Bank of America: Kimi K3 and other open-source large models are good for AI storage demand, reiterates "buy" rating on Micron (MU.US)

The latest research report from Bank of America Corp pointed out that the recent release of new generation open-source large models, including the Kimi K3 launched by Yue Zhi An Mian, further strengthens its long-term bullish logic on the AI storage market and Micron Technology, Inc. (MU.US). Boosted by the positive news, storage concept stocks surged collectively on Tuesday, with Micron Technology, Inc. (MU.US) rising over 8%, SK Hynix (SKHY.US) rising over 9%, and SanDisk (SNDK.US) rising over 10%. Analysts at Bank of America Corp led by Vivek Arya stated that the pricing of China's open-source large model APIs is highly competitive, with prices 5 to 350 times lower than Western models, but this more reflects a business model choice rather than a significant decrease in hardware costs. The analysts pointed out that although the new models have reduced GPU computing requirements through architecture optimization and inference efficiency improvements, the demand for high-bandwidth memory (HBM), DRAM, and other storage resources has not decreased but has increased with the continuous growth of model parameters and activation parameters. In addition, with each open-source model download, customers need to deploy the model themselves, thereby increasing the demand for HBM, DRAM, NAND, and other storage resources, which is not possible with closed-source models. Last week, China AI startup Yue Zhi An Mian, in which Alibaba Group Holding Limited Sponsored ADR (BABA.US) has invested, officially released the Kimi K3. This large model with 28 trillion parameters is described by the company as the world's largest open-source weighted model, with performance similar to Anthropic's latest flagship model Fable. Bank of America Corp believes that Chinese storage chip manufacturer Changxin Storage currently does not pose a substantial threat to Micron and reiterated a "buy" rating on Micron with a target price of $1550. The analysts stated that although Changxin Storage is actively expanding its capacity, accounting for only a low single-digit percentage to about 10% of the global DRAM wafer capacity, it mainly focuses on the consumer and standard DRAM markets and has not entered the high-end AI storage areas such as HBM3E and HBM4. Additionally, there is still significant uncertainty as to whether U.S. original equipment manufacturers (OEMs) can obtain government approval to purchase Changxin Storage products in the short term. Bank of America Corp also pointed out that Micron's stock buyback restrictions, due to government subsidies under the "Chip Act," are expected to expire around December 2026. With the restrictions lifted, the company's future annual free cash flow is expected to exceed $12-13 billion, and with a 40% capital return policy, an estimated $5-6 billion stock buyback can be implemented annually, equivalent to 5-6% of its current market capitalization of around $1 trillion. Regarding the market's concerns about China's low-priced AI model strategy, Bank of America Corp stated that this does not mean that AI infrastructure costs are decreasing simultaneously. For example, Tencent's Mixnet model API input price is only $0.06 per million tokens, while Anthropic's Claude Opus 4.8 is about $15; Kimi K3 charges about $3, but its performance is comparable to leading international models. However, the analysts emphasized that API prices more reflect business strategies rather than hardware costs. For example, each inference instance of Kimi K3 still requires approximately 1.4TB of HBM to run, while OpenAI's open-source model "oss-120b" has a parameter scale of only about one twenty-third of Kimi K3 and still requires approximately 63GB of model weight memory to run. The analysts stated that API prices are usually closely related to the model weight scale and the number of activation parameters, even for Chinese open-source large models. "Open-source" only means that the model weight is open, not a reduction in deployment costs; customers still need to purchase HBM, DRAM, NAND, and other storage hardware to complete local deployments. Furthermore, Bank of America Corp believes that China's AI model price advantage also comes from model architecture efficiency improvements and cost advantages. The analysts expect that China has a certain leading advantage in model architecture efficiency, and compared to overseas, it has approximately 1.5 to 2 times advantage in basic infrastructure costs such as electricity, labor, and land, with the remaining price differences possibly coming from national capital subsidies and financial support and free cash flow subsidies provided by cloud service providers such as Alibaba Group Holding Limited Sponsored ADR, Tencent, and Baidu Inc Sponsored ADR Class A (BIDU.US).