Samsung Electronics made 57 trillion KRW (≈$39B) in just three months
Samsung Electronics made 57.2 trillion KRW during the first three months of this year. Hard to picture, right? The Korean government spends about 60 trillion KRW a year on national defense, so Samsung alone made almost that much in just one quarter.
Revenue: 133 trillion KRW, operating profit: 57.2 trillion KRW. It nearly matched the company’s all-time record annual operating profit from all of 2018 (58.9 trillion KRW) in just one quarter, three months.
But do you know what these numbers really mean? Why did it suddenly start making so much money, is this temporary, and why is Nvidia — which makes a similar amount of money — valued at five times more than Samsung? Let’s unpack it one by one.
No. 1 Apple $50.9B · No. 2 Nvidia $44.3B · No. 3 Microsoft $38.3B · No. 4 Samsung Electronics ~$38.0B · No. 5 Alphabet $35.9B
2026 Q1 big tech operating profit — Samsung in 4th place?
Samsung Electronics entered the big tech top 5, overtaking Alphabet, Google’s parent company.
DDR5 vs GDDR6X vs HBM3E — compare at a glance
Check out the performance gap between DDR5, GDDR6 and HBM in the graph below.
DDR5 vs GDDR6X vs HBM3E — What’s the difference?
The three memories differ like highway lane sizes. You can see it just by looking. HBM is overwhelming.
| Metric | 💻 DDR5 | 🎮 GDDR6X | 🤖 HBM3E |
|---|---|---|---|
| Bus width | 64bit — 1-lane road | 384bit — 6-lane highway | 1,024bit — 32-lane highway |
| Bandwidth | 77 GB/s | 1,008 GB/s | 1,170 GB/s (15× DDR5) |
| Power consumption (24GB) | ~10W | ~60W (needs 12 chips) | ~15W (one stack is enough) |
| Structure | Flat layout (2D) | Flat layout (2D) | Vertical stacking (3D) – 16 floors like an apartment |
| Main use | Regular computers, servers | Gaming GPU, graphics | AI training/inference, data centers |
Why does AI devour this memory?
GPU compute speeds are nearly doubling every year, but the speed of fetching data from memory is only growing by about 25–30% annually. This is called the 'Memory Wall'.
No matter how fast a GPU can calculate, if memory can’t deliver data quickly enough, the GPU has to sit there waiting. It’s just like how cars can be incredibly fast on a highway, but if the tollgate is only one lane, everything gets jammed.
Large language models like GPT have to load hundreds of billions of numbers into memory and call them up constantly. HBM is effectively the only solution to this bottleneck.
The HBM market exploded from about $4 billion (5.6 trillion KRW) in 2023 to $46.7 billion (65 trillion KRW) in 2025 — more than 10x growth in just two years. Only three companies in the world can make it: Samsung Electronics, SK Hynix, and Micron.
In 2026, HBM4 mass-produced first in the world by Samsung Electronics has a 2,048-bit bus width (2x the previous generation) and bandwidth of up to 2TB/s. It’s like going from 32 lanes to 64 lanes.
Why do Samsung Electronics earnings ride a roller coaster?
Hover over the dots to see the exact figures. From the 2023 bottom to the 2026 rebound, the roller coaster is visible at a glance.
What makes this AI supercycle different?
Previous cycles were driven by demand like "people are buying more smartphones" and "data centers are expanding." Eventually, those markets become saturated. But AI is different. As AI models like GPT, Gemini, and Claude keep getting larger, memory demand is growing exponentially, and new sources of demand such as AI agents, autonomous driving, and robotics are continuously emerging.
On top of that, the structural supply shortage is hard to resolve. HBM consumes 3 times more wafers than conventional DRAM, yet its yield is only 65%. So even if fabs run at full capacity, the actual chip output remains low.
The Bank of Korea also described it as "the strongest supercycle since the 2000s." DRAM prices surged by as much as 10x in a single year.
① Demand explosion (new technology emerges) → ② Supply shortage · price spike (margin 55–60%) → ③ Capex boom → ④ Fab completion · oversupply · price collapse → back to ① (about a 4-year cycle)
Samsung vs SK Hynix — three reversals
In the HBM market, leadership between Samsung and SK Hynix has surprisingly changed hands multiple times.
2013: SK Hynix invents HBM
Developed jointly with AMD, SK Hynix completed the world's first HBM. Samsung had not yet joined the market.
2016: Samsung takes the lead with HBM2
Samsung became the first in the world to mass-produce HBM2. Samsung HBM2 was used in NVIDIA's V100.
2019: Samsung's critical misjudgment
Samsung management disbanded the HBM R&D team, saying there was "no market potential." At the same time, SK Hynix kept investing. That 2–3 year gap changed everything.
2022–2024: SK Hynix dominates
SK became the first to mass-produce HBM3 and secured exclusive supply for NVIDIA's H100. Samsung failed testing due to overheating issues. Market share: SK 62% / Micron 21% / Samsung 17%.
2026.2: Samsung mass-produces HBM4 first in the world
1c DRAM + in-house 4nm foundry + TSV low voltage = three cutting-edge processes deployed at the same time. 11.7Gbps (46% above the standard). However, volume is still Samsung 30% vs SK 70%.
Samsung Electronics vs NVIDIA — by the numbers
Hover over each item to see detailed figures. The difference in bar length shows the gap between the two companies.
If they make about the same amount of money, why is NVIDIA valued 5x higher?
CUDA controls the GPU well, so it handles calculations without trouble. That's why AI engineers worldwide work on CUDA. But CUDA only runs on NVIDIA GPUs. So if you want to do AI you need CUDA, and to use CUDA you need NVIDIA. Samsung's memory is a bit different; you can use something other than HBM and you don’t have to use only Samsung memory. So Samsung's core competitiveness is lower than NVIDIA's.
| Metric | 🏭 Samsung Electronics | 🧠 NVIDIA |
|---|---|---|
| Business model | Design + manufacturing (runs factories directly) | Design only (fabless, outsourced to TSMC) |
| Core moat | Factories + technology | CUDA ecosystem 2 million developers |
| Profit pattern | Cyclical (58T→6T→57T) | Structural growth (31 times in 4 years) |
So what does this actually mean for those of us living in Korea?
Samsung Electronics alone accounts for about 20% of Korea’s total exports and about 20% of the KOSPI’s market capitalization.
When semiconductor exports rise, more dollars flow into Korea → stronger KRW. For those sending money back home, it’s a double-edged sword. When the won strengthens, you get fewer dollars, yen, or baht when you exchange it.
During the 2018 supercycle, Korea’s economic growth rate was 2.9%; in 2023, a downturn year, it was 1.4%. Samsung Electronics’s earnings are not just one company’s report card — they are a thermometer for the Korean economy.
Whenever you hear "Samsung Electronics earnings" in the news, it’s a story connected to this country’s exports, exchange rates, employment, and your everyday life.
Even if two companies make the same amount of money, whether those profits can continue is what separates market valuations. Samsung Electronics is a 'factory that makes things well,' while NVIDIA is an 'ecosystem you can’t leave' — and the market assigns a higher value to the latter.