EG Industries (8907.KL): The Malaysian Factory Building AI Optics, and the Customer Who Could Take the Work Back

Every conversation about AI infrastructure eventually reaches the same unglamorous problem. The GPUs are fast; getting data between them is the bottleneck. Inside a modern data centre that data travels as light, through optical transceivers, and somebody has to build those modules by the hundreds of thousands.

One of the somebodies is $8907.KL, EG Industries, a Malaysian contract manufacturer that builds AI data-centre optical transceivers and co-packaged optics (CPO) for US customers. Fully diluted, it trades at 20 times core earnings. Its CEO has bought roughly US$1.2 million more stock since 15 September. The factor framework scores it StockRank 81 of 100, and the most interesting number on its scorecard is the weak one.

Why AI Data Centres Keep Buying Faster Light

The product cycle is simple to describe and relentless to live through. An 800G module moves 800 gigabits a second over eight lanes of 100G each. A 1.6T module doubles each lane to 200G. Every new GPU generation needs a faster module, so the industry upgrades not when the old modules wear out but when the chips beside them change.

There is a second force, and it is political. Chinese-made modules pay a 25 percent US tariff. Much of the optical module industry's design and manufacturing depth is Chinese, and its biggest buyers are American hyperscalers. That combination gives a Chinese module maker a strong reason to have the physical work done somewhere else, and Malaysia, with decades of electronics contract manufacturing behind it, is an obvious somewhere else.

One Customer, a Third of Revenue, and a US$242 Million Order

EG's largest customer accounted for about a third of FY2025 revenue, and is likely Cambridge Industries Group (CIG), a Chinese optical module maker with a Malaysian base. In May this year it placed a US$242 million order. By mid-2026 EG's yield on 800G modules had reached 80 percent, the kind of operational milestone that separates a line that ships from a line that is still learning.

A third of revenue from one customer is concentration by any definition. Whether that concentration is a risk or a moat depends on how hard it is for the customer to leave.

The Moat Is Requalification

Leaving EG means requalifying a generation.

An optical module is qualified as a whole: the design, the components, the assembly process and the line that builds it, tested against the customer's and the end buyer's specifications. Moving production to another manufacturer means repeating that qualification, and while it is repeated the new line ships nothing. Moving a qualified line costs months of output, in a market LightCounting expected (January 2026) to grow about 60 percent this year. In a market growing that fast, months of lost output are not a cost saving deferred, they are share handed to whoever kept shipping.

In Hamilton Helmer's 7 Powers this is Switching Costs, and it is the same mechanism we described in the Malaysian solder ball business at Mi Technovation: once a supplier is qualified into the recipe, the barrier to replacing it is the risk of changing, not the price of staying. Our explainer on what an economic moat actually is sets out why that kind of barrier tends to hold.

The Customer Who Could Take the Work Back

The obvious comparison is Fabrinet (FN), the largest non-Chinese option for this kind of work. At about US$16.6 billion and 35.5 times earnings, it is a different order of company from EG, and the market prices it as the established answer.

The less obvious comparison is the more important one. CIG, at roughly US$9 billion, is a second-tier Chinese maker that already has its own Malaysian base. It does not need EG to reach Malaysia. It can decide to bring manufacturing in-house and ramp down the work it farms out, and if it does, the requalification cost that protects EG from a rival contract manufacturer offers far less protection against a customer that already runs the process in the same country.

That is the uncomfortable shape of EG's position: between a giant that can win the work on scale, and a customer that can take it back on control.

Diversification, and the 30 Percent Catch

The diversification in progress is likely to be Micas Networks, an American networking company, through a US$100 million letter of intent for co-packaged optics. CPO moves the optics onto the same package as the switch silicon, the next step after pluggable modules, and a qualified position there would matter.

The catch is in who signed it. The LOI was signed by Genetronic, an EG subsidiary, and Genetronic is 45.7 percent owned by Micas. EG holds only 30 percent. It is recognised as a subsidiary on the basis of control, not ownership.

Follow the accounting through. Because EG controls Genetronic, it consolidates the subsidiary: Genetronic's revenue will appear in EG's revenue line in full. But EG's shareholders own 30 percent of Genetronic's profit, and the other 70 percent belongs to non-controlling interests, most of it to Micas, which is simultaneously the customer placing the order. A reader who sees CPO revenue arrive in EG's top line should look straight past it to the profit attributable to EG shareholders, because the two will tell very different stories.

The Dilution Arithmetic

In its fourth-quarter FY2026 results the CEO said “the group's transformation is delivering results.” Mass production of 1.6T modules is due by the end of 2026. The capital structure is where the excitement needs a discount:

• Warrants at RM0.90 add up to 48 percent more shares by December 2028. Fully diluted, the core P/E rises from 15.7 times to 20.2 times. The 20 times in the headline is the diluted figure, and it is the honest one to use.
• On FY2027 broker forecasts, fully diluted EG trades at about 12 times, against Fabrinet at 25.5 times forward. Half the multiple of the established player, for a company that is smaller, more concentrated, and exposed to its biggest customer's choices.
• A 0.3 percent dividend yield and no buybacks. The cash is going into capacity, which is defensible in a market growing this fast.
• The CEO has bought about US$1.2 million since 15 September. One insider buying with personal money is a real signal, if a single one. Our note on reading insider transactions explains why a cluster of buyers carries more weight than any one of them.

The MoatMap Scorecard: Q49 V58 M69, StockRank 81

Here is the EG Industries MoatMap StockRank, as of October 5, 2026, when the original thread was posted. The live page carries today's reading.

• Quality: 49/100. Middling, and the most informative number here. Contract manufacturing carries thin margins by nature, the customer base is concentrated, and the share count is set to grow. The framework is not persuaded that the transformation has yet shown up in the quality of the earnings.
• Value: 58/100. Moderate on trailing numbers. The 12 times forward figure rests on broker forecasts the factor does not use.
• Momentum: 69/100. Strong, as you would expect for a Malaysian company newly attached to the AI build-out.
• Composite StockRank: 81/100. Comfortably in the top fifth, carried by Momentum and Value rather than by Quality.

A Quality score under 50 inside a StockRank above 80 is the framework telling you that the market likes the direction more than the business has yet proven it. Our guide to factor investing covers why a composite led by Momentum deserves a different kind of attention than one led by Quality, and the Malaysian context sits in Best Malaysian Stocks 2026.

The Question Worth Sitting With

Every GPU generation needs a faster module. That is the growth story, and it is also the hidden weakness in the switching cost. Requalification protects a generation that is already qualified. It does far less to protect the next one, because a new generation has to be qualified somewhere regardless, and a customer deciding where to build 1.6T is not paying a switching cost to leave EG. It is simply choosing where to start.

Is the advantage EG's, or is it the tariff's, held on behalf of a customer that could build the same line itself?

The bull answer is that EG has already done the hard part: an 80 percent 800G yield, a US$242 million order, a 1.6T ramp due by the end of 2026, and a second customer in Micas. Each qualified generation makes the next qualification easier and more likely to stay, and at about 12 times forward earnings against Fabrinet's 25.5 times, the market is pricing EG as if the relationship were fragile.

The bear answer is that a third of revenue sits with a customer that has its own Malaysian base, that the 1.6T transition is the natural moment for that customer to rethink where it builds, that the CPO opportunity accrues 70 percent to somebody else, and that warrants will add up to 48 percent more shares by 2028. The tariff that sends work to Malaysia is a policy, and a policy can change in a way a switching cost cannot.

Companion Reading

EG sits inside our AI hardware and qualified-supplier clusters with three close neighbours:

• Mi Technovation (5286.KL) is the closest Malaysian twin: a supplier qualified into an AI supply chain, where the switching cost is the risk of changing and the growth arrives with each new generation.
• AT&S (ATS.VI) for the same concentration question in AI hardware: capacity built around a small number of very large customers, and what happens to the supplier when those customers change their plans.
• SK Hynix (000660.KS) for the bigger picture EG is part of: in AI computing the constraint is increasingly moving data, not computing it, and the suppliers of bandwidth are where the bottleneck pays.

The Bottom Line

EG Industries builds the light pipes of AI data centres: 800G optical transceivers at an 80 percent yield, 1.6T due by the end of 2026, and co-packaged optics on the way. Its moat is requalification, the months of lost output a customer accepts by moving a qualified line in a market growing about 60 percent a year. Its biggest risk is that the customer behind a third of its revenue, likely CIG, already has a Malaysian base and could bring the work in-house, and that every new module generation is a fresh chance to do so.

The price reflects both. Fully diluted it is 20 times core earnings and about 12 times FY2027 forecasts against Fabrinet's 25.5, the CEO is buying, and the scorecard reads Q49 V58 M69, StockRank 81: momentum and value ahead of quality. The diversification into CPO through Genetronic is real but only 30 percent EG's, and the warrants dilute by up to 48 percent by December 2028. For sizing a position where one customer and one policy carry this much weight, our guide to reviewing your portfolio for weak spots is the right framework.

For the full breakdown, including the customer mix, the Genetronic structure, the warrant schedule and the valuation walk, the EG Industries Deep Dive is the place to go.

Credit where due: thanks to @Iqbal_yusuf1994 for mentioning EG a few months before we wrote about it.

This article is opinion, not financial advice. The author may hold positions in names covered on MoatMap.