Samsung’s $62B Profit Is a Record: AI Chip Demand Outran Every Forecast
Samsung Electronics posted a record operating profit of $62 billion for the April-to-June quarter, beating analyst estimates as demand for its AI chips accelerated past what forecasters had penciled in.
It’s the biggest quarterly profit in the company’s history.
Also Read: Can €10 Billion Close Europe’s AI Compute Gap? Brussels Just Placed the Bet
The result extends a record run that started in late 2025, as hyperscaler spending on AI infrastructure compounded quarter after quarter.
Samsung AI Chip Business Drove The Beat
The profit surge was anchored by Samsung’s semiconductor division, which manufactures the high-bandwidth memory chips that underpin nearly every frontier AI accelerator in production today.
The South Korean technology giant’s results, reported Wednesday night U.S. time, confirm that the Samsung AI chip segment has become the company’s dominant earnings engine. High-bandwidth memory, or HBM, is a stacked chip architecture that places DRAM dies directly on top of one another and connects them through tiny vertical channels called through-silicon vias.
The resulting package delivers far higher data throughput per watt than conventional flat memory, making it essential for the matrix math that large language models and image generators perform at inference time.
Samsung competes with SK Hynix and Micron Technology in HBM supply. SK Hynix holds the dominant position in HBM3E, the current production generation, while the Samsung AI chip operation has worked to close the gap through yield improvements over the past two quarters.
The AI data center buildout has proved more durable than analysts expected entering this year.
Hyperscalers including Microsoft, Meta, and Alphabet have each raised capital expenditure guidance in their most recent earnings calls, signaling that GPU cluster expansion will absorb more silicon through at least 2027.
How Memory Became The AI Bottleneck
For much of the past decade, semiconductor analysts focused on compute as the limiting factor in AI training. GPUs and custom accelerators drew the headlines.
Memory was a commodity.
That framing inverted around 2024 as model sizes and inference batch sizes scaled faster than memory bandwidth could follow. A single Nvidia H100 GPU ships with 80 gigabytes of HBM2e or HBM3.
A rack of eight such GPUs demands memory subsystems that can sustain hundreds of terabytes per second of aggregate bandwidth. At that scale, DRAM packaging and stacking capacity became the new bottleneck, not compute die area.
Samsung AI chip revenue reflects that structural shift.
Memory now commands premium pricing that was previously reserved for leading-edge logic chips. Samsung’s gross margins on HBM are estimated to run well above its conventional DRAM business, though the company does not break out HBM financials separately in quarterly disclosures.
A Record Run Built On Hyperscaler Commitments
Samsung’s record profit does not exist in isolation.
The company entered this year with multi-quarter supply agreements from several large cloud providers, locking in volume at prices set before the current demand surge fully materialized. Those agreements provided revenue visibility but also capped the upside Samsung could capture from spot-market tightness.
The more significant dynamic is what comes after those agreements reprice.
Analysts tracking the HBM market expect renegotiations on multi-year supply deals to begin in earnest in late this year, with new terms reflecting the structural undersupply that has persisted since early 2025.
Samsung’s consumer electronics and display divisions also contributed to the quarter, though both segments operate on thinner margins and are more exposed to consumer spending cycles. The Samsung AI chip segment remains the earnings engine.
The South Korean KOSPI index fell sharply in mid-July on concerns about DRAM oversupply in conventional segments, creating an apparent contradiction with Samsung’s record operating print.
The resolution is straightforward: commodity DRAM used in laptops and smartphones faces price pressure, while the Samsung AI chip side of the business remains tight. Samsung straddles both markets, and the Samsung AI chip business is currently large enough to overwhelm weakness in the consumer segment.
Read Next: White House in Panic Mode Over Shocking AI Breach at OpenAI
