Samsung’s HBM memory sales drive 8x profit surge as AI demand triggers margin expansion

Samsung startled the semiconductor industry this quarter with an eightfold year-over-year jump in operating profit, driven by surging sales of high-bandwidth memory, or HBM, used in AI servers. The easy takeaway is that AI demand is lifting all chips. The more important one is narrower and more durable: a key slice of the memory market is no longer following the old commodity script. The result also fits into Samsung’s broader 2026 push to lead in AI chips.

For years, memory chips such as DRAM and NAND lived through punishing boom-and-bust cycles. Supply would rise, prices would crack, and bargaining power would drift toward hyperscalers and large buyers. HBM looks different. It is harder to make, available in limited volume, and essential for training and running advanced AI systems. That combination is giving suppliers something memory makers rarely keep for long: pricing power.

HBM prices rose more than 35% quarter over quarter, according to market trackers, even as legacy DRAM pricing stayed flat or weakened. Samsung’s chip division also recorded its sharpest margin improvement since before the last memory downturn, with the gains concentrated in its constrained, high-value HBM output. Customers building AI infrastructure are signing multi-quarter supply commitments above spot pricing, and demand has stayed firm despite broader worries around China exposure and global disruption.

That is the real signal in Samsung’s quarter. This was not a generic AI halo effect washing over the whole memory business. It was a clear sign that HBM, because of tight supply and technical difficulty, is carving out its own economics.

Why HBM is changing the rules for memory

HBM’s edge comes from a supply chain very few companies can run well. These chips require advanced 3D stacking, tight thermal control, and close integration with leading-edge logic processors. Expanding output takes major capital spending, but money alone is not enough. Yields, packaging expertise, and qualification with top customers matter just as much, and those are hard to build quickly.

That is why the market remains so concentrated. Samsung and SK hynix control more than 90% of HBM capacity, while Micron is still scaling and trails in volume and yields. In practical terms, buyers that need HBM for AI accelerators have limited options. Samsung’s aggressive positioning in memory also complements a wider strategy that now stretches from infrastructure to devices, including its AI-heavy Galaxy S26 lineup.

The proof is already commercial. Samsung’s HBM lines are reportedly booked into late 2024, and major customers, including Nvidia, have had to negotiate supply allocation months ahead. Meanwhile, standard DRAM and NAND, where supply is broader and the product is easier to substitute, did not show the same pricing strength or revenue lift. Even with US-China tensions unsettling other parts of tech, HBM shipments held up, with no clear signs in quarterly results of major order delays, clawbacks, or export restrictions cutting into demand.

That matters because AI systems cannot easily sidestep HBM. As training models grow larger and inference chips spread across data centers, memory bandwidth becomes a core performance requirement. Buyers can delay some purchases, switch vendors in other categories, or pressure suppliers where alternatives exist. In HBM, those levers are weaker. The result is a memory segment where customers are committing earlier, paying more, and accepting tighter allocation than the industry has seen in years.

This does not mean the old memory cycle has vanished. Capacity will expand, competitors will catch up, and pricing will eventually face pressure. But Samsung’s quarter shows something real has changed. In the most important AI-linked part of memory, technical bottlenecks and disciplined supply are shifting margin power back toward producers.

What to watch next

The next question is whether this new profit pool holds as supply ramps. If Samsung, SK hynix, and Micron all add meaningful HBM capacity by year-end, the key indicator will be contract pricing. Watch quarterly HBM price trends and announced fab or packaging investments more closely than broad industry optimism.

Second, track competitor progress. If Micron can push into double-digit HBM market share while keeping margins healthy, that would suggest this is becoming an industry-wide reset rather than a temporary Samsung advantage. Customer wins tied to Nvidia’s next-generation AI accelerators will be especially important.

Third, keep an eye on external shocks. Any disclosure from Samsung or peers around delayed HBM orders, export friction, or allocation changes tied to regulation would test how resilient demand really is. So far, buyers have kept ordering through the noise.

For now, Samsung’s quarter points to a simple conclusion: AI is reshaping where profits sit inside semiconductors, and HBM is one of the clearest places to see it. Anyone still treating all memory like a commodity is reading the market one cycle behind. That same AI strategy is increasingly showing up beyond chips as well, including Samsung’s reported move into smart glasses as part of a bigger 2026 AI push.