Table of contents
Kathy Xu, founder of Capital Today and often referred to as the ‘queen of Chinese VC’, has built her career on targeted investments that have proved to be masterstrokes: deals capable of repaying the entire fund with exceptional multiples.
His track record includes JD.com, NetEase and Meituan, each of which serves as a case study in patient capital, disciplined management of reserves and long holding periods. Now, in the age of AI, Xu has placed one of his boldest bets: ByteDance.
ByteDance as a full-stack challenger
During an informal chat with Stephanie Hui of Goldman Sachs Asset Management at the HKVCA’s Greater China Private Equity Summit in Hong Kong, Xu stated that ByteDance’s main advantage lies in its rare full-stack control.
Globally, only a handful of companies possess chips, data centres, language models (LLMs) and consumer applications. In the United States, Google fits this profile. In China, ByteDance stands out, with Alibaba a distant competitor.
The company’s strengths are striking and multifaceted. TikTok/Douyin’s recommendation engine generates huge cash flows, its reach among consumers is unrivalled, and it is expanding into e-commerce, local services, advertising and gaming. Ultimately, it aims to become a leading global player in the field of AI.
For Xu, ByteDance is China’s only realistic challenger to US hyperscalers: a platform positioned at the heart of the next-generation operating system for the AI era.
Investing in it is both a bet on exceptional returns and on China’s role in the global race for AI. For Capital Today, it is the cornerstone of an AI-focused portfolio strategy, designed to capitalise on the ‘tornado’ phase of adoption, during which growth can reach 200–300 per cent a year for several years.
The thesis on the AI operating system
Xu describes the potential of AI as an “OS moment”. Large language models combined with agents form the emerging layer of the operating system. Falling costs, driven by architectures such as the “mixture-of-experts” and open-source innovations like DeepSeek, are fuelling a boom in applications.
Xu points out that around 90% of the improvement in agents’ capabilities stems from updates to large language models (LLMs), which means that accountable agents must possess proprietary data and maintain ‘human-in-the-loop’ processes. The lessons from e-commerce apply here: super-platforms endure, whilst verticals without a ‘moat’ fade away quickly.
Basic models may become a commodity, but they will only deliver value to application-level agents if they are defensible, Xu said. Proprietary data and human oversight are non-negotiable.
Capital Today’s strategy
The construction of Capital Today’s portfolio reflects this worldview. The firm structures its funds to be between $600 million and $800 million in size to support four potential ‘home runs’, each of which requires up to $120 million in reserves. It avoids small, diluted positions, instead aiming to double and triple down on winners. Early bets on AI, such as Kimi and Zhipu AI, have already seen multiple top-ups over three years to build significant stakes.
His discipline is evident in past successes: JD.com was held for 12 years with a stake of around 40%; Yifeng Pharmacy for 16 years with a stake of around 30%. Meituan saw around 55% of Fund II’s capital reinvested in a concentrated position that significantly boosted returns. The lesson is consistent: the intensity, stake, reserves and long-term holding are controlled by the investor.
Xu argues that the greatest risk lies in misinterpreting the business model. To mitigate this, he emphasises the need for a thorough and rapid analysis of competitors and suppliers, as well as meeting with 15–20 founders within a short timeframe. His three-stage due diligence process—comprising vision assessment, benchmarking questions and in-depth interviews with the team—ensures conviction before capital is committed.
The lessons learned from the history of e-commerce are shaping its cautious approach to AI: avoiding undifferentiated vertical players and focusing on platforms with sustainable competitive advantages.
Lessons from Wahaha and NetEase
From the very beginning, Xu’s philosophy has been shaped by his encounters with visionary founders.
Her first investment, Wahaha, taught her the importance of ‘killer instinct’. In 1995, bottled water was virtually unknown in China. Wahaha’s founder, Zong Qinghou, spotted the opportunity after a trip abroad and, against conventional wisdom, set up three production lines at once.
Her first investment, Wahaha, taught her the importance of ‘killer instinct’. In 1995, bottled water was virtually unknown in China. Wahaha’s founder, Zong Qinghou, spotted the opportunity after a trip abroad and, against conventional wisdom, set up three production lines at once.
Years later, Xu stood by William Ding Lei of NetEase during a crisis. The company’s share price plummeted from around $5 to 60 cents in the midst of an SEC investigation, yet Ding remained steadfast. On his 30th birthday, he told Xu of his two goals: to build China’s leading online company and to generate profits for shareholders.
He changed direction, shifting from SMS services to online games, and spent years perfecting “Journey to the West II” until it became a hit. NetEase bounced back, securing Xu a return of 800%. From Ding, he learnt the value of resilience, quick learning and long-term ambition.
These stories have cemented his approach to assessing founders: a killer instinct, the ability to learn quickly, and a sense of responsibility towards shareholders. In the retail sector, this is complemented by frugality and care for employees. In the field of artificial intelligence, the archetype is shifting towards young scientists who coordinate agents within extremely lean teams. (Photo courtesy of VC-News)
Note to the reader: This article is licensed under the Creative Commons Attribution – NoDerivatives 4.0 International (CC BY-ND 4.0), which allows redistribution with attribution but no modifications.
ALL RIGHTS RESERVED ©
