The reflex read on chips right now is caution. The sector gave back part of a powerful spring rally just as it entered its weakest seasonal stretch of the year, but the buildout driving it is only accelerating.

The PHLX Semiconductor Index — as tracked by the iShares Semiconductor ETF (SOXX) — fell 11% since the start of the third quarter, historically its softest window, after an 88% second-quarter surge.

Yet, global cloud and AI infrastructure spending is still on track to approach $1.5 trillion by 2027, up 40 to 50% year over year, according to Bank of America's analyst Vivek Arya.

The analyst continues to expect that the long-term winners of the artificial intelligence infrastructure boom remain the companies most leveraged to AI capital spending.

"History suggests periods of consolidation are often followed by renewed momentum as investors regain confidence in the next leg of earnings and capex growth," Arya said.

The $1.5 Trillion AI Buildout Remains Intact

As visibility into 2027 spending improves during the second half of the year, Arya expects leadership to shift back toward companies directly tied to AI capital expenditures, including Advanced Micro Devices (AMD), Applied Materials (AMAT), Lam Research (LRCX), Micron Technology (MU), MACOM Technology Solutions (MTSI), Credo Technology Group (CRDO), and Marvell Technology (MRVL).

The bank indicates that hyperscalers remain focused on maximizing AI utilization rather than cutting infrastructure spending, suggesting that demand for chips powering AI data centers should remain robust.

Why Micron Is BofA's Top Pick

At the center of Arya's thesis is Micron, which he calls one of the market's biggest AI mispricings.

Memory now accounts for roughly 35% to 40% of AI cloud capital spending, more than double historical levels.

Yet memory stocks continue to trade at relatively modest valuation multiples because investors remain concerned that the industry's pricing will eventually revert to its traditional boom-and-bust cycles.

Bank of America disagrees.

Memory is moving "from a cyclical commodity to a strategic AI enabler," analyst Vivek Arya said.

Arya believes long-term supply agreements between memory suppliers and hyperscale customers are fundamentally changing the industry's economics by making pricing more durable and revenue streams more predictable.

That, in turn, could allow memory companies to evolve from cyclical commodity businesses into strategic AI infrastructure providers deserving higher valuation multiples.

The firm reiterated its Buy rating on Micron and maintained a $1,550 price target, implying roughly 59% upside from current levels.

Open-Source AI Isn't a Threat to Chip Demand

Arya also dismissed concerns that increasingly capable Chinese open-weight AI models could reduce semiconductor demand.

While lower-cost AI models may pressure software economics, Bank of America argues they should ultimately expand AI adoption by making inference cheaper and accelerating deployment across industries.

"The bigger risk is to model economics, not semiconductor demand," the report said, as broader AI usage would require more compute, memory, networking and power infrastructure over time.