Semiconductor ETF Inflows Offset 40% DRAM Plunge as Investors Buy the Dip

Investors have directed a combined $24.7 billion into four major semiconductor exchange-traded funds in less than a month, effectively neutralising the impact of a sharp market correction. This substantial influx of capital has occurred despite the Roundhill Memory ETF (DRAM) dropping nearly 40% from its peak on 22 June, with the fund retaining $23.4 billion in assets—only slightly below its late-June high of $25.9 billion. Other major funds, including the iShares Semiconductor ETF (SOXX), VanEck Semiconductor ETF (SMH), and leveraged Direxion Daily Semiconductor Bull 3X Shares (SOXL), also experienced substantial losses of 24%, 20%, and 61% respectively.

The sell-off, which saw key stocks such as Micron Technology, Marvell Technology, and Applied Materials give back gains, was driven by profit-taking following a strong second quarter and renewed concerns regarding the sustainability of artificial intelligence (AI) infrastructure spending. Specific inflows recorded were $8.8 billion for DRAM, $8.5 billion for SOXX, $5.1 billion for SOXL, and $2.3 billion for SMH. The DRAM figures highlight the forceful nature of the buying, as inflows have almost entirely offset the decline in asset value since the June peak.

These stocks had surged an extraordinary amount in a short period, with SOXX up 118% year to date and SMH up 86% before the correction. DRAM, which launched in April, had risen 191%, while the leveraged SOXL fund saw gains of 616% prior to the pullback. Although the recent decline has been steep, all four funds remain significantly higher on the year, suggesting that the market view of the long-term trajectory remains intact despite the short-term volatility.

The release of Kimi K3, a Chinese open-source AI model, has prompted comparisons to the 'DeepSeek episode' in early 2025, when the prospect of cheaper models briefly led the market to question the longevity of AI buildouts. Analysts note that while Kimi K3 is competitive on capability, it is not yet cost-competitive per task. This nuance complicates the narrative that cheaper alternatives will immediately reduce demand for expensive AI infrastructure, a key driver of the semiconductor sector's growth.

Despite these concerns, capital expenditure budgets at the largest technology companies continue to rise, and analysts maintain that the broader bull case for AI infrastructure remains largely unchanged. The persistent investor flows into semiconductor ETFs suggest that the current drawdown is being treated as a buying opportunity rather than a signal that the sector's growth cycle has ended. The market appears to be discounting the immediate impact of open-source developments on hardware demand, focusing instead on the continued expansion of data centre capabilities.