Everyone’s focused on Nvidia, Anthropic, Alphabet, TSMC and other flashy AI names.
But every technology boom eventually pushes deeper into the supply chain, and that’s usually where the next layer of winners starts emerging.
These three listed South Korean stocks sit in very different parts of the AI buildout, from chip packaging and semiconductor chemicals to industrial robotics.
They’re still moonshot style bets overall, but if even one of them becomes deeply embedded in the global AI stack, the upside could end up life changing for early investors.
Hanmi Semis (KRX: 042700)
There’s a part of the chip story most people don’t see.
We all talk about Nvidia designing the brains, or Samsung and TSMC stamping them out in massive fabs.
But once those chips come off the production line, they’re not ready to go. They still need to be cut, stacked, aligned, bonded, and packaged into something that actually works inside a server.
That’s where Hanmi lives.
In Korean industry terms, it’s a 반도체 후공정 장비 기업 — a back-end semiconductor equipment company.
In plain English, it builds the machines that put chips together after they’ve been made.
It’s not the glamorous part of the process, but it’s the part that decides whether everything else actually functions.
Take its TC bonders, for example.
These machines use heat and pressure to stack memory chips on top of each other, layer by layer. That’s essential for high-bandwidth memory, or HBM — the type of memory powering AI systems today.
Without that stacking process working perfectly, the whole AI pipeline starts to choke.
Then there’s its micro SAW equipment, which cuts semiconductor packages with extreme precision, and its inspection systems that make sure everything lines up at a microscopic level
It’s all highly technical work, but it sits right at the point where small errors turn into big losses.
For a long time, this part of the industry didn’t get much attention.
The real money, people thought, was in design and fabrication. But AI is changing that dynamic.
As chips get more complex and demand for HBM surges, the bottleneck is starting to move.
It’s no longer just about how many chips you can produce; it’s about how efficiently you can assemble them.
If demand keeps building, companies like Hanmi don’t just supply the industry, they start to shape its limits.
The speed, yield, and precision of packaging becomes a constraint on how fast AI infrastructure can scale. And when you sit at a constraint point in a global supply chain, your importance tends to rise quickly.
That’s where the “moonshot” angle for investors starts to come into focus.
If Hanmi continues to execute and stays embedded in the HBM supply chain, it could find itself in a position where growth is pulled forward by forces much bigger than the company itself.
Of course, it’s not a straight run
Hanmi still relies on continued investment from its customers, particularly memory makers, to keep the momentum going.
But if AI is the engine, and chips are the fuel, then packaging is the part that keeps everything from falling apart.
Hanmi sits right there, in the middle of that process.
SoulBrain (KOSDAQ: 357780)
There’s a layer in the semiconductor world that almost never gets talked about, but decides whether everything else works.
Before a chip can process AI models or power a data centre, it has to go through dozens of chemical processes — cleaning, etching, coating, layering.
All at atomic-level precision.
And that’s where Soulbrain sits.
In Korean industry language, it’s described as a 반도체 핵심 소재 기업, a company supplying core materials for semiconductor manufacturing.
Strip away the jargon, and what it really means is this: Soulbrain makes the ultra-pure chemicals that chips physically depend on to exist.
We’re talking about materials like TEOS, used to form thin insulating layers on chips, or CMP slurry, which polishes wafers down to near-perfect flatness.
There are also etchants that carve microscopic circuits into silicon, and precursors used in deposition processes that build up chip structures layer by layer.
If any of these materials are even slightly off in purity or consistency, the entire chip can fail.
That’s what makes the business interesting.
Once a material is qualified inside a semiconductor process, it’s very hard to replace. Switching suppliers isn’t just a commercial decision, it can mean revalidating entire production lines.
So companies like Soulbrain become embedded in their customers’ operations over time.
The story doesn’t stop at semiconductors either.
Soulbrain has been building out adjacent materials businesses in displays and batteries, including electrolytes for lithium-ion cells — the medium that allows ions to move between electrodes.
It’s a similar pattern: high-purity, high-reliability inputs sitting deep inside fast growing industries like EVs and energy storage.
For a long time, this part of the supply chain didn’t attract much attention.
But as processes shrink and move into more complex 3D structures, the demands on materials are increasing.
That’s where the moonshot angle for investors starts to build.
If advanced semiconductors and AI infrastructure continue scaling, the companies supplying the materials that enable those processes become more important.
And the deeper they sit in the production stack, the harder they are to displace.
Doosan Robotics (KRX: 454910)
There’s a part of the AI story that still feels like science fiction, right up until you realise factories are already preparing for it.
Everyone talks about Nvidia building the brains, or OpenAI training giant models.
But eventually AI has to operate in the real world, where machines need to move, react, and work safely around humans.
That’s where Doosan Robotics sits.
In Korean industry language, it’s a 협동로봇 기업, a collaborative robotics company.
Put simply, the company builds robots designed to work safely alongside humans in factories, logistics centres and industrial environments.
Its main products today are collaborative robotic arms, or cobots, used for welding, palletising, packaging, machine tending and repetitive industrial work.
Unlike old-school factory robots trapped behind giant safety cages, these are designed to operate around people.
Recently, Nvidia deepened ties with Doosan as the two companies discussed building what they call a robot execution platform.
Doosan has been developing something called an Agentic Robot Operating System, designed to help robots perceive their surroundings, and carry out tasks precisely.
The two companies are now aiming to showcase intelligent robotic solutions in 2027, followed by industrial humanoid robots by 2028.
Now, “humanoid robots” is one of those phrases markets instantly lose their minds over.
Factories, warehouses and logistics systems were all built for humans. Shelves, staircases, tools, pathways — everything assumes a human-shaped worker.
So instead of rebuilding entire factories for robots, companies are increasingly trying to build humanoid robots that fit existing human environments.
That’s where the Doosan’s moonshot angle for investors starts creeping in.
If AI becomes physical over the next decade, the winners may not just be the companies building the AI brains.
It could also be the companies building the execution layer between AI and the real world.
And Doosan is trying to position itself right where those two things meet.
——–
Now read: Korea’s AI Squid Game has a wild card: A mathematician turned founder
Follow us on X for more under-the-radar tech and biotech plays from Asia: https://x.com/ChairmanAsiaX
