In the last episode, we looked at how semiconductors fall into two broad types.
Chips that calculate, and chips that remember.
In the AI era, both matter. No matter how powerful a chip's calculating ability is, performance breaks down if data doesn't arrive fast enough.
That's why one type of memory chip keeps coming up in every AI conversation lately.
HBM — High Bandwidth Memory.
AI Needed a Wider Road
GPUs calculate at incredible speed.
But speed alone isn't enough.
If data isn't delivered fast enough, the GPU simply sits there, waiting.
Regular memory is like a single-lane road.
Cars can pass, but the moment traffic increases, everything backs up.
HBM is more like a multi-lane highway.
Multiple streams of data move at once, constantly feeding the GPU.
As AI models keep growing, this gap matters more and more.
AI no longer needs a road for one or two cars. It needs a road built for hundreds of cars moving at the same time.
But It's Extremely Hard to Make
HBM isn't simply "bigger" memory.
Multiple layers of memory are stacked with precision thinner than a human hair, connected through thousands of microscopic links.
Even a tiny misalignment can cause overheating, signal failure, or a defective chip.
You can't just build a new factory and start making it tomorrow.
It takes decades of accumulated process technology and production experience to manufacture it reliably.
And the Companies That Have That Technology Are in Korea
Here's what's surprising: the world-leading technology for HBM exists at a top level in two companies — in the same country.
Samsung Electronics and SK hynix.
In the overall DRAM market, Samsung Electronics holds the global No. 1 position.
But in HBM — the memory that matters most in the AI era — SK hynix holds more than half the market share, making it the clear global No. 1.
Two companies from the same country are leading both the broader memory market and the most critical AI memory market at the same time.
Of course, Korea doesn't dominate the entire semiconductor industry.
In foundry — the business of actually manufacturing chips designed by others — Taiwan's TSMC holds the world's No. 1 spot.
But when it comes to HBM, the memory that matters most in the AI era, Korea sits at the center of the global supply chain.
That's probably why NVIDIA CEO Jensen Huang has visited Korea twice — it's not a coincidence.
No matter how advanced NVIDIA's AI chip designs are, they can't reach full performance without HBM.
In the end, even the world's top AI companies now have to go through Korea's memory technology.
So, Why Korea — Again?
Competitiveness in the AI era ultimately comes down to one thing: how fast you can deliver data.
And right now, Korea sits at the center of that answer.
But this advantage didn't appear overnight.
Behind the reason the world now depends on Korean semiconductors lies decades of investment, technological development, and fierce competition.
In the next episode, we'll look at how Samsung Electronics and SK hynix rose to where they stand today.
Curious about Korea?
So am I.
Further Reading
- Counterpoint Research – Global DRAM and HBM Market Share: https://counterpointresearch.com/en/insights/global-dram-and-hbm-market-share

Comments
Post a Comment