Semiconductor equities came under pressure following Google’s disclosure of TurboQuant, a technique the company says can reduce memory requirements for AI inference by up to six times while maintaining accuracy. The market reaction was most pronounced in memory-related names, with Micron Technology and the broader Philadelphia Semiconductor Index (SOX) moving lower on the news.
Despite the near-term selloff, analysts at Bank of America contend that efficiency gains like TurboQuant are not the principal determinant of AI hardware demand. Instead, BofA emphasizes that AI capital expenditure - the amount companies allocate to servers, chips and related infrastructure - remains the key barometer for long-term demand in semiconductors supporting AI workloads.
BofA notes that compression and quantization approaches similar to TurboQuant are not entirely new; the bank points out that Google has published related technologies over the past 18 months. Over that same period, Google raised its calendar year 2026 capital expenditure outlook significantly - to roughly $180 billion, an increase of about 100% year-over-year versus a consensus projection of approximately $127 billion, which represented about a 38% year-over-year increase. The analysts use that divergence to underline their view that announced efficiency improvements have not curtailed cloud and AI infrastructure spending plans.
The bank said it continues to view demand for AI memory as robust. Rather than expecting a wholesale drop in memory consumption, BofA anticipates that a six-fold improvement in memory efficiency is more likely to translate into higher model accuracy and longer context windows - outcomes that could sustain or even expand demand for higher-performance memory in production AI systems.
On a broader timeline, the bank projects global AI-related capital expenditure to reach around $1.4 trillion by calendar year 2030, and it models a conservative capex intensity in the 25-30% range over time. That forecast underpins the firm’s subsector preferences and its stock-level recommendations within U.S. semiconductors.
BofA’s top U.S. semiconductor subsector rankings and selected stock picks
1) AI Compute - Bank of America places AI compute at the top of its preference list and highlights Nvidia, Broadcom and Advanced Micro Devices as key names. The firm notes industry activity that reinforces its view: European AI provider Mistral raised $830 million in debt to acquire thousands of Nvidia chips for a new data center, and sell-side firms including Wolfe Research and Cantor Fitzgerald have reiterated positive ratings on Nvidia citing ongoing AI momentum.
Broadcom, together with partner Carahsoft, secured a five-year, $970 million blanket purchase agreement with the U.S. Defense Information Systems Agency aimed at consolidating software contracts, and the company has begun volume shipments of its Tomahawk 6 switch chip. Advanced Micro Devices announced a collaboration with Celestica to develop the Helios rack-scale AI platform.
2) Semicaps (semiconductor capital equipment) - Semiconductor manufacturing equipment names sit second on BofA’s list, with Lam Research, Applied Materials and MKS Instruments singled out as favorite picks within the subsector.
3) AI Networking - The bank ranks AI networking infrastructure third, underscoring the importance of high-performance interconnects and switches in data center buildouts. Marvell Technology, Credo Technology and Macom Technology Solutions are listed as preferred stocks. Marvell has unveiled a new 260-lane PCIe 6.0 switch aimed at AI data centers and announced a long-term collaboration with Mojo Vision to support commercialization of a micro-LED platform.
4) Memory - Although BofA places the memory subsector fourth, the bank continues to see durable demand even as near-term pressure has emerged. Micron Technology is identified in this group. The analysts flag existing peak margin concerns for memory makers, and note that Micron trades at the low end of its historical five- to ten-times out-year price-to-earnings multiple based on a consensus $75-100 earnings-per-share range for calendar year 2027.
Top-five completion and other subsectors - Analog Devices and Allegro MicroSystems round out the top five positions in BofA’s sector rankings. Analog Devices has attracted several price target increases from firms such as Cantor Fitzgerald, TD Cowen and KeyBanc following a strong quarter and outlook driven by demand across data center, automotive and communications end-markets.
Electronic design automation firms Cadence Design Systems and Synopsys are ranked sixth by BofA. Synopsys recently launched new chip design and verification products that incorporate technology from its recent acquisition of Ansys. Consumer-focused semiconductor stocks occupy the seventh position in the bank’s subsector preference order.
BofA’s framework rests on the premise that capital spending behavior among cloud and hyperscale operators is the dominant force shaping semiconductor demand for AI applications. While algorithmic and systems-level efficiency improvements receive attention and can influence implementation choices, the bank’s view is that material changes in AI capex plans - rather than incremental efficiency gains - will be the primary driver of sustained market demand for chips, memory and related infrastructure.