Speed, Retention, and Token Costs: How AI Startups Can Win the Race

01 DeepSeek and the Future of Open‑Source AI

Founder Luo Yihang and investor Zhu Xiaohu discuss DeepSeek’s role in stimulating China’s AI ecosystem, noting that while it boosted enthusiasm, it has not yet become the nation’s "new infrastructure".

Zhu argues that open‑source models are crucial to prevent AI monopolies, citing rapid growth of Chinese model downloads on Hugging Face and predicting parity with U.S. closed‑source models within a year.

Luo adds that token costs differ between the U.S. and China, with Chinese users consuming fewer tokens due to lower prices and subsidies.

02 Manus – The Power of Go‑to‑Market

Manus is highlighted as a case study where superior go‑to‑market execution, rather than just product tech, gave Chinese startups a competitive edge in global markets.

Zhu stresses that speed and retention are the primary barriers for AI applications; without rapid deployment, even strong retention cannot compensate.

Luo introduces token consumption as a key metric for AI‑heavy businesses, treating high token usage as a proxy for AI intensity.

03 AI Coding – A Subsidized Game

Zhu warns that AI coding tools are dominated by large firms and operate at negative margins, making them unattractive for startups.

Luo notes that pricing models based on usage lead to poor retention for developers who can easily switch tools.

04 Robotics – Investing in "Work‑horse" Robots

The conversation shifts to robotics, with Luo describing investments in practical, high‑impact robots such as ship‑cleaning and massage devices that replace entire job functions.

Zhu emphasizes that successful robots must combine utility with revenue generation, not just novelty.

05 AI Hardware – Do Less, Not More

Zhu advises AI hardware founders to focus on simplifying products and ensuring mass production capability rather than adding flashy features.

Luo agrees, stating that hardware success depends on usability, emotional appeal, and clear ROI.

06 Globalization – Be a Proud Chinese Company

Zhu argues that Chinese AI startups should embrace their identity and compete globally without pretending to be foreign entities; success in the C‑end market is already strong, while B‑end expansion requires local sales expertise.

Luo asks about market prioritization, and Zhu suggests aligning strategy with team background—targeting the U.S. for large‑scale opportunities, Japan for niche markets, and Southeast Asia for early traction.

07 Valuation and Capital – High Valuations Reduce Margin for Error

Zhu cautions that inflated early valuations increase pressure and limit future fundraising flexibility.

Luo inquires about accessing more U.S. dollars, and Zhu notes that many Chinese LPs’ capital is locked in large private companies awaiting IPOs, suggesting Hong Kong as a viable listing venue.

08 AI Startup Speed – Triple the Pace of Mobile Internet

Zhu clarifies that VC expectations focus on cash‑back time rather than a 12‑24‑month exit, emphasizing rapid ROI on marketing spend.

Luo raises concerns that the best AI startups may need less early funding due to fast break‑even, and Zhu responds that early metrics like user engagement still guide investment decisions.

Both conclude that AI opportunities will emerge at three times the speed of the mobile era, requiring founders to look beyond the three main big‑tech pathways.