SK Hynix and Samsung Posted Record Earnings. The Selloff Is About What Comes Next.

The Headline Gets It Wrong - And Then Gets It Wrong Again

SK Hynix reported Q2 2026 operating profit of 60.54 trillion won, up 557% year over year, with a 76% operating margin. Samsung Electronics posted Q2 operating profit of 89.5 trillion won, up 1,814% year over year. These are not middling earnings. They are, by any historical measure, extraordinary.

Yet SK Hynix's Korean shares fell 9.6% on the earnings release day, after having already plunged 10.9% earlier in the week. Samsung's Korean stock slid 9.5%. Across the selloff that hit late July, SK Hynix, Samsung, and Micron shed a combined $462 billion in market value. The headline attribution - weak U.S. chip earnings have rattled memory stocks - gets the cause and the direction wrong at the same time.

The selloff was not triggered by weak earnings. It was triggered by three forward-looking fears: Nvidia's reported $250 billion financing backstop for an OpenAI data center project, which raised questions about how heavily chip vendors are underwriting their own customers; China's accelerating memory capabilities, highlighted by CXMT's strong IPO debut and progress on domestic DUV lithography tools; and the rise of low-cost Chinese open-source AI models like Kimi K3, which suggested future AI workloads might demand less compute intensity. All three are demand-side anxieties.

The actual story is supply-side.

The Earnings Miss Was a Timing Problem, Not a Demand Problem

SK Hynix missed analyst estimates - 60.54 trillion won in operating profit versus 64 trillion won expected, and 79.32 trillion won in revenue versus 84 trillion won expected. But the gap is not a sign that demand is softening.

Analysts attributed the shortfall to HBM4 shipments coming in below expectations, pushing revenue recognition into later periods. SK Hynix began mass shipments of HBM4 in the second quarter and plans to ramp production in the second half. HBM4 is the company's next-generation high-bandwidth memory - the stacked memory technology that powers AI accelerators from Nvidia and AMD - and its initial ramp was harder than the street's models assumed. That is a supply-constraint issue: advanced packaging yields, testing complexity, and the physical limits of moving a new product into volume production.

SK Hynix President Song Hyun-jong told investors on the earnings call that customer demand remained robust, with buyers continuing to ask for more memory. The company said both DRAM and NAND prices rose quarter over quarter. Revenue grew 257% year over year and 51% quarter over quarter. On a sequential basis, operating profit grew 61%. This is not a demand story. This is a supply ramp story - the market priced the supply ramp faster than the company could execute it.

Samsung tells the same story from the opposite direction. Samsung beat analyst estimates on operating profit (89.5 trillion won versus 88.13 trillion won expected) and reported record revenue of 171.5 trillion won, up 130% year over year. Samsung says it scaled up HBM4 sales and shipped the industry's first HBM4E samples to major customers. The company expects supply constraints to tighten further in 2027.

The Real Signal: Capex, Not Earnings

The market's reaction to record earnings is not about the quarter that just closed. It is about what the companies plan to spend in the quarters ahead.

SK Hynix said its 2026 capital spending budget is expected to reach the upper end of the 40 trillion won range, up from 30.2 trillion won in 2025. That is a roughly 32% increase in capex in a single year. The spending is anchored by the P&T7 facility in Cheongju - a $12.9 billion site that will become the world's largest dedicated HBM packaging and test facility. Construction began in April 2026 and is scheduled to finish by the end of 2027. The facility is designed to work with SK Hynix's nearby M15X fab, forming a coupled ecosystem for producing large HBM dies and packaging them into stacks.

At Computex 2026, SK Group Chairman Chey Tae-won said the company plans to double its memory wafer production capacity within five years. The scaling is weighted heavily toward HBM and advanced-node DRAM rather than commodity chips. As of Q1 2026, SK Hynix held approximately 58% of the global HBM market, with Samsung and Micron trailing at roughly 21% each.

Samsung is on a similar trajectory. The company said memory capex increased quarter over quarter as it expanded investment in its new Pyeongtaek fab and other infrastructure. Samsung has locked in long-term supply agreements with its five largest global data center customers and is close to completing deals with another five.

This is the structural question the market should be answering. Both companies are pouring capital into expanding capacity while simultaneously locking in long-term agreements with hyperscaler customers. The question is not whether demand exists today - the earnings confirm it does. The question is whether this capex trajectory will create oversupply by 2028-2029, or whether multi-year contracts will insulate pricing through the expansion cycle.

The Two-Market Split Is Emerging

The memory industry is bifurcating into two distinct sub-markets, and the split is not between SK Hynix and Samsung. It is between contracted HBM supply and spot-market memory.

SK Hynix has finalized long-term agreements - typically spanning five years - with around 10 customers. Samsung has agreements with five top global data center customers and talks nearing completion with five more. These contracts are pricing in multi-year visibility and reducing exposure to price cycles. The companies are not selling into a spot market; they are fulfilling booked capacity under long-term arrangements.

The remaining segment - conventional DRAM for consumer electronics, automotive, and non-AI server applications - remains subject to commodity pricing dynamics. That is where Chinese competition from CXMT and other domestic players could exert pressure.

The market is treating both segments as one. That is a misread. The long-term agreement base is the structural floor for SK Hynix and Samsung's AI memory revenue. The spot-market segment is the exposure to Chinese competition and cyclical demand swings. The two will not move in lockstep.

What the Nvidia Backstop Actually Signals

The reported Nvidia backstop for OpenAI's Ohio data center - a roughly $250 billion financial guarantee - triggered investor concern that chip vendors are underwriting demand that cannot stand on its own. The fear is that Nvidia is financing its own customers because organic demand cannot justify the infrastructure spend.

That reading misses the mechanics. The backstop is not a subsidy. It is a credit guarantee that leverages Nvidia's balance sheet to lower the borrowing cost for OpenAI's data center lease. Nvidia benefits because the project locks in a massive order for its chips over the facility's lifetime. The financing arrangement is a customer-retention tool, not a demand-creation scheme. It signals that hyperscaler capex is reaching a scale where traditional debt markets find the projects hard to underwrite on standalone cash flows - a function of the sheer size of AI infrastructure buildouts, not their economic viability.

The implication for memory suppliers is straightforward. If hyperscalers need vendor financing to build data centers, they also need guaranteed memory supply to fill them. That is why SK Hynix and Samsung are pursuing long-term agreements. The financing risk and the supply risk are the same risk.

Investor Takeaway

The selloff punished record earnings based on forward-looking demand fears. The actual driver of SK Hynix's earnings miss was a supply ramp delay, not a demand shortfall. HBM4 shipments were slower than the street assumed, pushing revenue recognition into the second half of 2026. Samsung's results, which beat estimates, confirm that the demand picture remains strong.

The more important question is what the capex trajectory implies. SK Hynix is increasing capex by roughly 32% this year and planning to double memory wafer capacity over five years. Samsung is expanding its Pyeongtaek fab and raising memory capex quarter over quarter. Both companies are simultaneously locking in multi-year supply agreements with hyperscaler customers.

The key issue is not whether AI demand remains healthy. It is whether the long-term contracts will be sufficient to absorb the capacity these companies are building, or whether the spot-market segment - where Chinese competition is intensifying - will face pricing pressure as new capacity comes online. The investor who understands that split understands the trade. The investor who treats SK Hynix's HBM4 ramp delay as a demand problem is reading the wrong signal.

What to watch: HBM4 production volumes in Q3 2026, SK Hynix's delayed shareholder return plan, and whether Samsung's HBM market share gains materialize as claimed. The supply ramp determines whether this is a temporary execution delay or the leading edge of a capacity glut.