PC shipments were declining because of higher bill-of-materials costs and that PC NAND demand was expected to decline through calendar 2027. Management subsequently clarified that mobile and PC ASPs were not expected to decline, even though further increases would be more moderate. This is stronger evidence than a generic statement about weak consumer demand: the supplier is describing component inflation as an active constraint on downstream purchases. The transmission mechanism operates through both margins and volumes. OEMs that absorb higher SSD and memory costs suffer lower hardware gross profit. OEMs that pass through the increases raise system prices and risk slower replacement cycles, lower configurations, or lost units. Procurement contracts and inventory can delay the accounting impact, but cannot permanently eliminate it. Slower price inflation merely reduces the rate at which costs worsen; it does not restore the previous cost base. HP’s latest reported quarter illustrates the distinction. Personal Systems revenue increased 18% while units declined 16%, with a 4.6% operating margin. The revenue and unit figures imply approximately 40% higher revenue per unit, reflecting pricing and mix rather than physical shipment growth. This does not isolate NAND’s contribution, but demonstrates why headline revenue growth is an unreliable measure of underlying PC demand in the current environment. The earnings sensitivity is substantial because hardware margins are thin. At a 4.6% segment operating margin, an illustrative 100-basis-point unrecovered cost increase would consume approximately 22% of Personal Systems operating profit, holding revenue and other costs constant. Lenovo’s 7.1% Intelligent Devices Group margin provides a larger cushion, although that segment includes products beyond PCs. These are sensitivities, not forecasts of actual margin deterioration. Important counterevidence prevents an unconditional company-wide bearish conclusion. HP recently raised guidance and reported improved memory availability, while Lenovo described effective supply-chain management and strong infrastructure growth. Kioxia’s call therefore challenges the durability of downstream margins and unit demand, not the accuracy of those already-reported results. The next relevant catalysts are the companies’ subsequent results and 2027 procurement and demand commentary. The longer-duration risk is that an AI-PC replacement narrative fails to overcome higher system prices. The cross-portfolio implication is that upstream memory longs and PC OEM longs are not equivalent expressions of AI adoption: the same NAND shortage supporting supplier profits can undermine customer volumes. SMARTPHONES: VALUE-SEGMENT OEMS FACE A MORE PERSISTENT COST AND AFFORDABILITY SQUEEZE (READ-THROUGH 4) Affected companies: Shenzhen Transsion Holdings (688036: China) and Xiaomi Corporation (1810: Hong Kong). Directional impact and magnitude: Negative, with high exposure for Transsion’s handset-centered business and meaningful but more diversified exposure for Xiaomi. The impact is likely to be greater in price-sensitive device categories than in premium models. Kioxia stated that shipments of lower-priced smartphones were moderating faster, while premium models remained resilient. It expects broadly unchanged smartphone NAND demand through calendar 2027 despite continued enterprise growth. The call also indicated that mobile NAND prices would increase more moderately rather than decline. The precise mechanism is a disproportionate increase in storage cost relative to the selling price and gross profit of an affordable handset. Raising retail prices can push devices outside local purchasing budgets. Holding prices constant compresses margins. Reducing storage specifications protects some cost but weakens the product proposition and can constrain NAND content growth. The investment implication is not simply lower handset units; it is an unfavorable trade-off among volume, specifications, and profitability. Transsion’s emerging-market focus makes affordability particularly relevant. The company’s geographic exposure increases sensitivity to local purchasing power and, where currencies weaken, the local-currency cost of globally priced components. This is a geographic exposure assessment, not a claim that every market is experiencing the same currency movement. Xiaomi’s value-oriented products face a similar pressure, although premiumization, services, connected devices, and other businesses diversify the consolidated outcome. Xiaomi’s published Q2 operating figures provide corroboration: smartphone ASP reached a record while shipments declined 26.3% year over year. The combination demonstrates that higher realized prices can coexist with substantial volume pressure. It does not establish that memory inflation caused the entire decline, but is consistent with Kioxia’s description of increasing demand sensitivity. The near-term catalyst is late-2026 holiday and promotional-season sell-through, followed by the pricing and specifications of 2027 models. Aggressive promotion may protect units at the expense of margins; weaker promotion may preserve unit economics while worsening volumes. The longer-duration implication is potential concentration of demand in premium products and better-capitalized vendors with stronger procurement capabilities. Premium resilience should not be confused with immunity to input inflation, and aggregate smartphone NAND demand remaining flat does not mean handset profitability remains stable. The most useful relative conclusion is that value-oriented handset exposure is more vulnerable than an undifferentiated smartphone-sector basket. SEMICONDUCTOR EQUIPMENT: TECHNOLOGY-CONVERSION SPENDING IS THE CLEARER BENEFICIARY THAN A BLANKET CAPACITY BOOM (READ-THROUGH 5) Affected companies: Lam Research (LRCX: United States) and Tokyo Electron (8035: Japan). Directional impact and magnitude: Positive for NAND-related etch, deposition, and selected bonding exposure; moderate at the diversified company level. The call supports the spending direction but does not identify incremental awarded orders or justify a precise revenue uplift. Kioxia identified BiCS8 and BiCS10 as investment priorities, discussed current-year expenditure somewhat above JPY 500 billion but below JPY 600 billion, and reiterated that its aggregate 3-year capital envelope was not intended to change materially. Management plans to use existing Y7 and K2 space before expanding K3, whose operations are targeted for fiscal 2029. The mechanism is process intensity and technology conversion, not merely more wafer starts. Advanced 3D NAND requires demanding high-aspect-ratio processing and extensive deposition. Architectures that separately fabricate and bond logic and memory arrays create additional bonding-related requirements. Lam has direct product exposure through its NAND etch and deposition portfolio, while Tokyo Electron has relevant etch, deposition, and wafer-bonding capabilities. These product exposures are verified; specific Kioxia equipment allocations were not disclosed. A critical qualification is that bit growth does not translate proportionally into wafer or equipment growth. More layers, higher density, better yields, and increased QLC usage can raise delivered bits without equivalent increases in wafer starts. Holding process and yields constant, increasing QLC from 30% to 40% of delivered bits reduces the required memory cells per delivered bit by approximately 2.7%. This simplified physical sensitivity illustrates why the low-20% bit-growth outlook cannot simply be applied to equipment revenue. The near-term positive is equipment demand associated with conversions and filling existing cleanroom space during late 2026 and 2027. K3 is a longer-duration opportunity, not an immediate fiscal 2029 revenue item for equipment suppliers: ordering, delivery, installation, and acceptance would occur at different stages before production. However, no sufficiently detailed schedule was disclosed to assign those revenues to particular quarters. Lam had already presented a strong NAND investment outlook on October 21, including opportunities across multiple storage tiers. Kioxia adds customer-side corroboration but does not establish an unanticipated increase in the industry spending envelope. The prior K3 announcement also made investment contingent on government support. The highest-conviction conclusion is selective process-equipment support, not an indiscriminate positive read-through to every semiconductor equipment category or an assumption that the entire future fab budget is newly incremental. HARD DISK DRIVES: NEAR-TERM SCARCITY SUPPORT COEXISTS WITH A LONGER-TERM QLC COMPETITIVE THREAT (READ-THROUGH 6) Affected companies: Western Digital (WDC: United States) and Seagate Technology Holdings (STX: United States). Directional impact and magnitude: Moderately positive for near-term mass-capacity storage demand and pricing resilience; negative over a longer horizon for performance-sensitive warm-storage opportunities exposed to QLC competition. The call does not support an immediate HDD replacement or earnings-collapse thesis. Kioxia attributed QLC adoption to expanding AI-generated data and described QLC SSD growth in warm storage. It highlighted mass production and ongoing qualification of its approximately 240 TB LC9 product and projected QLC at 30% of bits in 2027 and 40% in 2028. Simultaneously, it described persistent NAND tightness and no current signs of price declines. These observations create opposing but distinguishable effects. Rising data creation and retention support the requirement for storage capacity across both flash and HDD tiers. Expensive NAND can also preserve HDD’s economic appeal where acquisition cost per unit of capacity remains decisive. The near-term positive mechanism is therefore expanding storage demand combined with limited availability of an alternative technology, not a contractual transfer of Kioxia customers to HDD suppliers. The longer-duration negative is that high-capacity QLC improves the competitive proposition of flash for selected warm-storage workloads. Greater density, lower access latency, and potential infrastructure efficiencies can justify SSD adoption even when acquisition cost per terabyte is higher. This pressures the boundary of HDD’s addressable market rather than proving that bulk archival storage will migrate wholesale to flash. The peer disclosures argue strongly against treating QLC progress as evidence of an immediate HDD downturn. Western Digital’s October 22 release reported 55% year-over-year revenue growth and guided to further strong growth, citing both exabyte demand and pricing. Seagate’s October 27 release described strengthening demand and a multiyear growth phase, with its HAMR roadmap intended to expand capacity through areal-density gains. These improvements were already reported before Kioxia’s call. The relevant near-term catalysts are December-quarter demand, pricing, and capacity-roadmap execution. The longer-duration competitive indicators are actual enterprise QLC qualifications, workload-level total cost of ownership, and SSD pricing relative to HDD. Kioxia’s projected QLC percentage is a NAND bit-mix measure, not a percentage of the HDD market being displaced. Ownership also matters: Sandisk is the direct listed flash-manufacturing partner associated with Kioxia’s NAND platform. Western Digital’s operating read-through is primarily through HDD demand and competitive positioning, not through attribution of Sandisk’s flash operating profits. CLOUD INFRASTRUCTURE: NAND PROFITS ARE PARTLY A CUSTOMER CAPITAL-INTENSITY HEADWIND (READ-THROUGH 7) Affected companies: Microsoft (MSFT: United States) and (AMZN: United States). Directional impact and magnitude: Negative at the margin for infrastructure acquisition costs and free-cash-flow conversion; potentially meaningful within storage procurement but smaller and unquantified at the consolidated company level. This is not a standalone bearish conclusion on cloud demand or the companies’ overall earnings. The supporting call evidence is the combination of substantial NAND price increases, continuing enterprise ASP strength, longer-term allocation commitments, and growing storage requirements for AI. Kioxia’s approximately 80% operating margin is a supplier-side manifestation of the same scarcity that buyers must manage. Its commentary provides no evidence that cloud operators will receive near-term price relief. The transmission mechanism begins with a higher cost to acquire a given amount of usable storage. For owned infrastructure, that generally raises capital expenditure and initially affects free cash flow more directly than operating income. Depreciation follows after assets enter service. Lease or hosted arrangements can alter the accounting and timing, but do not eliminate the underlying economic cost. Procurement terms and inventory buffers determine how quickly supplier price movements reach the customer. Storage demand can also grow faster than accelerator counts as workloads retain more context, generated content, checkpoints, and retrieval data. Consequently, increasing NAND spending does not prove proportional growth in GPU purchases or cloud revenue. Conversely, the call does not establish that storage costs will crowd out a specific amount of accelerator investment. The required budget allocation and component-cost disclosures are absent. Microsoft and Amazon had already reported strong cloud growth before this call. Microsoft’s October 28 release showed Azure revenue growth of 48%; Amazon’s October 29 release showed AWS growth of approximately 39%. Amazon also disclosed trailing-12-month free cash outflow of USD 42.4 billion, primarily reflecting increased property and equipment purchases associated with AI. That existing cash-flow pressure should not be attributed specifically to NAND, but it makes additional infrastructure-cost inflation economically relevant. The near-term catalyst is subsequent capital-spending, cash-flow, and infrastructure-efficiency guidance, particularly as 2027 budgets become clearer. The longer-duration implication is a higher hurdle for monetizing storage-intensive AI workloads: utilization, retention policies, compression, tiering, and service pricing must improve sufficiently to offset rising physical infrastructure costs. Neither company was identified as a specific counterparty to the LTAs discussed on the call. The exposure is to the shared procurement market, not a verified Kioxia contract. The cross-portfolio conclusion is that upstream memory profitability and cloud-customer cash generation can diverge even while both report strong demand. Owning both therefore does not eliminate the economic consequences of storage inflation.