Mizuho highlighted a top memory pick for 2026 in a note Monday, arguing the company is best positioned to benefit from a shift in AI hardware demand that could reshape the market.

Analyst Jordan Klein said he sees Samsung as “the most compelling long idea” heading into next year, adding that the stock “could be a total home run next year.”

Klein pointed to a Digitimes report outlining the advantages of Groq’s language processing unit, or LPU, and the implications for high-bandwidth memory suppliers. 

According to the article, “LPUs significantly reduce data processing latency, enabling inference speeds faster than human conversational pace.” 

It added that Groq claims its LPU can execute large-language-model tasks “10 times faster than current solutions while consuming only one-tenth the energy.”

Mizuho believes this shift could broaden the AI memory market. While training workloads will still rely on HBM, “inference devices and on-device AI gadgets will increasingly utilize GDDR, LPDDR, and SRAM,” the report noted. 

Klein highlighted that the LPU’s design “enhances computing speed by leveraging high-speed memories like SRAM,” giving renewed attention to suppliers such as Cypress and Renesas.

The note also said Nvidia’s upcoming Rubin CPX chip will “forego costly HBM in favor of more affordable and readily available GDDR,” where Samsung holds clear leadership in next-generation GDDR7.

With generative AI expanding into video and high-resolution imagery, Klein pointed to rising demand for enterprise-grade QLC solid-state drives, saying Samsung stands to benefit across multiple segments as memory prices strengthen into 2026.