Storage continues to rise, but smart money is starting to position in this "flash memory chain"

Original Author: David

Introduction: SanDisk surged about 40x in 16 months post-IPO; A-share Longsys' Q1 net profit jumped 26x YoY... Storage is the hottest track in 2026, bar none. But since June, three giants—AMD, NVIDIA, and SanDisk—have done the same thing almost simultaneously:

Finding ways to use less expensive memory (DRAM) and offload tasks to cheaper flash storage (NAND). The leading stocks for this "flash replacement" undercurrent have already taken off, but the truly unpriced opportunities might lie in its upstream and downstream sectors.

Understanding AI's Constraint: The "Memory Tax"

A few numbers are enough to show how intense this storage rally has been so far.

SanDisk (SNDK), spun off from Western Digital in February 2025 with an IPO price of ~$38, reached around $2,000 by mid-June 2026—a nearly 40x increase in 16 months, with a P/E ratio of ~69x; not to mention Micron.

On the A-share side, Longsys reported a Q1 2026 net profit of 3.862 billion yuan, up 2644% YoY; GigaDevice's Q1 net profit was up 522% YoY, hitting its historical high with a daily limit up on June 17. The previous market consensus seemed to be just one sentence:

AI is so starved for storage, shortages will last until 2028, buy storage stocks and they'll go up blindly.

But while everyone celebrates the "shortage," the most influential companies are quietly planting landmines under this narrative.

On June 15, AMD announced the acquisition of a company called MEXT, whose core technology uses AI to make flash memory "pretend" to be DRAM;

Earlier, at CES and GTC in early 2026, NVIDIA launched CMX, moving the most memory-intensive data during AI inference to a flash layer; even earlier, in February, SanDisk and SK Hynix jointly initiated a new standard called HBF, aiming to fit flash memory into packaging previously reserved for high-end memory (HBM).

Seeing these three events together, the direction is identical:

Adding an intermediate memory layer for AI that's "cheaper than DRAM and faster than hard drives," reducing the need to pay for expensive DRAM. Overseas thematic investment research firm Citrini Research even coined a term for this phenomenon: "The AI Tax."

To understand this term and find investment targets accordingly, you first need to distinguish between AI's two types of "memory."

One is volatile memory, or DRAM, and its highest-end form, HBM (High Bandwidth Memory, used directly alongside GPUs). It's fast—the GPU can grab data instantly—but extremely expensive and has limited capacity.

The other is flash memory, or NAND—the solid-state drive (SSD) in your computer. It's cheap and offers large capacity, but it's slower. An imprecise but useful analogy: DRAM is like files spread on your desk for immediate access, while NAND is like goods in a basement warehouse—cheap and plentiful, but it takes time to fetch them.

As AI has evolved over the past two years, it's clear that the "desk" isn't big enough, and it's prohibitively expensive.

TrendForce data shows that DRAM contract prices rose over 90% QoQ in Q1 2026; Citigroup predicts average DRAM prices will increase 88% in 2026, and NAND by 74%. The root cause of the price hikes is AI:

NVIDIA's GPUs need data feeding; HBM is fast and expensive, so HBM consumes an increasing share of DRAM production capacity.

Data cited by Citrini shows that HBM's share of total DRAM wafer output rose from 2% in 2020 to ~21% in 2025, and is expected to reach 25% in 2026... meaning a quarter of memory production capacity is taken up by HBM, leaving less and pricier supply for others.

That's the origin of the "Memory Tax."

For AI to run fast, it's forced to pay an increasingly high "tax" for scarce, expensive memory. The tax is too heavy, naturally prompting efforts to avoid it. The only way to avoid it is to shift some of the workload originally burdening DRAM onto cheaper NAND flash.

The three actions by AMD, NVIDIA, and SanDisk mentioned earlier are essentially different paths for "tax avoidance." But the shared goal is to add an intermediate memory layer for AI that's "cheaper than DRAM and faster than hard drives."

The investment significance lies precisely in this "shift." Every task shifted to flash memory represents additional demand for the NAND supply chain. Storage leaders have already had their first rally based on "shortage-driven price hikes." "Flash replacing DRAM" is the second layer logic on top of the price increases.

This logic doesn't necessarily point to the already skyrocketing leaders, but rather to segments along the chain that haven't yet been priced for this new layer of logic. That's what's worth digging into.

Deconstructing the Flash Memory Supply Chain: Manufacturers Feast, Controller Makers Sell Shovels

The flash memory business, from wafer to SSD, is roughly divided into three layers. The further upstream you go, the more profitable and monopolistic it gets.

1. The most upstream are the NAND manufacturers, those producing the wafers themselves:

Samsung, SK Hynix (after the Kioxia merger), Micron, and SanDisk (spun off from Western Digital). They control capacity and profit the most during price upcycles.

1. The middle layer consists of module makers, who buy NAND die from manufacturers and package them into SSDs or memory modules for sale to end customers:

They don't manufacture wafers; they profit from processing and branding. Their earnings leverage can be even more dramatic than the manufacturers'. When NAND prices rise, their low-cost inventory instantly appreciates.

A-share companies like Longsys (Jiang Bo Long), Biwin Storage (Baiwei), and Demly (Demingli) belong to this layer. Longsys reported a Q1 2026 net profit of 3.862 billion yuan, up 2644% YoY; Biwin Storage grew 1567% in the same period.

However, this leverage is a double-edged sword. When NAND prices fall, inventory becomes a liability, and module makers are the first to feel the pressure in a downturn.

1. The most easily overlooked is the third layer: the controller chip:

An SSD contains not only NAND flash chips but also a "brain" responsible for managing data flow—the controller. It doesn't directly benefit from NAND price hikes, but as long as SSD shipments increase, controller demand rises.

Theoretically, this layer is the closest to a "pick-and-shovel seller" in the chain. The top two independent controller makers globally are Taiwan's Silicon Motion (SIMO) and Phison (8299.TW), while A-share Maxio Technology (688449) ranks third.

Currently, among these three layers, the manufacturers and module makers have already been fully priced in by the market's "price hike logic," with stock prices reflecting the current situation of shortages and rising prices.

"Flash replacing DRAM" is the second-layer logic layered on top of price hikes. It benefits not just price increases but, more importantly, the long-term amplification of SSD/flash memory shipment volumes.

The segments that should benefit most from this logic are precisely those driven by shipment volumes and haven't yet been swept up by the price hike rally, such as controller makers; and the newly generated incremental demand specifically created by HBF, which will be discussed next.

Truly Unpriced: The "Value Gap" of Controllers and the "New Cake" of HBF

Segments driven by shipment volumes, untouched by the price hike frenzy. Digging deeper reveals two areas.

First, the valuation gap of controller makers.

Maxio Technology (688449) is a prime example. It's the world's third-largest independent SSD controller maker, behind Taiwan's Silicon Motion and Phison. Its PCIe 5.0 controller is among the few in mass production domestically.

But as of April 2026, its market cap was still below its debut day, with its stock price significantly trailing module makers like Longsys and Demly... The reason likely isn't complex:

Controllers don't directly benefit from NAND price hikes. During the six-month NAND surge, capital flocked en masse to the module makers with the highest earnings leverage, leaving controllers on the sidelines.

However, this is precisely the divergence point between "price hike logic" and "shipment volume logic," in my view. NAND price hikes benefit manufacturers and module makers holding inventory; controllers don't share in that. But flash replacing DRAM drives a long-term amplification of SSD shipments. Every additional SSD sold requires one more controller.

If this line of reasoning holds, the beneficiaries are those driven by shipment volumes, not price hikes. Controllers are purer plays on this theme.

There are three key players in this layer:

Silicon Motion (SIMO) (US ADR): The global leader in independent controllers, with over 30% market share in consumer SSD controllers.

Phison (8299.TW) (Taiwan Stock): The global #2 independent controller maker; Kioxia's custom controllers come from them.

Maxio Technology (688449) (A-Share): The global #3 independent controller maker, the most technologically advanced domestic controller maker, and the one with the largest valuation gap.

However, the risks must be clearly stated. The controller market isn't a high-moat sector; there are many domestic players, and price wars are ongoing. Public data shows Maxio's R&D expense ratio is as high as 36%-38%, continuously squeezing profits. Being #3 globally doesn't guarantee high margins.

Second, the "New Cake" created by HBF.

First, what is HBF?

HBM is fast and expensive, also consuming a quarter of DRAM capacity. So SanDisk and SK Hynix devised a solution: stack NAND flash to create a replacement layer similar in form to HBM but with 8 to 16 times the capacity at a fraction of the cost. This is HBF (High Bandwidth Flash).

It doesn't compete with HBM's role. It acts as a "high-capacity warehouse" alongside HBM, specifically handling AI inference data that "doesn't fit in HBM, yet isn't worth relegating to cold storage."

The manufacturing process for HBF involves TSV (Through-Silicon Vias, creating vertical connections by drilling holes in chips) to stack multiple layers of NAND and then bond them together, a process similar to making HBM. This process will drive demand for advanced packaging, assembly & testing, and specialized materials. Technically related targets include:

JCET Group (600584), Tongfu Microelectronics (002156) (A-Share): The top two domestic assembly & testing leaders. The stacking and bonding processes used in HBF fall within their capabilities.

Wah Lee Industrial (688535) (A-Share): The only domestic company mass-producing GMC, a core material for HBM packaging. Its technology is applicable to compatible HBF processes.

However, this area seems more like unrealized expectations. Note the following points.

First, HBF is not yet in mass production. SanDisk's timeline is sampling in H2 2026 with initial equipment shipments in early 2027. Currently, all "benefits" are expectations with zero revenue impact on financials.

Second, the market size isn't as large as some imagine. According to forecasts cited by SK Hynix, the HBF market could reach ~$12 billion by 2030, compared to ~$117 billion for HBM in the same period. HBF is merely a complementary layer, not a disruptor.

Third, the A-share market has already seen a proliferation of "HBF concept lists", frequently including companies like E-shyton, PhiChem, ACM Research, and Quick. Most of these companies are only "theoretically possibly related," lacking actual HBF-related orders or process verification disclosures—classic cases of capitalizing on a concept.

These are different from truly connected companies like JCET and Wah Lee, where the "process connection is real." They must be evaluated separately.

Therefore, these two areas can be seen as near-term and long-term narratives under the same investment theme.

Controllers are a "value gap" that is "shipping now, but whose valuations haven't yet reflected the replacement logic," a solid foothold. HBF incremental demand is a "long-dated option with a compelling story, but monetization waits until 2027 or later," a more speculative bet with significant concept stock noise.

One Chart to See the Whole Market: Along the Supply Chain, Where Are the Targets, and Are They Expensive?

Summary of the deconstructed segments into one map.

Based on actual geographic distribution, this chain is concentrated in four markets: NAND manufacturers are in the US, Japan, and South Korea; controllers are in US ADRs and Taiwan stocks; modules, assembly & testing, and materials are almost entirely in A-shares. Hong Kong stocks have no pure-play flash memory targets, so none are forcibly included here.

Use this map with one coordinate in mind:

The further upstream (manufacturers), the more monopolistic and beneficial, but also the most fully priced and expensive. The further downstream (modules, controllers, assembly & testing, materials), the leverage and certainty vary, with some not yet priced for the "replacement logic."

Risks and Variables: Short-Term Tailwinds, Long-Term Sword of Damocles over DRAM

The most common misinterpretation of the "flash replacement" thesis is conflating short-term and long-term dynamics. They actually point in opposite directions.

In the short term (2026-2027), the replacement phenomenon is far from large-scale realization. The storage supercycle is still