UBS announced on Friday that it has reduced the combined weight of semiconductor and hardware holdings in its AI investment strategy to approximately 61% from near 76%. The move was presented as profit-taking after a rally - driven in part by Micron - lifted the semiconductor-focused SOXX index by 4% in a single session and by 10% over the month of June.
In a note from the bank's equity strategy team, UBS said the selling activity was concentrated in small- and mid-cap companies across a range of AI supply-chain niches. The areas cited include optics, baseboard management controllers, multilayer ceramic capacitors, substrates, cooling systems, advanced packaging and analog components. Those reductions represent a tactical trimming of exposure rather than an abandonment of the sector.
UBS emphasized that, despite the cutbacks, its AI strategy still maintains a material overweight to semiconductor and hardware stocks. The strategy remains roughly 20-25 percentage points overweight relative to the Nasdaq 100, which the bank estimates has about 42% of its exposure tied to AI semiconductors and hardware.
Concurrently, the bank said it materially increased allocations to more defensive segments of the AI ecosystem. Exposure to data center operators, telecommunications companies and selected payments names rose to around 20% from under 1% in early June. UBS said it targeted names with prudent balance sheets and stable dividend profiles as part of that shift.
UBS reiterated that the medium-term demand narrative for AI appears intact, supported by accelerated cloud adoption and growth in agentic AI workloads. Still, the bank cautioned that a notable decline in hyperscaler stocks could prompt management teams to cut capital expenditure commitments, a development that would pose downside risks to parts of the hardware and services supply chain. The note observed that the top five hyperscaler shares had fallen, on average, about 20% during June.
On specific semiconductor positioning, UBS said it continues to favor foundry exposure such as TSMC, semicaps including Applied Materials, memory names such as SK Hynix and select compute exposure. The bank also maintained a pronounced underweight to the so-called Magnificent Seven, which together account for roughly 18% of its portfolio weight.
Implications for markets and sectors: The rebalancing reflects UBS's effort to lock in gains in cyclical, supply-chain dependent semiconductor and hardware names while reallocating into income and stability-oriented parts of the AI ecosystem such as data centers, telcos and payments. The changes affect equipment suppliers, memory and foundry providers, and hyperscaler-related capex dynamics.