2Q25 Research Plan & Pre-Earnings Season Insights

Over the past three months, our research perspective has not only experienced an AI narrative shift but has grown increasingly optimistic about the sustainability of the AI story over the next 3-5 years.

• Coding has emerged as AI's first major application, potentially capturing 30% of the entire AI market by year-end. Looking at Anthropic's coding-related ARR alone: Cursor ($600M+), Claude Code ($200M+), Augment ($100M+), Replit ($100M+), Lovable ($50M+), and Bolt ($50M) already total $1.1B ARR. These products have low gross margins, and considering enterprise coding's use of Anthropic, coding represents 30-40% of Anthropic's ARR, likely reaching 30% of total LLM API ARR by year-end.

• Coding has resolved the AI ARPU debate, as we're seeing ROI-based payment trends increasingly outpace subscription models. Engineers aren't satisfied with Cursor's $20/$40 monthly plans—after exhausting 500 queries, many use cases generate $30 daily spend. With Claude Code, ARPU is even higher than Cursor.

• Clear ROI + visible TAM has filled the AI cycle's airpocket.

• The industry has shifted from supply chain-driven to application demand leading supply chain orders, creating a healthier ecosystem that's no longer fixated on supply chain metrics.

• The AI narrative is increasingly authentic, reminiscent of the internet boom's innovation wave. The past two years' startup trend emphasized high barriers/avoiding fundation model replacement; post-coding productivity iteration, the new trend embraces models and builds environments for LLMs. This has significantly reduced barriers to entry for LLM startups, driven exponential growth in new ventures, and enabled founders to demonstrate viable unit economics from day1.

From our preliminary research this quarter