Bank of America characterizes the semiconductor industry's recent price weakness as a transitory reset instead of a fundamental retreat in artificial intelligence demand. The brokerage noted that the Philadelphia Semiconductor Index (SOX) fell 11% in the third quarter following an 88% gain in the second quarter.
According to BofA, the current pullback is consistent with historically weak seasonal patterns for the sector. The firm recommends investors look to lower-beta semiconductor names to lead the market recovery in the near term - specifically naming NVDA, TXN, ADI, CDNS and SNPS as likely front-runners.
The bank projects a large expansion in global cloud and AI infrastructure capital expenditures, estimating spending will reach $1.5 trillion by 2027. That figure, BofA says, equates to a 40% to 50% year-over-year increase and is expected to be propelled by token expansion, broader agent adoption and infrastructure capacity constraints.
BofA also reports that hyperscale cloud operators continue to emphasize utilization and growth objectives over strategies that would prioritize depreciation-based optimization. This emphasis on scale and usage, the bank argues, supports the case for renewed demand across memory, compute, semiconductor capital equipment, optics and networking stocks as visibility into 2027 cloud budgets improves through the second half of 2026.
The emergence of Chinese open-weight large models from companies such as GLM, Kimi, DeepSeek and Qwen is another development highlighted by BofA. The bank notes these models have narrowed performance differences with U.S. frontier labs while providing lower inference costs. BofA views that dynamic as constructive for adoption; lower-cost intelligence should broaden usage and in turn raise demand for compute, memory, networking and power infrastructure.
Memory spending now represents a larger slice of cloud AI capital expenditures than in prior cycles, accounting for roughly 35% to 40% of such outlays, according to BofA. Despite the larger role memory plays in cloud AI capex, memory equities trade at about 10 times forward price-to-earnings ratios, the bank notes.
On individual coverage, BofA maintains a buy recommendation on Micron Technology (MU) with a price target of $155.
Overall, BofA frames the recent sector decline as a healthy reset that could clear the path for a resumed rally once spending visibility and supply-side dynamics align with ongoing growth in AI-driven workloads.
Summary
BofA sees the semiconductor pullback as temporary, points to seasonal weakness, expects lower-beta names to lead, forecasts $1.5 trillion in cloud and AI infrastructure capex by 2027, and highlights rising memory share of AI capex and positive implications from cheaper Chinese open-weight models.