Norimura's Rebel Call on Samsung and SK Hynix Could Rewrite the Memory Cycle

A Contrarian Call From the Top

A chorus of Korean brokerage houses has been lowering price targets for Samsung Electronics and SK Hynix all season, trading on a simple thesis: the memory upcycle is topping out, and a brutal downcycle is approaching. Then Nomura Securities and KB Securities walked in and said almost nothing had changed — except that the market is badly wrong.

Nomura held its buy ratings on both names, setting targets of 670,000 won for Samsung and 4.7 million won for SK Hynix. KB Securities kept its Samsung target at 600,000 won. Both firms argued that the structural forces reshaping the memory industry — AI-driven demand, long-term supply agreements, and compressed cycles — make a repeat of the vicious downturns that punctuated every previous generation of memory markets unlikely, if not impossible.

The headline number to watch: the reports suggest the downcycle may simply not arrive.

The Supply Squeeze Won’t End Anytime Soon

Nomura’s central argument rests on a supply-demand calculation that most short-term bearish analysts are underweighting. The firm estimates that global memory production capacity would need to double to 7.2 million wafers per month within four years — and triple to 11 million within six — just to meet projected demand from AI infrastructure buildouts. Even aggressive capacity expansion by Chinese manufacturers would not meaningfully close that gap.

The result: a supply shortage that Nomura sees persisting through at least 2028.

This is not a normal memory cycle. In previous downturns — the deep slumps of 2001, 2009, and 2016 — prices collapsed because excess capacity flooded the market while demand evaporated. Buyers waited. Suppliers panicked. Everyone lost margin. This time, the demand side is fundamentally different. Hyperscalers are not shopping around for the cheapest DRAM or NAND. They are locked into securing volume for AI data centers that are being built on timelines measured in quarters, not years.

The LTA Revolution

What is actually changing the industry is the rapid adoption of long-term agreements — contracts that lock in supply and price for periods of up to five years. Major customers, including Google and Amazon, are now joining Chinese giants like Baidu, Alibaba, and Tencent in demanding multi-year supply commitments. And they are willing to pay a premium to get them: Nomura reports that LTAs are being struck at ceiling prices up to 20 percent above prevailing spot rates.

These deals carry something unprecedented in the memory business: advance payments ranging from 20 to 30 percent of contract value, backed by stiff penalty clauses for either side that backs out. For Samsung and SK Hynix, this means a level of earnings visibility and revenue stability that the industry has never experienced. The classic memory boom-bust rhythm — where a single quarter of weak demand could erase billions in expected revenue — is being dampened, not eliminated, but materially reduced.

A veteran securities executive told Hankyung that this structural shift may be the most important factor re-rating these stocks, one that outweighs traditional cyclical metrics like spot price movements or inventory buildups.

Valuation Still Looks Cheap

Despite the recent rally in Korean chip stocks, both firms argue valuations remain deeply compressed. Samsung and SK Hynix shares are trading roughly 37 percent below their highs, and on a forward P/E basis, the average sits around 3 times earnings. For companies generating record profits with improving margins and expanding shareholder returns, those multiples are difficult to justify under any normal framework.

KB Securities projected that Samsung’s third-quarter operating profit could reach 107 trillion won and its full-year profit could exceed 221 trillion won — a year-over-year increase of roughly 587 percent, extending a streak of five consecutive quarterly profit records dating back to the fourth quarter of last year. The firm also flagged a potential inflection in Samsung’s foundry business: the ramp of 4-nanometer LPU production and improving yields could push the segment back into profitability as early as the third quarter, even before counting performance-based incentives.

Shareholder Returns as a New Anchor

Perhaps the most striking element of the bullish case is the focus on capital return. KB Securities calculated that Samsung has up to 110 trillion won in remaining shareholder-return capacity over a three-year window. Of that, roughly 70 trillion won could flow to cash dividends and 40 trillion won to share buybacks. If Samsung meets the conditions for Japan’s split-rate dividend tax regime and maintains a payout ratio of at least 25 percent, the cash dividend component could expand further.

On a half-year basis, KB sees the dividend yield climbing above 4 percent. That is not a speculative growth stock story. That is a mature industrial company returning cash at a pace that forces investors to reconsider whether these are cyclical memory plays or something closer to infrastructure businesses with pricing power.

The Won Risk Is Real But Overstated

Both firms acknowledged that a strengthening won poses a near-term headwind. SK Hynix’s third-quarter profit estimate was already trimmed from 86 trillion won to 77 trillion won to account for currency effects, and Samsung’s estimate was set at 107 trillion won. The mechanics are straightforward: these companies invoice in dollars but spend roughly 20 percent of revenue in won, so a 10 percent appreciation in the currency erodes operating profit by roughly 12 percent.

But Nomura pushed back hard on the macro concern. It argued that a stronger won is actually a symptom of the semiconductor supercycle itself — massive trade surpluses from chip exports will naturally bid up the currency. More importantly, it argued that the steady rise in average selling prices driven by HBM demand will more than offset any FX drag. In other words, the very strength of the memory market that generates profits also generates the currency pressure, and the two are not in conflict.

Who This Matters For

For global investors, the implication is larger than any single stock. If the memory cycle truly is entering a lower-volatility, higher-floor regime — sustained by multi-year contracts rather than quarterly spot-market gambling — then the entire valuation methodology applied to memory producers needs revision. Cyclical discounting assumes periodic crashes. The LTA model removes the assumption that crashes are inevitable.

Samsung and SK Hynix are the canaries in this new coal mine. Their pricing power, their contract structures, and their shareholder return commitments are the visible proof points. If the thesis holds, it changes how the market prices not just Korean memory stocks but the entire East Asian semiconductor supply chain — from equipment makers to packaging suppliers to the Chinese hyperscalers locking in supply.

If the thesis fails — if demand softens faster than expected, if LTAs break down, if Chinese capacity expansion accelerates beyond projections — then the current valuations look dangerously optimistic.

The market is still deciding which version of reality is closer to the truth. Nomura and KB Securities are placing their bets on the former. The rest of Wall Street has not yet caught up.