NextFin News - Samsung Electronics and SK hynix answered a fresh wave of AI skepticism with something the market cannot easily dismiss: record profits, record memory results, and a demand outlook that still points to server-heavy AI spending rather than a fading hype cycle. Samsung reported second-quarter revenue of KRW 171.5 trillion and operating profit of KRW 89.5 trillion on July 30, while SK hynix posted revenue of KRW 79.3 trillion and described the quarter as driven by AI-related memory demand and tight supply. The message was simple. The AI buildout has not stopped at chips that do the computing; it is still being bottlenecked by the memory and packaging layers that make those chips usable at scale.
The market’s doubt is understandable. Semiconductor memory has never stopped being cyclical, and every strong quarter invites the same suspicion: that prices are peaking just as supply responds. But the current numbers are forcing investors to confront a more specific possibility. The profits are cyclical. The bottleneck is more durable. Samsung said its memory business posted an all-time high in quarterly revenue and operating profit, while the Device Solutions division generated KRW 89.2 trillion in operating profit. SK hynix said revenue grew 51% sequentially and 257% from a year earlier to KRW 79.3 trillion. Omdia, in a July 30 forecast, raised expected 2026 semiconductor revenue growth to 94.1% year over year and said memory ICs are now expected to account for more than half of total semiconductor revenue.
That combination matters because it reframes the whole AI trade. The question is no longer whether demand for artificial intelligence exists. It is whether the chain of supply that serves it can keep up. Samsung said demand in the second half of 2026 should remain robust, centered on servers, supported by continued AI infrastructure capex and broader adoption of agentic AI. That is a very different demand profile from smartphones or PCs. It is enterprise spending, tied to data-center buildouts, and it tends to run through longer procurement cycles. SK hynix’s result points in the same direction: the company is not benefiting from a one-off restock, but from a market that still cannot produce enough AI-oriented memory quickly enough to clear demand.
That is why the debate around Samsung and SK hynix has shifted from earnings beats to industrial structure. A memory upcycle is usually self-defeating because higher prices bring new capacity, and new capacity eventually crushes prices. This time the mechanism is slower. High-bandwidth memory requires more advanced integration, more complex packaging, and a narrower set of suppliers than ordinary DRAM. When the bottleneck sits there, the supply response does not behave like a normal commodity cycle. It takes longer for the industry to catch up, and by the time it does, the next generation of AI systems may already have moved the requirement higher again.
That does not mean the cycle disappears. It means the profit cycle and the supply cycle are no longer perfectly aligned. Samsung and SK hynix can print record numbers because the market is still short of the exact memory products AI servers need, but the same concentration of profit is what eventually invites more investment, more qualification work, and more competition. The near-term story is therefore strong earnings. The medium-term story is whether those earnings reflect a one-time shortage or a reordering of the semiconductor stack.
The Market Is Pricing Scarcity, Not Just Demand
Samsung’s quarter was not just better than feared; it was better in the part of the company that matters most for AI. Consolidated revenue reached KRW 171.5 trillion and operating profit KRW 89.5 trillion, both all-time highs. The company’s memory business also posted record quarterly revenue and operating profit, and Samsung said its outlook for the second half remains anchored by server demand and AI infrastructure spending. The market’s consensus bar was already high, with estimates around KRW 172.65 trillion for revenue and KRW 88.13 trillion for operating profit, so the profit line mattered more than the top line. Samsung did not need a blowout to change the story. It needed to show that record margins can survive even when the wider electronics cycle is uneven.
That is exactly what a scarcity market looks like. The value is not in broad unit growth. It is in access to constrained capacity. Samsung said the memory business achieved another record-breaking quarter by proactively addressing AI demand despite limited capacity, focusing on server products. SK hynix described the quarter in similar terms, with AI-related memory demand and tight supply doing the heavy lifting. Those statements are important because they identify where pricing power comes from. It is not a generic semiconductor recovery. It is a bottleneck in the memory architecture used by AI accelerators and server systems.
The same logic explains why the wider semiconductor forecast has become so aggressive. Omdia’s 94.1% growth forecast for 2026 semiconductor revenue is not a routine cyclical call; it is a sign that the market sees memory and AI infrastructure as the dominant marginal source of industry growth. Memory ICs are expected to account for more than half of total semiconductor revenue in 2026. That is a striking number because it tells you the center of gravity has shifted. The economic value of the AI stack is not only accruing to the most visible accelerator makers. It is also accruing to the suppliers that control the scarce memory inputs those accelerators need.
“In H2 2026, the Memory Business expects robust demand centered on servers stemming from continued AI infrastructure capex and broader adoption of agentic AI,” Samsung said in its second-quarter release.
The key words are “servers,” “AI infrastructure capex,” and “agentic AI.” Samsung is not talking about a consumer refresh cycle. It is talking about enterprise buildouts that can stay firm even if phone and PC demand are mixed. That is the mechanism behind the current trade. AI spending creates demand for compute, but compute creates demand for memory, packaging, and a narrower set of components with fewer substitutes. The money flows first to the companies that can ship those bottlenecked parts.
There is a second-order consequence the market may be underpricing. If HBM and server DRAM stay tight, cloud providers and AI platform companies face a higher bill for each new deployment. That means the memory bottleneck is not only a supplier story. It can become a tax on the pace of AI rollout. A more expensive infrastructure stack does not stop AI spending by itself, but it can slow the rate at which some projects reach scale and force buyers to redesign systems around the available supply. In other words, the tight memory market can support the chipmakers now while eventually constraining the very customers that are driving the boom.
Why This Looks Structural, Even If The Profits Are Cyclical
The distinction matters. The profit surge is clearly cyclical. Memory has always been one of the semiconductor industry’s most volatile segments, and every boom eventually attracts enough supply to cool the market. But the constraint underneath this boom is more structural than the usual DRAM story because the choke point is no longer just wafer output. It is HBM, advanced packaging, qualification cycles, and the limited number of suppliers able to serve the highest-end AI systems at scale.
That is why the usual “it will normalize” argument is only half-right. Yes, supply will eventually respond. No, it will not respond at the same speed as a standard memory cycle. Samsung and SK hynix are not only selling more memory; they are selling memory that sits at the center of a more complicated, more technically demanding, and more relationship-driven supply chain. That shifts the cycle’s length. It also changes the way buyers behave. When a part is hard to replace and slow to ramp, customers sign longer deals, secure supply early, and accept less flexibility. That is a structural change in procurement, even if the quarterly earnings still swing like a cycle.
Samsung’s own figures illustrate the point. The Device Solutions division generated KRW 89.2 trillion in operating profit in the quarter, while the memory business posted all-time highs in revenue and profit. SK hynix reported KRW 79.3 trillion of revenue, a level that would have been difficult to imagine before the AI infrastructure buildout intensified. These are not numbers that come from a normal consumer upgrade pattern. They come from a market where a small set of technical bottlenecks can concentrate value in a few suppliers.
The strongest objection is that this is still the old memory story in new clothes. The counter-thesis says buyers will delay purchases if prices stay high, competitors will expand capacity, and the market will eventually revert to oversupply, just as it always has. That view deserves respect because memory has repeatedly punished those who extrapolated too far. It is also plausible that the current AI cycle eventually cools if hyperscalers slow capex or if model economics fail to justify continued spending at the same pace. Nothing in Samsung’s or SK hynix’s results repeals that law.
But the burden of proof has shifted. To falsify the structural thesis, you would need to see more than one weak quarter. You would need to see HBM and server DRAM pricing soften for several quarters in a row, server-order language turn cautious, and the companies’ outlook statements stop referencing robust demand centered on servers. If AI infrastructure spending cools and the memory market’s pricing power rolls over at the same time, then the old cyclical pattern has won. Until then, the supply constraint remains the more convincing explanation.
“This is projected to keep the market undersupplied, despite partial demand moderation in mobile and PCs,” Samsung said.
That line is the strongest evidence that the cycle has changed shape. Samsung is effectively saying the weaker parts of the memory market no longer set the tone for the whole industry. Mobile and PCs can soften without breaking the AI memory story because server-grade products are still short. The market can still cycle. But the center of the cycle has moved.
What The Trade Means For Korea, Customers, And The AI Stack
The first-order effect is straightforward: Samsung and SK hynix benefit from tight supply and strong AI-related demand, while their customers pay more for the memory they need. The second-order effect is wider. If the bottleneck persists, the value created by AI spending shifts toward the least visible parts of the stack — the memory layers, the packaging houses, and the supply-chain specialists that can clear qualification hurdles. That is why the current debate is bigger than two earnings releases. It is about where the margins in AI are settling.
For Korea, the concentration is both a strength and a risk. A market dominated by memory profits can look spectacular when prices are rising, but it becomes sensitive to any hint that the cycle is easing. A slowdown in server demand would not only affect Samsung and SK hynix. It would ripple through equipment suppliers, packaging partners, and index weights tied to semiconductors. The more the local market depends on one narrow branch of the global AI capex cycle, the more every change in guidance matters.
For AI customers, the pressure is different. They can absorb higher memory costs for a while, but the bill eventually matters. If HBM remains scarce, cloud companies and AI platform builders face a choice: pay up, redesign, or slow deployment. That is why the memory bottleneck can become an indirect brake on the pace of AI rollout even if it does not dent the near-term enthusiasm around AI itself. The bottleneck is upstream, but the cost lands downstream.
The base case is that tight AI memory supply will keep Samsung and SK hynix’s results strong through the next several quarters, even if consumer electronics remain mixed. The upside case is that demand stays stronger than expected and the bottleneck lasts long enough to extend pricing power further into 2027. The downside case is that AI capex moderates, buyers push back, and memory pricing begins to normalize before the market has fully internalized how much of the current profit surge depends on scarcity.
The most important signal to watch is not simply another earnings beat. It is whether both companies keep describing demand in the same language: server-heavy, AI-led, and capacity-constrained. If that language weakens, the market will have its answer. If it holds, the skeptics will have to accept a harder truth: the AI boom is no longer just about computation. It is about the parts that make computation possible.
The memory cycle is still cyclical, but AI has made the bottleneck structural enough to keep the earnings upside alive longer than the doubters want to believe.
Explore more exclusive insights at nextfin.ai.
