Key Takeaways:
Bank of America's Vivek Arya says the semiconductor sector's 11% third-quarter pullback is a seasonal reset, not a structural breakdown, and that memory stocks trading at 10 times forward earnings are severely undervalued relative to their growing share of AI infrastructure spending.
The Philadelphia Semiconductor Index surged 88% in the second quarter before retreating 11% in the third quarter, a decline Bank of America analyst Vivek Arya characterized as a "summer reset" in a July 6 report. He forecast a rebound in the autumn, arguing the correction aligns with historically weak seasonal patterns and does not reflect deterioration in AI demand.
"Memory is now 35% to 40% of cloud AI capital expenditure, two to three times its historical share, yet memory stocks trade at sub-par 10 times forward earnings," Arya, who reiterated a Buy rating on Micron Technology with a $1,550 price target, said. He called the pullback "a healthy reset, not a structural change in AI demand."
Global cloud and AI infrastructure capital expenditure is on track to approach $1.5 trillion by 2027, representing 40% to 50% growth from current levels, according to the report. Arya identified memory (Micron), computing (Advanced Micro Devices, Intel), semiconductor equipment (Applied Materials, Lam Research, KLA, Teradyne), optics (Macom Technology Solutions) and networking (Credo Technology, Marvell Technology) as sub-sectors poised to regain market leadership as visibility into 2027 cloud spending improves through the second half of 2026.
The report directly addressed investor anxiety over Chinese open-weight AI models, which have rapidly closed the gap with US frontier labs at significantly lower inference costs. Third-party benchmarks as of July 4 show US models from Anthropic and OpenAI still leading, but Chinese models now occupy eight of the top 16 positions. The highest-ranked is GLM 5.2, developed by Z.ai, a 750-billion-parameter model with a million-token context window.
Chinese open-source models pressure software margins but boost chip demand
Arya argued that the rise of Chinese models such as GLM, Kimi, DeepSeek and Qwen creates genuine pressure on AI software profit margins but acts as a tailwind for semiconductor demand. Lower-cost intelligence expands use cases and deployment breadth, ultimately driving higher demand for compute, memory, networking and power infrastructure.
"The bigger risk is in model economics, not semiconductor demand," Arya wrote. Nvidia is actively engaging with the open-source community, a move that helps broaden its hardware ecosystem and capture smaller-scale AI adopters who lack direct access to frontier labs, the report noted.
Memory valuations face a structural repricing opportunity
The report's strongest conviction centers on memory chips. Arya said the market underestimates the industry's transition toward long-term contracts and more predictable pricing models. Investor concerns about pricing sustainability, new supply and customer concentration have kept valuations compressed, but the report argues this view is outdated.
"Memory is evolving from a cyclical commodity to a strategic AI infrastructure component," Arya said. "Valuation multiples should expand to reflect that transition."
The bullish thesis gained support from multiple Wall Street firms on Monday. UBS analyst Nicolas Gaudois lifted DDR contract-pricing forecasts to 32% quarter-over-quarter in the third quarter of 2026 and 18% in the fourth quarter, calling the DRAM market undersupplied "until at least the second quarter of 2028." Citi added Micron to its 90-day upside catalyst watch, and JPMorgan strategist Mislav Matejka told clients the semiconductor upcycle is "not peaking anytime soon," with meaningful new supply unlikely before 2028.
Memory stocks rebounded sharply Monday after Thursday's selloff. SanDisk rose 5%, Western Digital gained 5% and Micron climbed 3%. The Roundhill Memory ETF advanced more than 6%. The moves followed a brutal Thursday session triggered by a report that AI startup Anthropic was in preliminary talks with Samsung to design custom AI chips, which traders interpreted as a potential competitive threat to US memory makers.
Not everyone shares the bullish view. "Big Short" investor Michael Burry has disclosed a short position in Micron based on a valuation bubble thesis, sitting directly opposite Bank of America's assessment. Micron trades at 22 times trailing earnings with a forward multiple of 7 times, while SanDisk carries a trailing price-to-earnings ratio of 60 times.
Two catalysts loom this week. Samsung reports second-quarter results on Tuesday, offering a critical read on high-bandwidth memory pricing and demand. SK Hynix is set to list on the Nasdaq on July 10, an event some traders expect could drive rotation out of Micron into the lower-priced HBM market leader.
For investors, the BofA report provides a framework for navigating the semiconductor pullback. If the "summer reset" thesis holds, memory stocks at 10 times forward earnings offer a valuation entry point before autumn catalysts — Samsung's earnings, SK Hynix's listing and improving 2027 capex visibility — potentially re-rate the group. If Samsung's results fail to validate the pricing thesis, the bears may gain fresh ammunition.
This article is for informational purposes only and does not constitute investment advice.
