Most China tech ETFs give you zero exposure to the companies actually building Nvidia's AI infrastructure
I've been trying to add some China tech exposure to my portfolio and realized something that surprised me. The two most commonly mentioned options, KWEB and CQQQ, are almost entirely internet and e commerce names. Tencent, PDD, Meituan, Baidu. Those are fine companies but they have basically nothing to do with the AI hardware buildout that's driving the current cycle. After GTC 2026 this week I went down a rabbit hole on Nvidia's supply chain. Jensen Huang projected $1 trillion in AI chip orders through 2027 and the systems being built need massive quantities of 800G optical modules to connect GPUs inside each rack. TrendForce estimates shipments jump from 24 million to 63 million units next year. Two Chinese A share companies, InnoLight (300308.SZ) and Eoptolink (300502.SZ), supply roughly 60% of Nvidia's orders in this category. Eoptolink's revenue was up 283% in the first half of 2025. On the autonomy side BYD and Geely just joined Nvidia's L4 self driving platform. Cambricon, China's leading AI chip startup, is targeting 500,000 chips in 2026 with ByteDance as its biggest customer. None of these companies are in KWEB. None are in CQQQ. They're A share listed which means most international ETFs simply don't hold them. I ended up finding CNQQ which tracks a Solactive index of about 100 Chinese tech companies across both A shares and Hong Kong. It holds InnoLight at 3.5%, Eoptolink at 1.3%, Cambricon at 2.3%, and BYD at 1.9%. The index weights companies partly by R&D intensity which is why it skews toward hardware and manufacturing rather than pure internet plays. About 50% of the portfolio is in A shares and while there is some overlap with KWEB and CQQQ on the Hong Kong listed internet names like Tencent and Alibaba, the A share hardware and manufacturing holdings that connect to the AI infrastructure theme are almost entirely absent from those other funds. There are real risks. It's a newer fund with limited live track record. CPO (co packaged optics) could eventually reduce demand for pluggable modules, though analysts say that's at least three years out. And any China allocation carries geopolitical tail risk that's impossible to fully price. But if you're looking for China tech exposure that actually connects to the AI infrastructure theme rather than just being another internet basket, the difference in underlying holdings between these ETFs is worth understanding. I was genuinely surprised how little overlap there is once you look past the top line "China tech" label.