Micron Q3 2026 Earnings Ignite an AI Memory Rally Across Asia: KOSPI, SK Hynix, Samsung and Kioxia Surge

Micron reported $41.46 billion in fiscal Q3 2026 revenue on June 24, up 346% year-over-year against a consensus near $35.6 billion. The order book carried the real signal: management announced 16 strategic customer agreements representing roughly $100 billion in minimum contracted revenue, structured with take-or-pay clauses across its high-bandwidth memory roadmap, and that single line did more to reprice Asian semiconductor equity than any margin figure could.

What followed was a cross-border repricing event that ran in one overnight cycle. South Korea's KOSPI, which had collapsed close to 10% two sessions earlier, opened more than 5% higher on June 25 and tripped a buy-side sidecar within minutes of the open, with SK Hynix and Samsung Electronics leading the bounce, while in Tokyo, NAND maker Kioxia broke through ¥100,000 per share. The transmission from Boise, Idaho to Seoul and Tokyo carried a lesson that extends past memory chips into every market priced on the AI capital-expenditure thesis, including the digital-asset venues that trade against the same dollar-liquidity conditions.

 

The income statement is best read as a sensor measuring a memory market that is operating under hard physical capacity limits and firm pricing, rather than as a conventional earnings beat. Revenue of $41.46 billion arrived against $23.86 billion the prior quarter and just $9.30 billion a year earlier, adjusted EPS landed at $25.11 versus roughly $20 expected, and gross margin printed a company record near 84.9% as combined data-center revenue cleared $25 billion in the quarter; an annualized run rate above $100 billion with DRAM carrying about 76% of the total mix.

 

 

Margin expansion of that magnitude does not come from operational efficiency but from scarcity, which is why CEO Sanjay Mehrotra's repeated statement that the company can satisfy only a fraction of medium-term demand matters more than the margin itself; a structural supply deficit hands the seller pricing control across the entire contract book. When a manufacturer pre-sells its calendar-2026 HBM in full and then signs $100 billion of minimum-revenue floors on top of that, the cyclical question that has governed memory valuation for two decades, namely when the cycle rolls over, gets pushed out beyond the visible horizon.

Micron is the only U.S.-domiciled HBM manufacturer, which makes its filings a live proxy for two companies that do not report on the same calendar, SK Hynix and Samsung, and the market does not wait for Korean earnings to confirm the read, it prices Micron and front-runs the implication immediately.

Two sessions before the print, the KOSPI fell roughly 10% from an all-time high near 9,115 and hit circuit breakers as Samsung and SK Hynix each dropped more than 12%, and the reason the index fell that hard is structural rather than sentimental, because Samsung and SK Hynix together account for close to half of the KOSPI's entire market capitalization, which means that when both memory names move in tandem the index carries no internal hedge and correlation becomes the dominant risk factor.

That same concentration ran in reverse on June 25, when the KOSPI opened about 2.7% higher and extended past 5%, approaching 9,000 intraday and triggering a buy-side sidecar the exchange's program-trading halt that fires when KOSPI 200 futures move 5% or more as SK Hynix jumped roughly 9–12% and Samsung gained around 4–5%. A cold reading requires one correction to the easy narrative, however, because this was not a clean V-shaped recovery: foreign investors remained heavy net sellers and extended a multi-session outflow streak while domestic retail and institutions absorbed the supply, the early surge was not fully held into the close, and two non-Micron catalysts reported Samsung buyback speculation and SK Hynix's U.S. listing news contributed materially to the bid, which means attributing the entire move to one earnings report would misread a multi-factor rally that Micron triggered without single-handedly driving.

Tokyo told a parallel story built on different inputs, as the Nikkei 225 climbed back toward its record near 72,781 led by memory and chip-equipment names, but a weakening yen and a sharp drop in crude Brent fell under $73, its lowest since February were independent tailwinds for an import-dependent economy and both pre-dated the Micron release. Kioxia, the NAND maker that recently overtook Toyota as Japan's most valuable listed company, rose roughly 7–8% and pierced ¥100,000 per share, while Tokyo Electron, a wafer-fab equipment supplier, gained about 8% on the logic that an extended AI capex cycle lifts equipment orders, and both Advantest and SoftBank followed.

The more durable consequence concerns capital formation rather than price, because a persistent gap between U.S. and Asian equity multiples for identical exposure is now pulling Asian hardware giants toward Wall Street listings. SK Hynix moved to raise roughly $29 billion through a U.S. American Depositary Receipt offering reports place the figure between $26 billion and $29.4 billion issuing new shares with trading targeted for mid-July pending regulatory clearance, while Kioxia has signaled its own U.S. depositary-share plan reportedly aimed at spring 2027.

The arbitrage is straightforward, since American AI capital assigns a richer multiple to memory exposure than Asian markets do for the same underlying cash flows, and bringing equity to where the deepest institutional liquidity sits is a direct attempt to close that discount. There is a second-order read here for anyone tracking global liquidity, because the AI hardware complex is now actively competing for the same pool of dollar liquidity that funds risk assets broadly from equities, credit, and the digital-asset markets that have spent this cycle trading as a high-beta expression of dollar-liquidity conditions which means that when $29 billion of new memory paper targets U.S. desks, it draws from a finite allocation that other risk assets are simultaneously bidding for.

Every contracted-revenue floor and sold-out-supply claim deserves the opposing case stated with equal weight, beginning with the fact that cyclicality has not been repealed: memory remains the most cyclical segment in semiconductors, and while take-or-pay clauses smooth the near term, they do not abolish the cycle, because coordinated capacity additions across Micron, SK Hynix, and Samsung are the historical mechanism by which every memory upcycle ends, and the supply that feels scarce today is being built right now.

Concentration compounds that exposure, since a KOSPI that draws roughly half its value from two correlated names is a structurally fragile index, and the June 23 crash and the June 25 rebound are the same risk factor expressed with opposite signs, an investor long the KOSPI for AI exposure is in practice running a concentrated two-name memory bet wearing the costume of a diversified index. The contracted order book introduces a further dependency, because the $100 billion in minimum revenue is only as sound as the hyperscalers standing behind it, and take-or-pay obligations assume those counterparties remain willing and able to pay through any moderation in AI capex; if the spending thesis driving Microsoft, Meta, Amazon, and Google softens, the contracts get tested at precisely the moment the spot market also weakens, layering correlated counterparty risk on top of correlated price risk. The foreign-flow divergence reinforces the caution rather than resolving it, since a rebound bought by Korean retail and institutions against persistent foreign selling carries a specific failure mode whenever domestic leverage is the marginal buyer absorbing foreign exits.

 

The semiconductor sector can no longer be modeled through isolated domestic lenses, because a constrained hardware ecosystem with multi-year supply lockups transmits a guidance revision in Idaho into the equity risk premium in Seoul and Tokyo inside a single overnight session, and the Micron print both stabilized the KOSPI after a violent correction and reinforced that physical AI hardware still sits at the base of the entire trade, beneath the models, the tokens, and the narratives layered on top of it. For the crypto-native reader the connection is liquidity rather than analogy, since the same dollar-liquidity backdrop that funds $29 billion ADR raises and bids up HBM suppliers also sets the tide for digital assets, which means that when the hardware layer absorbs capital this aggressively, the liquidity it consumes is worth tracking directly.

Revenue of $41.46 billion (up 346% YoY), adjusted EPS of $25.11, and a record gross margin near 84.9%, reported June 24, 2026, with both figures beating consensus of roughly $35.6 billion in revenue and about $20 in EPS.

Micron is the only U.S.-based HBM manufacturer and trades as a real-time proxy for SK Hynix, Samsung, and Kioxia, so strong guidance signaled that AI memory demand and pricing remain firm and prompted an immediate repricing of its Asian peers.

It opened about 2.7% higher on June 25 and extended past 5%, triggering a buy-side sidecar two sessions after a roughly 10% crash, though the gains were volatile and only partly sustained as foreign investors net sold against domestic buying.

Micron disclosed 16 strategic customer agreements representing roughly $100 billion in minimum contracted revenue, using take-or-pay structure across its HBM roadmap, which transfers demand risk to the customer and converts volatile revenue into something closer to contracted cash flow.

Revenue of $50.0 billion plus or minus $1.0 billion, against roughly $43 billion expected, with gross margin near 86%.

To arbitrage the multiple gap between U.S. and Asian markets for identical memory exposure, with SK Hynix targeting a roughly $29 billion ADR raise and Kioxia signaling a U.S. depositary-share plan for 2027, both seeking deeper dollar liquidity.

Cyclicality, because coordinated capacity additions across the three major memory makers historically end every upcycle, and the contracted-revenue floors depend on hyperscaler counterparties continuing to pay through any AI capex slowdown.

Indirectly, through dollar liquidity, since large memory ADR raises and an AI hardware complex absorbing capital draw on the same liquidity pool that funds risk assets, including digital assets that trade as a high-beta expression of liquidity conditions.