Bank of America’s semiconductor analyst Vivek Arya reiterated a Buy rating on Micron Technology on July 21, maintaining a $1,550 price target that implies roughly 66% upside from current levels. The core thesis: China’s AI ambitions aren’t a threat to Micron’s memory business. They’re a tailwind.
Why Chinese AI models need more memory, not less
The specific concern centered on models like Moonshot’s Kimi K3, a Chinese open-source AI system that has impressed the industry with its capabilities. The worry was straightforward: if China can build competitive AI models more cheaply, maybe the world doesn’t need as much expensive high-bandwidth memory.
Arya’s rebuttal is built on actual hardware specs. The Kimi K3 model requires approximately 1.4 TB of high-bandwidth memory per serving instance. That’s a staggering amount of HBM for a single deployment, and it suggests that as these models proliferate across data centers, demand for advanced memory products doesn’t shrink. It multiplies.
The cheap API pricing that Chinese AI companies have been offering comes from operational efficiencies, not from using less silicon. This distinction matters enormously for Micron’s business model. The company’s growth story hinges on selling HBM3E and the next-generation HBM4 products. If Chinese AI advancement required less memory, that would be a problem. If it requires the same amount or more, deployed across a larger number of instances, Micron’s addressable market just got bigger.
China’s domestic chipmakers aren’t the threat investors fear
Arya’s analysis suggests that the CXMT threat narrative doesn’t hold up under scrutiny. CXMT is focused on commodity DRAM production, the kind of memory that goes into everyday electronics rather than cutting-edge AI accelerators. The company is projected to begin initial HBM output by late 2026, but it lags significantly behind Micron in the kind of advanced memory technology that AI applications actually require.
Micron’s revenue share from China dropped from approximately 14% in 2023 to around 7% in 2025 as domestic competition bit into commodity memory sales. But Arya frames this as manageable, since the high-margin AI memory business, where Micron holds a technological lead, is where the growth story lives.
The buyback catalyst and cloud spending wave
Beyond the China thesis, Arya highlighted a potentially massive capital return story. After certain CHIPS Act restrictions expire around December 2026, Micron could engage in share repurchase activity reaching $50-60 billion per year.
The backdrop for this optimism is a projected $1.5 trillion in cloud and AI capital expenditures through 2027. Every new data center rack, every new GPU cluster, every new inference deployment needs memory, making companies like Micron direct beneficiaries of the buildout regardless of which AI model architecture ultimately wins.
What this means for investors watching AI infrastructure
The BofA note essentially argues that investors have been pricing in a China risk that doesn’t match the technical reality. Chinese AI models require massive memory allocations. Chinese chipmakers aren’t producing competitive HBM products at scale. And the overall trajectory of AI infrastructure spending continues to accelerate.
The risk to watch isn’t China’s current capabilities. It’s the pace at which CXMT and other domestic players close the technology gap on HBM production. Late 2026 is the projected timeline for initial output, and if that accelerates meaningfully, the competitive dynamics could shift faster than BofA’s model anticipates.