Japan Electronic Materials: Hidden HBM Play

I discovered this name few weeks back and told few people in close group about this being under-radar. In last few weeks it has almost gone up 50%, yet the valuations are just at 20x TTM P/E, so it is still worth discussing this.

The AI boom is increasingly constrained not by compute alone, but by memory, and this is likely to remain true for the foreseeable future.

The reason is straightforward: large language models rely on massive amounts of data and parameters that must be accessed by GPUs with extremely low latency. It’s not enough to have powerful compute; the computational components of an LLM must be fed data fast enough to keep the GPU fully utilized. If memory bandwidth becomes the bottleneck, the entire system slows down.

This is why High Bandwidth Memory (HBM) has become one of the hottest segments in the semiconductor industry. HBM is essentially ultra-high-performance RAM designed to sit very close to the GPU, enabling extremely fast data transfer. As model sizes grow and inference workloads scale, the demand for faster and denser memory rises sharply.

Beyond HBM, broader storage infrastructure is also becoming increasingly critical. The proliferation of AI models means exponential growth in data, training datasets, fine-tuned model checkpoints, embeddings, logs, user prompts, video content, and contextual memory layers. All of this must be stored, retrieved, and processed efficiently.

In short, while compute gets most of the attention, memory, both high-speed memory like HBM and large-scale storage systems, remains one of the most important structural constraints in the AI ecosystem.

When investors talk about the AI revolution, the conversation usually starts and ends with the giants: Nvidia for the chips and SK Hynix or Micron for the High Bandwidth Memory (HBM).

But there is a critical “invisible” step between manufacturing a silicon wafer and shipping a finished AI module. Before an HBM stack can be sold, it must be tested. And as chips become exponentially more complex, the equipment required to test them is becoming exponentially more expensive.

Enter Japan Electronic Materials (JEM). While they aren’t a household name, they are the gatekeepers of quality for the global memory market. Here is why this quiet Japanese firm has become a critical pivot point in the AI supply chain.

---

1. What Does JEM Actually Do?

At its simplest, JEM manufactures probe cards.

To understand why this matters, you have to understand the physical problem of chip manufacturing.

• The Tester: The machine that checks if a chip works, is the size of a large refrigerator.

• The Chip: The contact points on a modern AI chip are microscopic, thinner than a human hair.

You cannot plug a refrigerator-sized machine into a microscopic chip. You need a bridge.

The Probe Card is that bridge. It is a high-tech interface, roughly the size of a dinner plate. On the top, it connects to the massive testing machine. On the bottom, it holds thousands of microscopic needles (probes) that physically touch the silicon wafer.

Think of the testing machine as a doctor and the probe card as the stethoscope. The doctor cannot diagnose the patient without the instrument touching them. JEM makes the high-precision stethoscopes required for the world’s most complex silicon patients.

2. The Tailwind: It’s Not Just Volume, It’s Complexity

The investment thesis for JEM isn’t just that more chips are being made. It is that the type of chips being made requires significantly more expensive probe cards.

JEM’s business is split into two worlds: Volume and Value.

• The Volume (Standard Memory): This is your standard DRAM for PCs and NAND for phone storage. It keeps the lights on.

• The Value (HBM & AI): This is the explosion.

An HBM (High Bandwidth Memory) chip is a beast compared to standard memory. A standard DRAM chip might have 40,000 pins to test. An HBM stack can have over 150,000 pins.

Because of this density, chipmakers cannot use cheap, older-style needles. They must buy JEM’s high-end MEMS (Micro-Electro-Mechanical Systems) cards.

• Standard Card Price: ~$30k – $50k

• HBM Card Price: ~$100k – $200k+

JEM isn’t just selling more units; they are selling units that are 3x to 4x the price of their legacy products.

3. The “Burn-In” Factor: Why Customers Are Panic-Buying

The most fascinating driver for JEM is the inefficiency of HBM manufacturing.

In standard memory, you test a chip once; if it works, you ship it. HBM is different. HBM involves stacking 8 or 12 dies vertically. If you stack 11 good chips on top of 1 bad one, the entire stack is ruined.

To prevent this, manufacturers like SK Hynix are enforcing “Known Good Die” (KGD) standards. They are stress-testing (burning in) every single layer intensely before stacking.

This has two effects:

1. Higher Consumption: “Burn-in” physically wears out the probe needles faster.

2. Recurring Revenue: Probe cards are consumables. They dull, bend, and break. The more aggressively manufacturers test to save their HBM yields, the faster they burn through JEM’s cards.

4. The Memory Pure-Play

The probe card market is an oligopoly dominated by handful of players: FormFactor (USA), Technoprobe (Italy), and JEM (Japan).

However, JEM is unique because it is a Memory Pure-Play.

• FormFactor is heavily exposed to Logic (Intel/CPUs).

• Technoprobe is focused on Mobile/Logic.

• JEM is almost entirely reliant on the Memory market.

In previous years, this lack of diversification might have been a risk. In the age of AI, where HBM is the single biggest bottleneck in the supply chain, this concentration is JEM’s greatest asset. They are more sensitive to the HBM supercycle than any of their competitors.

Conclusion

JEM represents a classic “pick and shovel” play in the AI gold rush. They do not design the chips, and they do not run the fabs. They simply provide the consumable equipment that is mandatory for the production of HBM.

As long as HBM remains the bottleneck for AI accelerators, and as long as chipmakers are terrified of stacking “bad” dies, the demand for JEM’s high-margin MEMS cards will continue to outpace the broader market.

In recent quarter they have reported 70% revenue growth and 3x jump in operating profit. The stock trades at 20x TTM P/E. While Formfactor and Technoprobe are at 2x-3x valuations.

The key risk to watch out here is entry of new players in probe cards market. TSE from Korea has recently entered and Korean memory giants Samsung & SK Hynix might shift to local player if it proves to be credible. I am also watching them and might create a basket of two positions.

---

Disclaimer: This post is for informational purposes only and does not constitute financial advice. Always do your own due diligence.