Paradis
AI & Technology Investment Research
Topco Scientific is hard to go wrong with, thanks to really competent management and how far it has come in the past five or six years on the back of semiconductor supply chain shifts. Per @Blinklebloop, it is Shin-Etsu's primary agent for silicon wafers and photoresists in Taiwan and parts of China, looks relatively cheap at about 15 times earnings for its supply chain position, and management expects double-digit revenue growth this year. x.com
Topco Scientific is cheap at about 15x earnings because it is a decent company riding a tailwind, and it still has legs after starting to run. It is Shin-Etsu's primary agent for semiconductor materials such as silicon wafers and photoresists in Taiwan and parts of China, and has a silicone JV with Shin-Etsu. Management expects double-digit revenue growth this year, with exposure to CMP slurries, water treatment and advanced packaging. Such companies tend to outperform with half the volatility. … Show more
Some semiconductor stocks have returned to reasonable valuations as fundamentals improve, making recent declines buying opportunities. Many had traded at two to three times their median valuations; the correction brought some back toward normal, while sustained AI spending should benefit suppliers for years. x.com
Soitec holds a monopoly on the Photonics SOI wafers the AI photonics trade runs through, yet trades at just 3.6x forward EV/revenue. Silicon photonics chips in AI data centers, from transceivers to Nvidia switches and Broadcom ASICs, are built on it via its proprietary Smart Cut process; GlobalWafers exits by 2027 and Shin-Etsu sells little. Optical interconnect grows from about $10 billion to over $70 billion by 2030, and top customer Tower Semiconductor targets 5x capacity by end-2026. x.com
Shin-Etsu Chemical is shifting from a chemical conglomerate to a chip-supply bottleneck, owning lithography and chemistry patents essential to the glass-substrate era. In 1,000W+ AI data centers glass is already required, as organic substrates warp and crack under the heat of Nvidia's Rubin systems. Intel and Samsung have moved pilot lines into volume production, with Shin-Etsu supplying bonding chemistry and masks. Cyclical PVC cash funds its electronics push, supporting a re-rating. x.com … Show more
Soitec will recover because it is literally the foundation for expanding co-packaged optics capabilities and is too big to stay down. x.com
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
TSMC is the one company the AI chip boom cannot happen without, since every advanced chip runs through its fabs and Apple, Nvidia, AMD and Qualcomm all depend on it. Q1 profit rose 58% to a record and revenue grew 35% to $35B, with high-performance computing now 61% of sales. It is spending $52-56B on capex this year to build capacity competitors are a decade behind replicating. x.com
TSMC's push to localize its supply chain is bullish for Taiwanese small-cap equipment makers, while Samsung and SK hynix squeezing suppliers hurts Korean substrate makers. TSMC is evaluating local suppliers such as Gudeng, GPTC, VisEra and GPM for CoPoS and panel-level packaging as bottlenecks shift to packaging and materials. Samsung and SK hynix reportedly seek H2 2026 substrate price cuts to reverse a 3%–4% increase, pressuring Simmtech, Korea Circuit and Daeduck Electronics. x.com x.com
SUMCO is the cleanest play on the silicon wafer upcycle, because as a pure-play 300mm wafer maker at breakeven, even a modest price inflection creates massive earnings leverage. It holds about 40% global share in leading-edge logic. Per SemiAnalysis, advanced-logic epi wafer demand accelerates next year toward nearly 1M wafers a month by 2028 while HBM absorbs oversupply, and 2026 wafer-maker capex is only about 30% of peak. New contracts for 2028 onward may price epi wafers about 20% higher. … Show more
TSMC and ASML are good longs because their bottlenecks are the least easily solved, so the excess returns from owning them should prove more durable through a cycle. Bottlenecks elsewhere tend to get solved, which is why a significant improvement in another part of the chain is nice but not revolutionary, and should already be assumed by anyone underwriting semis or AI demand. x.com
