Samsung AI Memory Chip Profit Surge Explained: Beyond HBM

Samsung is expected to report Q2 2026 operating profit of around 86 trillion won (approximately $56 billion), roughly 18 times what it earned in the same quarter a year ago, according to an LSEG SmartEstimate drawn from 30 analysts weighted by track record, The Star reported yesterday. That would mark a third consecutive record quarter, driven by what analysts describe as a prolonged shortage across the entire memory market, not just premium AI chips.

The result matters because the profit surge is broad-based. Analysts cite rising prices and constrained supply across conventional DRAM, NAND flash, and high-bandwidth memory simultaneously, a combination that makes this upcycle harder to reverse than a typical single-product spike. The biggest risk to that picture, analysts say, is any slowdown in AI infrastructure spending by major cloud providers.

Samsung DRAM and NAND prices are driving the earnings story

Samsung's Q1 2026 results set the stage. Revenue hit 133.9 trillion won and operating profit reached 57.2 trillion won, both all-time highs, with the Memory division posting an 86% quarter-on-quarter sales increase, Samsung's Q1 earnings release showed earlier this year. Q2 is expected to extend that trajectory by a wide margin.

The price data behind those numbers is striking. Citi Research found that average DRAM selling prices rose 44% quarter-on-quarter in Q2, while NAND prices rose 53% over the same period, The Star reported. Nomura projects further increases of 24% in DRAM and 25% in NAND for Q3, supported by demand from both data centers and consumer devices.

What is driving those increases across the full product stack, rather than in HBM alone, is the expansion of AI workloads into a broader range of computing tasks. Analysts and Samsung itself attribute the demand surge to agentic AI applications that require not just fast inference chips but large pools of conventional memory and storage, The Star reported. Supply cannot keep pace with any part of that demand.

Analysts expect undersupply to persist into 2027, in part because the four new fabs Samsung and SK hynix are building in South Korea are unlikely to produce meaningful output this decade, TECHi reported last week. New chip capacity takes years from groundbreaking to volume production, and fab megaprojects have a long record of slipping, not accelerating.

The shortage has pushed valuations across the sector to historic levels. Samsung, SK hynix, and Micron shares have risen 158%, 273%, and 242% respectively this year, driving all three above $1 trillion in market value, The Star reported. That simultaneous re-rating is a signal that investors read the shortage as an industry-wide condition, not a Samsung-specific story.

How Samsung is positioned within the upcycle

Samsung began mass production and commercial shipment of HBM4 in February 2026, claiming an industry first. The chip delivers total memory bandwidth per single stack increased by 2.7x compared to HBM3E, reaching a maximum of 3.3 terabytes per second, with 40% better power efficiency, according to Samsung's product announcement at the time. For data center operators managing power budgets alongside compute density, those are meaningful numbers.

In Q1, Samsung initiated what it described as the industry's first mass product sales of HBM4 and SOCAMM2 for NVIDIA's Vera Rubin platform, Samsung's Q1 earnings release confirmed. The company projects its HBM sales will more than triple in 2026 compared to 2025. HBM4E sampling is expected in the second half of this year, with custom variants reaching customers in 2027, according to Samsung's earlier announcement.

On the contracts side, Samsung disclosed in April that it had signed multi-year binding supply agreements with customers seeking to lock in access to memory, without naming those customers or disclosing terms, The Star reported. Those agreements could make revenue more predictable if prices cool, though the actual terms remain unknown.

For NAND, Samsung said at its full-year 2025 earnings earlier this year that it plans to scale up high-performance TLC products targeting Key Value SSD demand for AI inference workloads. That is the requirement to store and retrieve large model outputs quickly at scale, a segment Samsung sees as a structural growth opportunity alongside HBM.

The risks embedded in the forecast

Analysts identify potential delays to AI infrastructure investment as the single biggest risk to the current memory market trajectory, The Star reported. JPMorgan noted that while investors broadly agree memory supply and demand fundamentals remain tight, many question whether AI memory's rising share of cloud provider capital expenditure, estimated at 52% in 2026 and projected above 70% in 2027, can hold at those levels, the same report said. Microsoft has guided full-year 2026 capital spending to around $190 billion and Meta raised its own 2026 range to $125-145 billion, TECHi reported last week. Any reversal or deferral in those commitments would weaken demand across all memory categories.

That anxiety is not abstract. Korean memory stocks sold off hard last week, with Samsung falling as much as 9% intraday and SK hynix dropping nearly 15%, before both recovered within hours. Samsung rebounded 8.7% the same session, TECHi reported. A rout and recovery that fast shows how reactive sentiment has become to any rumor of oversupply or capex softening.

There is also a cost factor that does not show up cleanly in consensus earnings estimates. In late May, Samsung averted a large-scale semiconductor strike by agreeing to direct 10.5% of the semiconductor division's operating profit into special employee bonuses. Some analysts estimate cumulative bonus provisions could exceed 40 trillion won, and the timing of when those obligations are recognized in Samsung's accounts is a material variable for the Q2 reported figure, The Star noted.

The same price increases powering Samsung's memory profits are squeezing its mobile division, where higher component costs have outpaced recent handset price increases. Analysts said further smartphone price increases may be needed in the second half of the year, The Star reported. It is a real cost, though one that sits well below the scale of the memory gains.

What the Q2 release will confirm

Samsung's Q2 preliminary results are expected imminently. The figures will likely confirm that the AI memory chip profit surge has continued to broaden, with price gains across DRAM and NAND reinforcing HBM rather than substituting for it.

The forward indicators worth tracking are DRAM and NAND average selling prices each quarter; any guidance changes from Microsoft, Meta, or other major hyperscalers on AI infrastructure capital expenditure; the precise timing of Samsung's bonus provision recognition and how it affects reported versus underlying earnings; and whether new fab capacity anywhere in the supply chain shows signs of arriving ahead of schedule.

Samsung and SK hynix last week pledged to invest a combined 3,200 trillion won in Korean chip capacity through 2040, The Star reported. That commitment signals confidence in long-term demand, but it also reflects how much the boom's duration will depend on whether cloud AI spending holds its current pace.