Glass substrates will likely become a big investment theme in 2027, especially if we remain in a bull cycle, where themes tend to get fast-tracked I’ve already done some coverage, including an interview with LPK’s CEO In case anyone wants to check it out

Musk said in 2017 that AI would likely cause WW3 One has to wonder how long before AI starts creating serious tensions between global powers Sadly, it feels like a matter of time Militaries would use ASI if it were available. It's simple game theory, and it could end terribly

$TSM and $ASX Break Ground on Taiwan’s First Advanced Packaging Cluster TSMC, ASE, Foxconn, and more than 20 semiconductor companies have begun developing Taiwan’s first advanced packaging supply-chain cluster in Kaohsiung. The first phase is already fully allocated ASE expects advanced packaging and testing capacity shortages to persist for another 3–5 years. Most of its planned capacity additions over the next three years are already booked. The company has seven new factories under construction, three undergoing equipment installation, and two additional plants planned TSMC will establish two facilities focused on CoWoS and panel-level packaging (PLP) technology validation, while Foxconn will develop an AI computing and smart manufacturing base Separately, ASE’s $3.4 billion advanced testing park is scheduled to begin operations in April 2027, with annual production value expected to reach $5.6 billion at full capacity

Unitree Launches A Humanoid Robot Hand for $6,000 Unitree Robotics is attacking the dextrous hand bottleneck with the Dex5-S, a dexterous robotic hand with 22 degrees of freedom, up from 20 in its previous model The new hand weighs 620 grams, supports a maximum payload of 2 kg and a continuous working load of 1 kg. The Pro version adds tactile sensing Prices start at $6,000 per hand

$FSLR Continues Patent Lawsuits Against Solar Rivals First Solar plans to withdraw its ITC complaint and resume patent lawsuits against $CSIQ, Jinko Solar, $TE, and Trina Solar in US courts First Solar holds TOPCon patents but manufactures CdTe thin-film solar cells, giving it little incentive to settle through cross-licensing Patent litigation, combined with US tariffs, is making it harder for foreign TOPCon manufacturers to enter the US market Some manufacturers are turning to alternative technologies to avoid the dispute, including PERC and heterojunction (HJT) solar cells

Development by 2028 and mass production by 2030 was always the predicted timeline, to be fair $LPK, Absolics, Rapidus, and SEMCO were aiming for mass production by 2028, but that seems unlikely, at least not at a large scale across the industry

TrendForce: US Data Centers Could Face a 170GW Power Gap by 2030 TrendForce projects global data center power demand capacity to triple from 161GW in 2026 to 490GW by 2030, driven by accelerating AI server deployments AI servers are expected to account for 54% of total demand by 2030, up from 33% in 2026, with their power requirements increasing from 54GW to 264GW The critical turning point is 2028, when grid connection delays could leave data centers with installed computing equipment but insufficient electricity to operate it The full 171.7GW gap is not an established electricity shortage. Of this figure, 124GW represents demand beyond Lawrence Berkeley National Laboratory's existing forecast. TrendForce estimates actual grid delivery shortfalls of 49GW against that reference scenario The largest projected regional shortfalls are concentrated in PJM (19.5GW), ERCOT (11.2GW), and MISO (7.6GW) PJM alone could see data center load growth of 29GW by 2030, against a deliverable baseline of just 9–10GW Grid upgrades, transformers, substations, and transmission infrastructure require years to complete, while permitting and interconnection delays add further constraints $IREN $HUT $CIFR $GLXY $CORZ $KEEL $CLSK

China's SiC Shortage Pushes Orders Toward Taiwan 6-inch SiC production is particularly constrained, with some manufacturers operating at full utilization. China's 8-inch capacity expansion remains limited by production yields and customer qualification requirements AI data centers are driving demand for 1,200V power devices as power architectures transition toward 800V systems. China also plans to expand its EV charging network to 28 million facilities by the end of 2027 Taiwanese beneficiaries include Episil (3707.TWO), Foxconn's HYS (parent: Diodes (DIOD), and VIS (5347.TWO). Episil leads in 6-inch production, while VIS is approaching mass production. HYS and Diodes are also expanding into 8-inch SiC Most of China's additional SiC demand is being absorbed domestically, leaving limited capacity for overseas customers

$MU Taiwan Union Moves Toward Strike Vote Micron’s Taiwan labor dispute is escalating after a second mediation meeting failed on September 21. The Taoyuan union plans to hold a strike vote over disagreements concerning profit-sharing and employee bonuses The union previously proposed allocating 15% of operating profit to employee profit-sharing. Micron announced a NT$1 million ($31,544) one-time bonus on September 11, but the union criticized the decision for being made without prior consultation Micron says it cannot commit to financial changes before its September 30 earnings release and October 9 board meeting. A separate mediation session with its Taichung union is scheduled for October 22

Physical AI Projected to Grow 20× by 2036 Nelson Tsay, president of Ennoconn ( a Taiwanese industrial computing company within the Foxconn Group, expects physical AI to take off in 2026 Tsay cited research projecting the physical AI market to grow from $6 billion in 2026 to $120 billion by 2036, a 35% CAGR. Humanoid robot shipments are expected to exceed 50,000 units in 2026, up more than 700% YoY Ennoconn operates three US factories with more than eight SMT lines, over 40 SMT lines in Europe, and manufacturing facilities in Southeast Asia

Agents Are A Blessing For NAND Prices The market expected NAND prices to peak in Q3 2026 > Instead, NAND prices have taken a very different direction The reason? Hyperscalers' enterprise SSD demand is nearly double earlier forecasts Hyperscaler NAND inventories stood at about 14 weeks in July, below the normal level of 15 weeks, with inventories expected to decline further in Q4 2026 What's driving this demand? 1. Agents AI agents perform repeated retrieval operations, access large databases, and generate substantial amounts of intermediate data Unlike traditional chatbots, agents can execute multiple tasks autonomously, repeatedly accessing and processing information. This increases the amount of data that needs to be stored and retrieved Cloud providers are consequently expanding their use of high-capacity QLC enterprise SSDs for vector databases, retrieval, and caching QLC NAND allows them to store more data at a lower cost per bit, making it particularly attractive for these workloads 2. KV-cache offloading AI models need to remember the context of a conversation or task to avoid processing the same information repeatedly. This information is stored in what is known as the KV cache Traditionally, this data is kept in expensive HBM or DRAM. However, not all of it needs to remain in high-speed memory at all times By moving less frequently accessed KV-cache data to SSDs, cloud providers can reduce memory costs while maintaining access to much larger amounts of context TrendForce specifically identifies DeepSeek-related architectures in China as a driver of growing demand for high-capacity QLC enterprise SSDs for KV-cache offloading

The CPU-to-GPU ratio rumors are moving markets Agents are changing the game Muse is already the most successful agent product Retail is now adopting agents after enterprise. The long-tail adoption is happening before our eyes $AMD and $INTC have gold in their hands

I'd be shocked if $NVDA ends up gaining market share in 2027 Especially with $AMD moving from mid-sized clusters to gigawatt-scale deployments I think AMD could still surprise the market

How on earth would 2–3 GW of compute generate only $17B in revenue for $IREN in 2030? Something is wrong here Also, modelling adjusted EBITDA for a neocloud? I’d like to see the full report. This snapshot feels extremely superficial