In every gold rush, there are two games being played. One is the miner’s game, chasing upside and attention. The other is the shovel maker’s game, selling capacity and constraints. In tech, the second game tends to win over long time horizons because it sits closer to the parts of the system that are scarce and expensive to replicate.
AI is a clean example... a lot of companies right now are wrappers around foundation models. Some of them will build real businesses, but structurally they live upstream of nothing they control. If the platform changes pricing, rate limits, or distribution, the wrapper takes the hit. They may or may not hit gold.
The shovel makers are the ones building the physical and industrial base that the whole ecosystem is pinned to: compute, networking, cooling, chips, packaging, tooling, process know-how, and maybe even, robot.
For years, I repeated a single statistic as evidence that Malaysia is central to semiconductors: Malaysia exports 13 percent of the world’s semiconductors. The number is true, but it can mislead you. Export share is throughput. Throughput is not the same thing as value capture. You can move a lot of product and still sit in a part of the chain where margins are structurally capped.
I went to a semiconductor conference in Penang with that statistic in my head. Not as a patriotic slogan, just as an implicit assumption: we must be closer to the core than people think. After a day of walking the booths, listening to talks, and seeing what companies were actually selling, the conclusion was straightforward. Malaysia is important, but our strength is in operational delivery, not in the parts of the value chain that set prices through IP or scarcity.
The 13% Paradox
Malaysia’s semiconductor footprint is largely ATP: assembly, testing, and packaging. The wafers get fabricated elsewhere, then they come here to be diced, tested, packaged, and shipped. That is real industrial capability. It creates jobs, operational skill, and a place in the supply chain. It is also a segment where competition is intense and margins are thinner because the work is easier to benchmark and easier to move if cost or policy changes.
The simplest economic model for this is the Smile Curve. At the left end is R&D and chip design, where IP and differentiation create pricing power. At the right end is market control, distribution, and services, where customer lock-in and brand capture margin. In the middle is manufacturing and assembly, where performance is measured by yield, cycle time, defect rates, and cost. It is crucial work, but it is harder to defend structurally because buyers can multi-source and squeeze.
This is what the middle-income trap looks like in an industrial context. Wage levels rise, so the cheap-labor advantage fades. But moving into the ends of the curve requires more than capital. It requires a compounding capability stack: research depth, specialized tooling, process IP, and a dense talent pipeline. If those do not accumulate locally, you stay globally relevant but economically capped.
The Temptation to Skip Steps
The most common reaction I hear is frustration. "Why don't we just build our own chips? This is a dangerous fantasy. You cannot wake up one day and decide to build a Ferrari engine if you haven't yet mastered how to cast the steel. History shows us that every semiconductor giant started by doing the "boring" work better than anyone else.
The problem with that plan is not the ambition. It is the missing mechanism. Semiconductor leadership is path dependent. You do not get there by declaring an outcome. You get there by accumulating process learning and institutional knowledge over decades, then using that foundation to climb into adjacent capabilities.
Japan is a good reference because it did not begin at the frontier. Early on, Japanese firms, like Sony and Toshiba, licensed key transistor technology and then got exceptionally good at manufacturing discipline. Yield improvement, quality control, process repeatability. Those sound boring, but that is the point. In the 1980s Japan dominated DRAM, a commodity product where small yield advantages compound into massive market advantages. The lesson is not “make memory chips.” The lesson is that process mastery creates leverage. Once you can operate at high yield with low variance, you can move into more complex, higher-margin categories.
China’s OSAT sector shows a similar ladder. JCET started small and spent decades scaling packaging and test. The breakthrough was not a sudden invention. It was operational competence plus capital, deployed at the right time. In 2015, JCET acquired STATS ChipPAC. That deal accelerated capabilities and customer access, but the acquisition only worked because the base layer was already strong. There was already competence to absorb the complexity.
Skipping steps fails because the step you skip is usually the one that builds the capability you need.
The Strategy: Be the Undisputed Champion of ATP
Malaysia needs to learn this lesson. We cannot skip to the finish line.
The National Semiconductor Strategy (NSS) has correctly identified that our immediate goal shouldn't be to build a Fabrication plant (Fab) to compete with TSMC. That is a $20 billion game we cannot afford to lose.
Our goal must first be to become the Undisputed Champion of Advanced ATP.
Packaging is no longer just "wrapping." In the era of AI, packaging is the new bottleneck. NVIDIA's H100 chips are limited not by how fast they can be printed, but by how fast they can be packaged using "CoWoS" (Chip-on-Wafer-on-Substrate) technology. This involves stacking chips in 3D structures and connecting them with hair-thin wires.
Why We Can’t Just "Copy" Taiwan
The standard advice is often: "Why doesn't Malaysia just build its own Fabrication plants (Fabs)?"
It is not that simple. A modern Fab is not a factory; it is a fortress of physics. The supply chain is so complex that even the United States is struggling to re-shore it despite throwing billions of dollars at the problem.
A Practical Strategy: Advanced Packaging
This is where Malaysia’s strategy can actually make sense. Building a leading-edge fab ecosystem to compete with TSMC is not a near-term move. A modern fab is a multi-billion-dollar system that depends on an entire surrounding ecosystem: equipment, chemicals, ultra-pure materials, research institutions, and a deep bench of specialized engineers. You cannot assemble that quickly, and you definitely cannot do it cheaply.
Advanced packaging is different. It is adjacent to what Malaysia already does, and in the AI era it is increasingly a bottleneck. Packaging is no longer just a plastic shell. It is part of the performance path.
Look at what happens in high-end AI compute. Technologies like Chip-on-Wafer-on-Substrate exist because the system-level problem is no longer “one chip is fast.” It is integration, bandwidth, thermals, yield, and reliable scaling. The market rewards whoever can deliver advanced packaging at high yield, at volume, with tight reliability. That is not a branding story. It is manufacturing physics.
If packaging capacity is constrained globally, then packaging becomes pricing power. That is how you move up a curve without pretending you can teleport to the top.
The Real Constraint: Talent Compounds or It Doesn’t
If you strip away slogans, the binding constraint is talent density and capability accumulation. Advanced packaging demands specialists: process engineers, materials, test, mechanical and thermal design, quality systems that run at scale. Those skills are not downloaded. They compound through long careers and tight feedback loops in production.
Malaysia also faces a straightforward labor market problem. Engineers can move to higher-paying markets and capture immediate upside. If local roles do not offer comparable technical growth, compensation, and prestige, retention becomes structurally hard. Over time that becomes a ceiling. Not because Malaysians are not capable, but because the system does not compound expertise locally.
If Malaysia wants to climb, talent pathways have to be treated like infrastructure, not like a footnote. <talk about National Semiconductor Strategy (NSS)>
The 2 Million Dollar Part Theory
The most profound lesson I learned was this: Do not try to make the product. Make the crucial part for the product.
A semiconductor manufacturing system can cost hundreds of millions of dollars. Malaysia does not need to build the entire machine to capture value. It can capture value by becoming world-class at a critical subsystem inside the machine, the part that is hardest to replicate and most painful to be without.
Penang already has examples. ViTrox builds inspection and machine vision systems. That functionally sits in the metrology and yield control loop. Test socket companies build the interfaces that connect chips to testers, a niche that is not glamorous but is indispensable and precision-heavy. These are leverage points. They work because the engineering difficulty is real and the learning curve is steep.
You don't need to be the king of the entire hill. You just need to be the only person selling the shoes required to climb it. That is how you become a default dependency instead of a low-cost option.
The Trifecta Solution
To get there, we need what I call "The Trifecta":
- The Student:Committed to mastering the hard sciences, knowing there is a light at the end of the tunnel.
- The Government: Funding R&D, not just "D" (Development).
- The Corporation: Providing the sandbox for innovation and a guaranteed career path for students who obtain the specialized degrees the industry demands.
We need to stop trying to do everything(AI, Fab, Design, OSAT) all at once. We need to focus. We need to find the one boring, difficult, high-precision part of the supply chain that the rest of the world cannot live without, and we need to be the champion in the world at it.
That is how Japan did it. That is how China did it. And that is how Malaysia will escape the middle-income syndrome.
Identity as Strategy
When I walked into the conference, I assumed the 13 percent statistic spoke for itself. By the end of the day, it was clear that the number describes throughput, not value capture. It measures how much flows through Malaysia, not how much of the margin stays here.
That distinction matters because the middle-income trap is not a mystery. It is what happens when a country specializes in a low-margin layer of a value chain, then tries to outspend or out-innovate incumbents without the underlying capability stack. In semiconductors, value concentrates in design, tooling, process IP, and the ecosystems that compound around them. Assembly and test can be globally important while still structurally capped on margin.
So the practical question is not “How do we become the next TSMC?” It is “Which adjacent capability can Malaysia compound from where we already sit?” Advanced packaging is a plausible answer because it is close to our base, globally constrained, and rewarded with higher margins when executed at high precision and high yield. The same logic applies to narrower, high-leverage components inside the chain: inspection, test interfaces, metrology, and the specialized subsystems that become bottlenecks.
In that frame, identity stops being a slogan and becomes a strategy. A country’s identity, economically, is its comparative advantage made durable: the set of capabilities, suppliers, talent pathways, and standards it can deliver reliably at world-class levels. If Malaysia wants to escape the middle-income trap, the work is not to chase breadth. The work is to pick a small number of technically hard niches where learning curves are steep, then invest long enough that the advantage compounds and the market starts treating Malaysia as a default dependency rather than a low-cost option.