TSMC manufactures chips for most of the leading tech companies while Intel still struggles with its own foundry. It makes NVIDIA's GPUs, Apple's phone chips, AMD's processors and Google's TPUs, plus chips for Amazon, Microsoft and Meta. x.com  x.com 

Xintec offers TSMC's co-packaged optics testing bottleneck at a huge discount. TSMC's COUPE volume production starts this year with wafer-level testing flagged as one of three critical scaling challenges, and TSMC plans to pursue CPO with its subsidiary Xintec, which is ramping wafer testing capacity for second-half 2026 growth. 90% trial yields on 310mm pilot lines suggest technical risk is falling faster than its roughly $2.18B market cap reflects. x.com 

TSMC may lose pricing power as performance gains shift from shrinking transistors to stacking, interconnects and optics, while logic folding could make older processes competitive. Despite strong leading-edge demand, investors may overestimate the profit share its node leadership can retain. Its cautious expansion and past reluctance toward advanced packaging suggest a risk-averse culture that protects returns but could leave it late to disruptive technologies and key customers. x.com 

Amkor and ASE gain as TSMC's advanced-packaging spillover grows, making this the season for outsourced packaging. The share of spillover customers at TSMC has risen drastically, so the Amkor partnership was the viable option, and ASE has huge opportunities too. A post the author replies to calls packaging the biggest bottleneck, with chiplets, compute-memory stacking and silicon photonics driving growth. x.com 

TSMC's land deal for two 1.4nm fabs and an advanced packaging facility in Taiwan removes a near-term roadblock to sub-2nm deployment. The Longtan Science Park expansion zone grew from 89.63 to 104.19 hectares, and TSMC estimates $49 billion of initial investment for four sub-2nm plants. That supports Apple's expected use of 1.4nm A22 Pro chips in 2028 iPhones and unblocks advanced-node spending as sub-2nm becomes a competitive necessity. x.com 

FormFactor's production-ready wafer-level CPO probe gives it a test opportunity at a major foundry, but that business depends on one customer. CEO Mike Slessor says the CPO piece flows through a foundry worth 10% of sales; he did not name it, though an analyst named TSMC. Triton tests photonic wafers before stacking; post-stack test remains difficult. FormFactor expects CPO test revenue to significantly exceed $20m this year and estimates a $400m market by 2030. x.com 

TSMC sits in the part of AI's profit chain that everyone has to pass through. AI profits are split extremely unevenly: Nvidia earned $130.4 billion, several times everyone else combined, while server makers like Dell earn thin margins and model makers bleed, with OpenAI losing $8 billion on $13 billion of revenue. The best bets are high-barrier picks and shovels like Nvidia GPUs, TSMC's advanced packaging and HBM; server hardware and most AI apps likely destroy value. x.com 

Himax's microlens monopoly is real but narrower than the market thinks. As the Hunterbrook and Citrini report claimed, and Tianfeng Securities confirms, it is the sole microlens supplier for TSMC's COUPE Gen 1-2. That is not a monopoly on Nvidia's co-packaged optics: Himax is not among Nvidia's 11 official CPO partners, TFC, Corning and Orbray build CPO-grade fiber array units without it, and 80% of revenue is still display drivers. That leaves it a small, low-conviction long. x.com  x.com 

TSMC sits at the end of the money trail as agentic AI reprices the chip supply chain, and Google's $40 billion Anthropic deal is the clearest proof. That deal is effectively a prepaid inference contract for 5GW of TPU capacity over five years, and every TPU flows through TSMC, where advanced-node demand runs about 3x supply. Agentic AI burns 10-100x more compute per session than a chatbot. CoWoS packaging is the tighter bottleneck, with prices rising 10-20% a year, double logic wafers. x.com  … Show more

ASML, TSMC and BE Semiconductor are the safer, easier-to-understand way to own AI because they sit in the industrial part of the chain. The big question in AI is timing, which no one can call, and Nvidia's software moat is impossible to understand or trust with confidence. A quoted post notes that Crésus sold Nvidia, expecting faster AI progress to favor the owners of models and factories over chip designers. x.com