TR42: Beards, Wafer Fabrication Equipment, Claude Code

Doug shares his complete conversion to Claude Code (Opus 4.5), describing it as a fundamental shift in how he works. He demonstrates hitting usage limits on the paid tier, automating chart creation, building custom bots, and using Obsidian as a persistent knowledge store. The discussion covers practical workflows, skill issues versus tool limitations, and the addictive nature of AI-assisted coding.

• •Claude Code Opus 4.5 represents a step-change improvement - 'one shots everything' and handles complex tasks end-to-end
• •Practical applications: automated all charts for year-end outlook, built 5+ custom bots, created IPO tracking systems with automatic Excel exports
• •Obsidian + Claude Code workflow: daily notes, persistent markdown storage, cross-referencing deep research outputs, git sync between work and home
• •Key insight: 'Everything is a skill issue now' - limitations are primarily in how you prompt and structure requests, not the tool itself
• •Hitting $1000/month usage limits on Claude Code Max tier through intensive daily use across multiple concurrent projects

Dylan outlines Semi Analysis's major 2026 initiative to build comprehensive China research coverage, tracking Chinese wafer fab equipment manufacturers, chemical suppliers, memory players, and chip companies with the same detail as Western firms. The team is actively hiring analysts with deep China expertise and planning detailed WFE share analysis.

• •Launching dedicated China research practice covering WFE manufacturers, chemical suppliers (photoresist, etchants, precursors), and chip companies
• •Hiring Chinese nationals with specialized knowledge - example candidate discussed CXMT and niche memory players for 25 minutes
• •Upcoming post on China WFE share across different fab types with year-by-year projections
• •New mandate requiring 50% Chinese semiconductor manufacturing equipment in certain fabs driving research need
• •Expanding coverage up the stack: fabs, memory, logic (both trailing edge and front end)

Discussion of Sholto's AI-themed RTS game built entirely with Claude Code in one week, demonstrating the power of AI-assisted development. The game features US vs China factions, tech trees for AI research, robotics, and handles battles with 50,000+ units - all created without writing a single line of code manually.

• •Complete RTS game built in one week using only Claude Code prompting - zero manual coding
• •Game mechanics: industrial age through AI timelines, tech trees for model research, robotics, internet, hacking
• •US vs China factions with different modifiers (China: production/power advantages)
• •Performance breakthrough: handles 50,000 vs 50,000 unit battles, more performant than traditional RTS games
• •Developer approach: 'became a zombie, just typed ideas and clicked yes' - pure prompt-driven development

Analysis of 2025-2026 semiconductor market dynamics, including Oracle's controversial debt raise timing, TSMC capacity constraints, and Intel's unexpected recovery through NVIDIA's N3 wafer purchases. Discussion covers WFE spending projections and capital cycle continuation.

• •Oracle's debt raise timing criticized as 'own goal' - could have waited 3-6 months for better terms, triggered CDS concerns
• •NVIDIA buying all N3 wafers from TSMC forces Apple to Intel, 'Jensen saved Intel' by stealing Apple's capacity
• •TSMC expansion limited by construction workers and deployment friction, not just capital - 'can't employ 200% of Taiwan'
• •WFE market potentially reaching $140B, driven by AI infrastructure buildout
• •Capital cycle assessment: 'not at the end, drunkenly dancing somewhere in that timeless night' - expects continuation through 2026

Vision for API-first software consumption where traditional UIs become secondary to programmatic access. Discussion covers search APIs, Bloomberg's potential transformation, and the shift toward generative UIs that adapt to individual users.

• •APIs becoming more valuable than traditional interfaces - 'would rather pay for Bloomberg API than terminal'
• •Search APIs critical infrastructure: Google blocking SERP API, Brave emerging as alternative
• •Future prediction: computers ship with just Claude Code API, users build their own software/OS
• •Generative UI vision: interfaces adapt to individual consumption patterns, color preferences, information density needs
• •Neo Cloud founder 'super pilled' on generative UI - every interface will be token streaming customized to user

Comprehensive overview of Semi Analysis's aggressive hiring plans across multiple regions and specializations. Focus on building teams in Singapore, China research analysts, AI adoption tracking, and technical infrastructure roles.

• •Hiring 5+ junior analysts in Singapore for regional coverage and operations
• •Multiple China analyst positions: chip companies, analog, server companies, data center, grid modeling
• •AI analyst roles: adoption tracking, tokenomics, competitive analysis
• •Technical staff hiring: cluster operations, infrastructure (candidate who's never used Windows or OSX)
• •Sales positions and general 'if you're good, reach out' - rapid team scaling across all functions

Announcement of Semi Analysis's memory model product launch (currently beta for full-service customers) and hints at upcoming physical operations expansion in first half of 2026. Discussion of CapEx investments and infrastructure buildout.

• •Memory model launched in beta, available only to full-service customers initially
• •Mysterious first-half 2026 initiative involving physical operations and CapEx
• •Plans for potential conference/event with video content and major announcement
• •Infrastructure investments: Mac minis for headless bot operations, local compute for cron jobs
• •Expansion beyond pure research into physical/operational capabilities - 'somebody else has CapEx, we have CapEx'