ASML

South Korea's AI Memory Crash

The article argues that the AI supercycle is a stack of asynchronous S-curves where bottlenecks rotate; supply is multi-year while demand reprices quarterly, producing localized, violent drawdowns at overstretched nodes. In late July 2026 three distinct shocks hit the stack: a semiconductor/memory selloff, a financing squeeze, and a durability/geopolitical shock centered in Seoul. Micron, SK hynix and Samsung reported record operating margins (81%, 76%, 70%) and said memory will remain undersupplied beyond 2027, yet Korea’s stock market plunged as China’s ChangXin Memory Technologies (CXMT) completed a large IPO and reports surfaced of a domestically built immersion DUV lithography tool. The piece frames the event as a test of the memory node’s durability: Korea’s economy is highly concentrated in memory, and a state-backed entrant with scale and capital threatens the disciplined under-build that sustains the memory toll.

What Will More AI Intelligence Do?

The essay argues that although AI has achieved superhuman ability in narrow tasks (solving open math and cryptography problems), the broader economic and social impact has been more incremental than some expected. One hypothesis is that intelligence faces diminishing returns because the information extractable from data is bounded or costly to obtain; critics propose governance and frictions also slow change. The author highlights three mechanisms by which AI could still drive large productivity gains: replicability (running many agents in parallel), roboticization combined with energy/battery improvements, and AI’s ability to extract and diffuse distributed tacit knowledge or discover “cloud laws” — complex regularities humans cannot easily formalize. The piece cites examples (Zeiss/ASML mirrors, rare-earth refining) and surveys views from researchers including Francois Chollet and Arvind Narayanan.

AI model distillation and cross-border IP debate

Exponential View analyses concerns about Chinese AI labs allegedly distilling the outputs of leading US models, the technical feasibility of external distillation, and the unsettled legal status of model outputs as intellectual property. The newsletter cites allegations by Anthropic that DeepSeek, Moonshot and MiniMax used millions of Claude chats, notes Stanford’s Alpaca admissions about training on frontier-model outputs, and discusses industry responses and supply-chain developments (e.g., China achieving ~41% domestic AI chip supply in 2026). The piece frames distillation as a long-standing ML technique, distinguishes internal vs external distillation, and highlights the policy and national-security implications of treating model outputs as protected IP.