SerpApi.Org
Chip stocks are red hot again as investors come around to the idea that something once viewed as a threat to the artificial-intelligence trade might actually be a further spark.
A key debate in the market has been around the impact of Kimi K3, an open-source AI model from China-based Moonshot AI, which performs competitively to leading, closed U.S. models on certain benchmarks despite financial and technological limitations. Its ascent has revived questions about whether other model developers are vastly overspending on AI hardware and computing power.
Nancy Tengler, CEO of Laffer Tengler Investments, compared the market reaction to the selloff spurred by China’s open-source DeepSeek model last year that created “a tremendous opportunity” for investors who bought the dip.
Read on: Why cheap Chinese AI models could actually be a boon for Nvidia, Micron and other chip stocks
Shares of memory makers Micron Technology and Sandisk surged 12.2% and 14.4%, respectively, on Tuesday. SK Hynix’s American depositary receipts jumped 13.8%.
Storage makers Western Digital and Seagate Technology rose 12.5% and 11.1%, respectively.
The moves higher follow a sharp selloff among chip stocks last week that dragged the PHLX Semiconductor index into a bear market.
Brian Mulberry, chief market strategist at Zacks Investment Management, said that chip-stock valuations “have reached a point where you can once again get involved and participate in the upside potential” from expectations for earnings per share growth. He added that earnings results from manufacturers outside of tech are “helping boost confidence” in chip stocks.
Bank of America analyst Vivek Arya said China’s recent releases of open-source models support his “bullish thesis on memory” — and on Micron’s stock.
Chinese companies are charging developers much less than Western companies to access and use their open models, Arya said in a Monday note. He views that “as reflective of business-model choices,” and not of the cost of hardware.
The most important point, in his view, is that although Chinese model makers have been able to reduce compute intensity using different techniques for efficiency, the models still “require the same or more memory as their model weights and active parameters increase.” A model’s parameters are the variables it uses to learn and recognize patterns during training, and they serve as a sort of memory. Kimi K3 has 2.8 trillion parameters, 50 billion of which are active.
Additionally, Arya said every download of an open-weight model leads to another customer’s need for memory to run it on their hardware — something that isn’t the case with closed models.
Chinese models also have an advantage over U.S. ones when it comes to costs, Arya said, pointing to efficiency in architecture and lower prices for physical infrastructure such as electricity, labor and land in China. He attributed other cost differences to government subsidies and free cash flows at Chinese cloud-service providers such as Alibaba and Baidu
Meanwhile, Arya said he doesn’t see a threat to Micron from Chinese memory-chip maker ChangXin Memory Technologies which is expanding capacity of dynamic random-access memory.
See more: How a homegrown Chinese chip maker became the memory industry’s biggest wild card
Unlike Micron, SK Hynix and Samsung Electronics, the world’s leading producers of DRAM, CXMT does not focus on high-bandwidth memory, which is in high demand from AI chip makers like Nvidia, he said. The Chinese company’s ability to serve U.S. customers also remains unclear, Arya added.
Hendi Susanto, a portfolio manager at Gabelli Funds, said that although memory companies have seen share prices rise significantly in recent months, “the investment case remains compelling,” as demand is expected to continue outpacing supply through next year. That means “favorable pricing and strong earnings” can continue, he said.
The PHLX Semiconductor Index gained 5.2% on Tuesday, for its best day in just over a month, according to Dow Jones Market Data.
