To The Moon

Citron’s argument is not new in memory cycles, but the timing is interesting. Every memory upcycle eventually attracts a “supply illusion” thesis because historically, memory has been the most cyclical segment in semiconductors. The key question now is whether this cycle still behaves like the old PC and smartphone-driven cycles, or whether AI has structurally changed demand.

---

1. What the “supply illusion” thesis is really saying

Short sellers are likely arguing three points:

1. Front-loaded AI orders Hyperscalers may be over-ordering storage and memory to avoid shortages, creating temporary demand spikes rather than sustainable consumption.

2. Capacity eventually catches up NAND historically swings from shortage to oversupply quickly once fabs ramp output.

3. End-demand outside AI remains soft PCs, smartphones, and consumer electronics are not in a strong recovery, meaning traditional demand pillars are weak.

If true, pricing power fades once inventory normalises, compressing margins sharply. That has happened repeatedly in past memory cycles.

---

2. Why this cycle may be structurally different

The counterargument, and why the market has re-rated storage names, lies in AI workload behaviour.

AI data centres fundamentally change storage usage:

(a) Data intensity is exploding

Training and inference require:

massive datasets

checkpoint storage

vector databases

persistent model storage

Unlike consumer devices, AI workloads continuously generate and retain data.

(b) AI shifts storage from consumer-driven to infrastructure-driven

Historically:

NAND demand followed gadget sales.

Now:

Demand follows data centre buildouts, which are multi-year capital projects.

Infrastructure spending tends to be stickier than consumer upgrades.

(c) Storage is becoming a bottleneck again

GPU utilisation depends on feeding data efficiently. Slow storage reduces expensive GPU productivity. Hyperscalers increasingly treat storage as part of the compute stack rather than a commodity add-on.

That changes pricing dynamics.

---

3. Does Citron’s short thesis materially weaken the sector outlook?

It introduces cycle risk, but does not necessarily invalidate the structural bull case.

Think of two overlapping realities:

Short-term Long-term

Inventory swings possible AI data growth secular

Pricing volatility Higher baseline demand

Sentiment-driven pullbacks Structural TAM expansion

Memory stocks rarely move in straight lines even during strong secular trends. Sharp corrections are normal.

---

4. What investors should actually monitor now

Instead of reacting to the headline short call, watch leading indicators:

1. NAND pricing trends Spot price stabilisation matters more than shipment volume.

2. Hyperscaler capex commentary If AI data centre expansion continues, storage demand persists.

3. Inventory levels Rising inventories across suppliers would support the short thesis.

4. Enterprise SSD demand AI inference deployment should drive sustained enterprise storage upgrades.

If pricing holds despite capacity additions, the “supply illusion” argument weakens quickly.

---

5. Does this change the broader sector view?

Not fundamentally, but it changes the risk profile.

The sector likely transitions from:

early-cycle re-rating (multiple expansion) → to

execution phase (earnings must justify valuations).

In other words, upside becomes more selective and volatility increases.

---

Bottom line

The short thesis highlights a real historical risk: memory has always been vulnerable to oversupply. However, AI may be raising the floor of demand rather than eliminating cyclicality.

The storage rally is probably not fragile, but it is also unlikely to be smooth.

A useful way to frame it:

If AI infrastructure spending continues, pullbacks become cyclical corrections within a secular uptrend.

If hyperscaler spending slows, memory stocks will react disproportionately because operating leverage cuts both ways.

The debate now is not whether demand exists, but whether the market has temporarily priced peak conditions too early.