Yesterday, we sent you the decoder. Today, this is the call.
We published the forecast and evidence in full for Tessara members last week. After the print, we will grade ourselves against the reported numbers and add it to Tessara’s public track record.
Know someone watching this print? Forward this to them.
SK Hynix reports 2Q results on Wednesday, July 29 (Asia morning).
We made our forecast, which sits above consensus:
₩70 trillion or more: 20% probability
₩68 trillion or more: 40% probability
₩65 trillion or more: 65% probability
Central estimate: ₩66–67 trillion
(Wall Street consensus: ~₩65 trillion)
The forecast is aggressive relative to the Street. Our thesis is simple:
July traded as though memory demand had broken. We think the market is pricing the wrong failure mode.
Nothing in the KIS preview that triggered the selloff, or in the physical data that followed, supports an industry-wide demand break. The disagreement is much narrower: did blended DRAM realization land near +30%, as the bear case assumes, or closer to +40%, as we expect?
That one variable decides the print.
Part I: The Setup
July priced the wrong failure mode
Memory stocks sold off sharply in July as investors absorbed a bearish preview from Korea Investment & Securities (KIS).
KIS reduced its second-quarter estimate for SK Hynix to approximately:
₩80.9 trillion in revenue
₩60.4 trillion in operating profit
SK hynix shares responded with their worst Seoul trading session on record.
The market reaction was interpreted as evidence of a broader memory-demand problem.
But that is not what the KIS estimate actually says.
KIS still forecasts a record operating margin of approximately 74.6%, with margins expected to rise further. Its lower estimate rests primarily on a narrower assumption: SK hynix may capture less of the recent market-price increase than investors had expected.
The core assumption is blended DRAM price realisation of approximately +29%, down from KIS’s earlier estimate of roughly +50%.
The reason is contract structure.
HBM pricing is largely fixed through annual agreements. A substantial portion of conventional memory is also sold through longer-term contracts. Market prices can therefore rise faster than the prices recognised in reported revenue during any single quarter.
That is a legitimate concern.
But it is a pricing-realisation concern, not a demand-collapse thesis.
The selloff turned a company-specific pass-through question into an industry-wide demand narrative.
We think that is too bearish.
Today, the fundamental setup for SK Hynix is asymmetric. The shares have fallen 30-40% in recent weeks.
If SK hynix prints near our base case, the main earnings risk behind the selloff would be removed. The broader thesis that memory supply remains tight would still hold.
PART II: THE ARGUMENT
SK Hynix in thirty seconds:
SK Hynix is the world's second-largest memory maker and, with Samsung and Micron, one of only three companies that matter in DRAM. Memory is effectively the entire business
The company sells into three major pricing regimes:
Conventional DRAM, where pricing moves through quarterly and longer-term contracts
NAND, where pricing is more market-sensitive but still affected by contract timing and product mix
HBM, where pricing is largely fixed through annual agreements and customer-specific programmes
The quarter therefore combines three different forms of price transmission.
Conventional DRAM and NAND are moving through one of the strongest pricing environments in the industry’s history.
HBM remains structurally tight, but its reported pricing is less responsive within the quarter because contracts reset more slowly.
This distinction matters because HBM can fall as a percentage of revenue even while HBM revenue itself continues to grow.
If conventional DRAM and NAND rise faster, the denominator expands.
A lower HBM revenue share would therefore not, by itself, indicate weakening HBM demand.
Three observations support an outcome closer to our estimate:
SK Hynix's own pass-through. In Q1, conventional DRAM contract-price benchmarks rose 93 to 98%. SK Hynix reported blended DRAM pricing up in the mid-60% range, implying realized pass-through of roughly two thirds. That ratio is an anchor, not a rule. But the W60.4 trillion case requires pass-through to fall toward one half just as contracts were resetting upward at historically strong levels.
The physical export tape. Korean customs shows a strong physical quarter rather than an industry-wide demand break.
The uncovered peer. Nanya Technology, which has far less long-term-agreement cover than SK Hynix, reported second-quarter revenue up 68% from Q1. It doesn't isolate pricing, since bits, mix and currency sit inside the number. But it shows the strength implied by the benchmark reaching reported results where contract timing was less restrictive, and it narrows the remaining question to the size of SK Hynix's contract discount.
What customs establishes
Korean memory exports rose from $39.8 billion in Q1 to $62.3 billion in Q2, up 57%, with June the strongest month in the series.
Commodity DRAM's implied price per kilogram rose 58%, almost exactly matching the independent contract-price trackers, while HBM-class shipped weight increased 15%. The quarter was supported by both higher prices and higher physical shipments.
Customs does not tell us SK Hynix's reported revenue or operating profit. The data combine Samsung and SK Hynix, track exports rather than accounting revenue, and cannot isolate company-specific realization, product mix or cost conversion.
But it means an industry-wide Q2 demand collapse is difficult to reconcile with the physical export data.
PART III: The Tessara Model
Tessara is a living model of the AI buildout, with memory as one of its core systems. Its SK hynix forecast is built from three linked layers:
Physical volume
Market pricing
Internal conviction
Volume comes from the physical ramp. Our HBM line register shows existing capacity still ramping (M15X has been adding wafers since February) and no major new line reaching full output before the second half of 2027. Supply arrives gradually, not as a wave, so the company's own shipment guide is the volume forecast. The near-term risk is execution (yield, qualification, packaging), not a sudden supply wave.
Pricing comes from tightness; realization comes from contracts. The major memory categories SK Hynix sells read at or near peak tightness on our constraints board, giving suppliers substantial pricing power. Roughly half the book sells under longer-term agreements that cap this quarter's pass-through: why our price assumption sits below the benchmark, and why the gap reads as deferred rather than lost.
Conviction comes from the evidence flow. Dated equipment orders, customer prepayments, customs data and the cross-maker inventory check point toward a strong physical quarter. Our cycle watch reads tight today, with one honest amber: contract-price increases are decelerating, which is why our aggression stays in this quarter's numbers and not in forward promises.
The bridge: Q1 actuals to the Q2 estimate
(Our model allocates the DRAM guide between conventional DRAM and HBM and estimates all pricing, mix, FX and conversion inputs)
The bridge above shows exactly where the competing outcomes come from. Blended DRAM realization near +30%, with NAND near +50%, reproduces the bearish case of roughly W80 to 81 trillion of revenue and W60 to 61 trillion of operating profit. At the other end, applying Micron's recent ~98% incremental conversion to our revenue estimate produces roughly W69 to 70 trillion of operating profit.
Why 40%, and not 20 or 60
Our central estimate is about W66.5 trillion, but our uncertainty around it is skewed upward.
The largest unknown is how much of the quarter's market-price increase SK Hynix actually recognized in reported results. If realization proves stronger than our base case, three parts of the model are likely to improve together.
NAND realized pricing. We assume NAND realized pricing increased 65%, modestly below the independent tracker range of 70 to 75%. Our 65% assumption therefore recognizes the strength of the market while retaining a discount for company-specific realization.
HBM mix. A larger contribution from 12-layer HBM3E shipments, or an earlier contribution from HBM4, would raise the revenue mix. Mid-year reports of HBM4 timing revisions are why we treat the schedule as a registered mechanism check rather than building an aggressive HBM4 contribution into the base case.
Operating conversion. We assume 88 to 89% of incremental revenue reaches operating profit. That is below the roughly 98% conversion Micron recently reported. Micron's result is not a clean benchmark: its fiscal window, product mix and cost structure differ from SK Hynix's. But it demonstrates how much operating leverage is possible when pricing rises faster than costs.
The AI buildout today is a 6 sigma event in human history. We have no historical template. This up-cycle has repeatedly punished forecasts anchored too closely to history.
Our registered call
Operating profit: ₩66–67 trillion
Probability of at least ₩65 trillion: 65%
Probability of at least ₩68 trillion: 40%
Probability of at least ₩70 trillion: 20%
The setup in one line:
July priced an industry demand break. The evidence points instead to a company-specific pricing-realisation dispute inside a physically strong quarter.
It's also a test of whether a live model of physical constraints, pricing regimes and company exposure can identify where consensus is wrong/missing
That is what Tessara is being built to test. Let's see!
Full forecast and model: https://tessara.tech/views/sk-hynix-wrong-failure-mode
Cheers,
Teng
This article is for informational and research purposes only. It is not financial advice, investment advice, or a recommendation to buy or sell any security. Tessara Research does not publish price targets. The views expressed here reflect our analysis at the time of publication and may change as new evidence arrives. Readers should do their own research and consult a qualified financial adviser before making investment decisions.