Two years ago, Samsung’s semiconductor division was bleeding. Memory chip prices had cratered, inventory was piling up, and analysts were writing obituaries for the DRAM cycle. Samsung even took the extraordinary step of cutting production — something the world’s largest memory chipmaker almost never does voluntarily. The division posted its first operating loss in over a decade.
Fast forward to this week: Samsung just reported 57.2 trillion won ($37.8 billion) in operating profit for Q1 2026. That’s not a typo. That’s a 755% year-over-year surge and an all-time quarterly record — not just for Samsung, but for any semiconductor company in history. And the Device Solutions division, the same semiconductor arm that was hemorrhaging cash in 2024, accounted for roughly 54 trillion won of that total. That’s 95 cents of every dollar Samsung earned.
The AI Memory Supercycle Is No Longer a Forecast — It’s Here
The numbers tell a story that’s almost absurd in its clarity. Consolidated sales hit 133 trillion won ($88 billion), up 68% year-over-year. DRAM prices surged 90-95% quarter-over-quarter. NAND flash followed. And Samsung’s high-bandwidth memory (HBM) shipments — the chips that go directly into Nvidia’s AI accelerators — are selling out faster than Samsung can produce them.
This isn’t just a Samsung story. It’s confirmation that the AI infrastructure buildout has entered a phase where the companies making the physical components — the memory, the chips, the interconnects — are capturing value at a rate that makes even the AI model companies look modest by comparison. OpenAI is reportedly losing $14 billion a year. Samsung just made $37.8 billion in a single quarter.
Follow the Money: Why Memory Chips Are the Real AI Winner
Here’s the part most people miss about the AI hardware stack. Everyone talks about GPUs — Nvidia, AMD, the custom silicon arms race. But every AI accelerator needs memory. Lots of it. An Nvidia B200 GPU doesn’t work without HBM3E chips stacked on top of it, and right now, there are only three companies on Earth that can make those chips: Samsung, SK Hynix, and Micron.
That’s an oligopoly feeding what might be the largest capital expenditure cycle in technology history. Amazon is spending $200 billion on AI infrastructure this year. Microsoft, Google, and Meta are each pouring in north of $60 billion. Every dollar of that capex translates into demand for memory chips — and the supply side simply cannot keep up.
DRAM prices are projected to climb another 60% in Q2. If that holds, Samsung’s Q2 could be even bigger than Q1. We’re watching a supercycle that makes the 2021 crypto-driven chip boom look like a rounding error.
Samsung’s Comeback Is Also a Revenge Story
There’s a competitive angle here that matters. For the past 18 months, SK Hynix had been eating Samsung’s lunch in the HBM market. Hynix was first to qualify its HBM3E chips with Nvidia, first to ship in volume, and first to lock in long-term supply agreements with the hyperscalers. Samsung was late. Its early HBM3E yields were reportedly poor, and there were whispers that Nvidia had deprioritized Samsung as a supplier.
That narrative is now dead. Samsung’s Q1 numbers suggest it has not only caught up on HBM yields but is shipping at a scale that’s driving record revenue. The company confirmed that HBM shipments grew “significantly” quarter-over-quarter, and analysts at Korea Herald report that Samsung is now the second-largest HBM supplier by revenue, closing the gap with Hynix fast.
Samsung also has a structural advantage that Hynix doesn’t: it makes its own logic chips, OLED displays, and consumer electronics. That diversification was a drag when memory prices were low. Now it’s a hedge that lets Samsung invest more aggressively in capacity expansion without betting the entire company on one product line.
The Risk Nobody Wants to Talk About
Before you go all-in on the Samsung supercycle thesis, there’s a geopolitical shadow hanging over this story. The ongoing U.S.-Israel conflict with Iran has disrupted shipments of helium and neon — gases critical to semiconductor manufacturing. Helium is used in chip fabrication cooling systems, and neon is essential for the lithography lasers that etch circuit patterns onto silicon wafers.
South Korea imports a significant portion of its specialty gases from suppliers with Middle Eastern exposure. If the conflict escalates further, Samsung’s fabs could face production constraints in the second half of 2026 — precisely when demand is expected to peak.
There’s also the tariff question. The U.S. just introduced legislation targeting China’s chip industry, and while Samsung isn’t a direct target, its massive manufacturing operations in China (Samsung runs one of the largest NAND flash fabs in Xi’an) could get caught in the crossfire. Trade policy uncertainty is the one thing that could cool this supercycle before the market does.
What This Means for the Broader Market
Samsung’s Q1 is a data point that reshuffles how you should think about the AI value chain. The conventional wisdom has been: Nvidia captures the GPU margin, cloud providers capture the inference margin, and everyone else fights for scraps. That’s wrong now.
Memory is capturing a disproportionate share of AI infrastructure spending because it’s the bottleneck. You can design a faster GPU, but if you can’t get enough HBM to pair with it, the GPU sits idle. Samsung, Hynix, and Micron have pricing power that even Nvidia would envy right now — and unlike GPU demand, memory demand scales linearly with every model training run, every inference query, every AI agent deployed at scale.
Samsung’s shares jumped 5% on the earnings preview. But if you think about the trajectory — DRAM prices up another 60% next quarter, HBM demand showing no signs of slowing, and Samsung’s foundry business potentially picking up Nvidia packaging orders — the stock move might be just the beginning.
The Verdict
Samsung’s $37.8 billion quarter isn’t just a corporate milestone. It’s a signal that the AI hardware boom has matured past the “picks and shovels” metaphor and into something more like an industrial revolution — where the companies controlling the raw materials of intelligence are printing money at a pace the software layer can’t match.
Two years ago, Samsung’s chip division was the cautionary tale. Today, it’s the proof point. The AI supercycle is real, it’s accelerating, and the biggest winners aren’t the ones building the models — they’re the ones building the memory those models can’t run without.