To The Moon

$S&P 500(.SPX)$ and $Dow Jones(.DJI)$ both closed at record highs. The Dow rose 0.99% to 49,462.08, historically breaking through the key psychological level of 49,000. The S&P 500 also set a new record, gaining 0.62% to close at 6,944.82.

As January goes, so goes the year. When January closes positive, the S&P 500 is higher 89% of the time, with an average gain of 17% and an average maximum drawdown of 10.5%. When January is negative, average returns fall to -1.8%, with only a 50% hit rate and deeper market drawdowns.

Beyond the January Effect: Capital Is Looking at Memory

Capital is accelerating its rotation away from extremely crowded AI mega-cap leaders and spreading deeper into the supply chain. This “decentralized” rotation not only improves market breadth, but also triggers a broad-based revaluation across the entire semiconductor supply chain.

$SanDisk Corp.(SNDK)$ surged 27.56%, $Micron Technology(MU)$ jumped 10.02%, $Western Digital(WDC)$ soared 16.77%, and $Seagate Technology PLC(STX)$ climbed 14%—all posting double-digit gains.

SanDisk and Western Digital focus on high-speed flash storage; Micron is the leading player across HBM and DRAM; Seagate represents long-term memory for AI workloads.

At CES, NVIDIA unveiled its “Inference Context Memory Storage Platform,” sending a very clear signal: AI has entered a new phase of long context + intelligent agents. Inference is no longer just about how fast GPUs can compute, but whether memory can be fast enough, cheap enough, and scalable enough to keep up.

In the past, AI memory roles were clearly divided:

• HBM sat right next to GPUs for ultra-fast computation—but at a high cost.

• DRAM served as working memory for data turnover.

• NAND (SSDs) acted as remote warehouses for long-term storage.

Now the challenge is different. When models need to process hundreds of millions of tokens and maintain “long-term memory,” the KV cache generated during inference can balloon to the terabyte level. Storing everything in HBM is prohibitively expensive; pushing it to remote storage leaves GPUs idle, wasting compute power.

What NVIDIA has done this time is essentially create a new intermediate memory layer between GPU memory and hard drives. Through the BlueField-4 DPU, enterprise-grade NAND flash is integrated directly into racks and positioned closer to GPUs. SSDs—once purely backend components—are now participating in real-time inference for the first time.

By pulling NAND into the core of AI inference at CES, NVIDIA has moved storage from the background to the front line. The entire storage sector has been revalued in one stroke.

Historically, storage stocks were treated as cyclical plays, because no one could accurately estimate how much storage an AI data center would need. Now, as storage becomes a standardized rack-level component, demand can be linearly modeled based on GPU rack counts for the first time. Once demand becomes quantifiable, valuations naturally rise.

At the same time, supply-demand imbalances provide additional catalysts. HBM is crowding out capacity, compressing supply of general-purpose DRAM and pushing prices into a sustained upcycle. NAND, meanwhile, is upgrading from “cheap, high-capacity storage” to a “quasi-memory layer,” driving a re-rating in both volume and pricing. This is why the market lifted flash, DRAM, and integrated storage companies simultaneously on the same day.

So the question now is:

1. Is it too risky to chase memory stocks at these levels?

2. Which name do you favor most in the memory space—SNDK or the all-around leader Micron?

3. Is it better to buy at the start of the year, or wait for the January effect to play out and add during earnings season?

4. Do you believe in the January effect?

Leave your comments to win tiger coins~