These 10 Chinese stocks are powering U.S. data centers

Chinese mainland-listed firms are emerging as key AI infrastructure suppliers, with the 22V Research report saying the U.S. relies on Chinese companies for almost 30% of imports of AI-related products. The article highlights strength across Shenzhen-listed names in the data center supply chain, including CATL, Victory Giant, Sanhuan Group, Foxconn Industrial Internet, Sungrow Power, InnoLight, Eoptolink and TFC, many of which have at least doubled over the past 12 months. Foxconn Industrial Internet said cloud computing revenue surged 88.7% in 2025 to 602.68 billion yuan ($88.67 billion), underscoring momentum in AI-linked hardware demand.

Analysis

The market is still underpricing the extent to which AI capex is a systems-integration trade, not just a compute trade. If chip exports face friction, the spend simply migrates into the highest-beta domestic enablers: optics, power conditioning, PCBs, and passive components. That creates a more durable earnings runway for mainland suppliers because these parts are embedded into server architectures, are harder to substitute quickly, and often carry longer qualification cycles than the headline semis.

The second-order winner set is broader than the obvious “AI supply chain” basket. Rising fiber density, MLCC content, and power-management intensity imply that component inflation can outpace unit volume growth, so revenue expansion may exceed what end-demand data center build rates suggest. The losers are global server OEMs and Western component vendors that rely on China-based manufacturing capacity; if Chinese firms keep gaining share in lower-tech but indispensable subsystems, the margin pool shifts away from assembly and toward domestic Chinese suppliers with scale and pricing leverage.

The key risk is that this becomes a crowded thematic trade before it becomes a fundamental one. These names have already rerated sharply, so the next leg depends on order-duration visibility into 2027-28 and on whether export controls or licensing slow the flow of AI infrastructure demand through Chinese supply chains. A sharper-than-expected U.S. restriction on Chinese electronics content would hit optics and PCB names first, but the more probable near-term reversal is simply multiple compression if growth decelerates even modestly after the current re-rating.

Contrarian take: consensus is treating this as a China-AI hardware beta trade, but the cleaner expression may be scarcity pricing in niche components rather than broad “China tech” exposure. The strongest setup is to own the bottleneck suppliers with the most content-per-server and the least replaceability, while fading the broader basket where enthusiasm has outrun earnings revisions. In other words, the trade is less about who makes AI and more about who controls the plumbing.

Key Decisions for Investors

• Go long INNOLIGHT / EOPTOLINK on 3-6 month horizon into any 5-8% pullback; these are the cleanest expression of rising fiber intensity and should keep compounding if datacenter order books remain stretched into 2027-28.
• Pair trade: long Chinese optical/PCB leaders, short a diversified China hardware ETF or broad Shenzhen tech proxy to isolate the bottleneck beneficiaries from the crowded momentum basket; target 15-20% relative outperformance if earnings revisions continue to lead price.
• Buy call spreads on FOXCONN INDUSTRIAL INTERNET with 6-12 month tenor; the setup is favorable if cloud/AI server assembly continues to reaccelerate, but the spread structure limits damage if multiple compression hits the whole group.
• Avoid chasing CATL at current levels; its inclusion in the AI basket is more thematic spillover than direct AI leverage, so upside is more limited and downside from factor rotation is higher than for pure-play component suppliers.
• Set a tactical hedge via long JPM / short Chinese AI hardware names if policy headlines intensify; the benefit is that trade-policy volatility can hit mainland supply-chain equities faster than it changes global AI demand.