Memory prices fall, Citigroup significantly cuts Micron's target price!

Can improvements in AI inference efficiency potentially feed back into higher storage demand?

Citi said it has sharply lowered its target price for Micron Technology by 17% due to the recent noticeable softening in DDR5 DRAM spot prices, but it maintains a Buy rating and leaves all earnings forecasts unchanged, believing that the long-term logic behind AI-driven storage demand has not been fundamentally shaken.

According to the Chase Wind Trading Desk, in a research report published on March 31, Citi analyst Atif Malik lowered Micron’s target price from $510 to $425, a 17% decrease. As of the close on March 30, Micron’s share price was $321.80, still offering about 32% upside relative to the new target price.

According to report data, the spot price of mainstream DDR5 16GB DRAM has recently dropped by about 6%. This round of pullback is mainly attributed to the market’s concerns about Google’s TurboQuant technology. The technology is believed to potentially compress AI inference’s consumption of memory, which in turn has sparked worries in the market about the outlook for storage demand.

Micron and peers have already begun negotiations with hyperscale cloud computing companies for strategic long-term agreements spanning 3 to 5 years. The agreement framework covers locking in baseline procurement volumes, setting up prepayment arrangements, and a quarterly pricing adjustment mechanism based on market conditions, which is expected to provide effective support for contract prices.

DRAM spot prices under pressure directly trigger target price downgrades

Since the beginning of the year, mainstream DRAM spot prices overall have been on a downward trajectory, with the DDR5 16GB product showing particularly sharp weakness and falling by about 6% recently.

Based on this, Citi lowered the valuation benchmark for Micron from 6x to 5x the bottom-cycle forward P/E multiple, and using the expected peak earnings per share in 2027 as the basis, it arrived at a new target price of $425, which aligns with the historical valuation bottom range of 5x to 6x during prior DRAM upcycle periods.

It is worth noting that the firm keeps all of its financial forecasts for Micron unchanged: its core EPS forecast for fiscal year 2026 is $58.46, and for fiscal year 2027 is $94.55. The current share price implies an expected P/E ratio of about 5.5x for 2026 and about 3.4x for 2027, placing it at a relatively low level historically.

Long-term agreements provide structural support for contract prices

Although the spot market is under pressure, the downside risk to contract prices is relatively controllable.

The report indicates that Micron and its DRAM storage peers are in talks with hyperscale cloud computing companies regarding strategic long-term agreements for 3 to 5 years. The terms include locking in baseline purchase volumes, establishing prepayment mechanisms, and dynamically adjusting quarterly prices according to market conditions.

The above long-term agreement framework is expected to build structural support for contract prices. About 79% of Micron’s revenue comes from its DRAM business, making the trajectory of contract prices critical for its visibility into profitability.

TurboQuant shock is viewed as similar to DeepSeek; long-term may reverse or pull demand back

In response to the main driver of the recent spot price decline, the report conducted a targeted assessment of the TurboQuant technology.

TurboQuant is a model compression technology developed by Google’s research team. It optimizes the computation of KV cache (Key-Value Cache) through new quantization approaches, including PolarQuant technology and the QJL algorithm, thereby reducing memory usage during the AI inference process.

The report believes that TurboQuant’s impact on storage demand is similar to the earlier DeepSeek event: on the surface, efficiency-improving technologies reduce the compute and memory costs per single AI inference, but the lower usage cost will further release application volume, ultimately increasing the overall total demand for compute power and memory.

Looking at historical patterns, cheaper technologies often end up stimulating demand for more advanced technologies, and the AI sector is no exception. Based on this, the firm judges that the recent pullback in spot prices more reflects short-term market sentiment disturbances rather than a trend-level reversal in AI storage demand.