Chinese funds push AI hardware holdings to 60% in Q2

Chinese active fund managers poured into AI hardware stocks in the second quarter, pushing combined holdings in electronics and communication sectors to 60% of total AI exposure while cutting software positions to historic lows.

"The rotation reflects a clear preference for near-term AI monetization through infrastructure spending over longer-duration software bets," the Xingzheng Strategy team, a China-based institutional research group, said in a note.

Funds boosted domestic computing chain holdings by 10.4 percentage points to 18.4% and North America-linked computing chain positions by 8.6 points to 24.1%. AI upstream materials saw the sharpest rotation, with holdings jumping 5.6 points to 12.9% as managers targeted semiconductor materials, PCB substrates and copper-clad laminates.

The repositioning shows institutional investors expect AI capital expenditure — estimated at roughly US$800 billion from global hyperscalers — to flow primarily into hardware supply chains over the next 12 to 18 months, leaving software and application-layer stocks at risk of continued outflows.

Inside the hardware rotation

Within the hardware basket, fund managers showed a clear preference for domestic computing names over their North American counterparts. The 10.4-point increase in domestic chain exposure outpaced the 8.6-point gain for the North America chain, even though the latter still commands a larger absolute share at 24.1%. The Xingzheng team tracks 41 domestic computing chain stocks — spanning CPU, GPU, materials, equipment, packaging, testing and memory — against 47 North America chain names covering optical modules, PCB, fiber optic cable and liquid cooling.

The divergence was most pronounced in upstream networking hardware. Optical modules, PCB and switches all saw significant inflows, while fiber optic cable was cut. On the chip side, memory chips and semiconductor materials and equipment drew the heaviest buying, while packaging, testing and analog chip positions rose only modestly and remain at historically low allocation levels.

AI materials emerge as a new battleground

The 5.6-point jump in AI upstream materials to 12.9% of holdings marks the most notable broadening of the AI trade. Within this category, fund managers added semiconductor materials including aluminum nitride, CMP polishing materials and sputtering targets; PCB materials such as ABF/NBF film and tungsten rods; and copper-clad laminate inputs including electronic fabric, copper foil and spherical silica powder.

The cuts were equally telling. Optical module materials — including rotatory optical plates and indium phosphide — were reduced alongside optical fiber preforms, liquid cooling materials and PTFE resin for CCL. The pattern suggests managers are rotating from earlier-stage AI infrastructure plays into deeper supply chain names that benefit from volume production scaling.

Software and applications left behind

The selloff in AI software and applications was broad and deep. Within the mid-layer model and software services category, only computer systems, EDA and operating systems saw marginal buying. AI applications fared even worse — only e-government software received small inflows, while most sub-sectors saw their allocation and overweight ratios fall to historically extreme lows.

Humanoid robots, satellite communications and autonomous driving — once high-conviction AI application bets — were all reduced, with their allocation ratios now sitting at historically moderate-to-low levels. The only consumer hardware categories that gained were AI smartphones, AI PCs and AI wearable devices.

What this means for investors

The data suggests Chinese institutional investors are treating the AI theme as a hardware cycle rather than a software platform shift — a bet that the US$800 billion in hyperscaler capital expenditure will translate into revenue for component and material suppliers before it reaches application-layer companies. For investors tracking the China AI trade, the key question is whether the hardware rotation has room to run or whether software names have been oversold to levels that could attract contrarian buyers.

This article is for informational purposes only and does not constitute investment advice.