Samsung and SK Hynix are awash with AI cash — will shareholders get more back?

The AI memory boom has transformed the earnings power of South Korea’s two biggest chipmakers. Now investors are asking a different question: how much of that cash should be reinvested, and how much should come back to shareholders?

Samsung Electronics has already guided to roughly KRW 195 trillion in third-quarter revenue and KRW 107.4 trillion (AUD $115.37 billion) in operating profit, while SK Hynix posted record second-quarter revenue of KRW 79.3 trillion and operating profit of KRW 60.5 trillion. AI servers, high-bandwidth memory and tighter DRAM supply have driven the surge.

Shareholders have benefited from the rally too. Samsung shares were up about 87% year to date, while SK Hynix has roughly doubled from its late-2025 level despite a sharp pullback from recent highs.

The next catalyst may therefore depend less on proving that AI demand exists and more on what management does with the cash it is generating. Samsung and SK Hynix both need to keep spending aggressively on HBM and advanced manufacturing, but investors are also being offered substantially larger dividends and buybacks.

The memory cycle has moved from shortage recovery into an AI-driven profit boom.

Samsung’s semiconductor division generated KRW 127.5 trillion of second-quarter revenue and KRW 89.2 trillion of operating profit, helped by record memory results and stronger HBM4 sales. The company expects server DRAM, enterprise SSD and HBM demand to remain strong as hyperscalers continue spending on AI infrastructure.

SK Hynix has benefited even more directly from its leadership in high-bandwidth memory. Its second-quarter revenue rose 257% year on year, while operating profit increased 557%. Cash and cash equivalents reached KRW 88 trillion, debt fell to KRW 18.6 trillion, and the company finished the quarter with KRW 69.4 trillion of net cash.

HBM remains central because it carries much higher value than conventional memory and is tightly linked to AI accelerators. SK Hynix began mass shipments of HBM4 during the second quarter, while Samsung has also ramped HBM4 sales and shipped HBM4E samples to major customers.

The earnings question is therefore no longer whether AI memory is important. It is how long that pricing power lasts and how much capital both companies need to preserve their position.

The numbers show why capital allocation is becoming a bigger part of the investment case.

Samsung has committed to returning 50% of free cash flow generated over 2024–2026 and now expects its 2026 shareholder return to reach KRW 90–110 trillion. That includes around KRW 30 trillion of cash dividends planned for the third quarter, with further dividends or share buybacks to be determined after the full-year result.

SK Hynix has gone further. It approved a KRW 40 trillion share repurchase and cancellation program in August and raised its shareholder-return target from within 50% of cumulative free cash flow to more than 50%. Additional dividends remain under consideration.

At the same time, neither company is backing away from investment. Samsung plans more than KRW 110 trillion of facilities and R&D spending this year, while SK Hynix recently approved KRW 54 trillion for additional DRAM and NAND capacity in Yongin and Cheongju.

The two stocks are often treated as similar AI-memory trades, but their capital-allocation pressures are different.

SK Hynix has the cleaner exposure to HBM and has built a strong position supplying high-end AI memory. That leadership is generating extraordinary cash flow, but maintaining it requires continued investment in new fabs, packaging and next-generation HBM.

Its August buyback announcement shows management believes it can do both. The company said its intrinsic value was not fully reflected in the share price even as it committed KRW 40 trillion to repurchases and cancellations.

The risk is that shareholder returns become too aggressive just as competition increases.

Samsung has a broader problem. Memory is driving profits, but the group is also funding foundry, advanced packaging, consumer electronics, mobile devices, and other long-term businesses. Its semiconductor strategy includes memory, foundry and packaging rather than HBM alone.

That gives Samsung more diversification but also more competing uses for cash.

For traders, that means the companies can react differently to similar news. A larger payout from SK Hynix could be welcomed if investors believe HBM investment remains fully funded. The same announcement could raise concerns if it appears to come at the expense of capacity expansion.

Samsung faces a similar trade-off, but with a much wider investment portfolio.

The timing is becoming more important after such strong share-price gains. Samsung has substantially outperformed the KOSPI this year, while SK Hynix has also experienced a huge re-rating as AI memory profits surged. Once a large part of the earnings recovery is reflected in the share price, investors naturally start looking for the next reason to keep buying.

Higher dividends and genuine share cancellation can provide one. South Korea’s broader corporate value-up reforms have also increased attention on shareholder returns. The government has encouraged listed companies to improve capital efficiency, raise payouts and communicate more clearly with shareholders as part of an effort to reduce the long-standing valuation discount applied to Korean equities. New dividend-tax rules introduced in 2026 have strengthened that focus further.

Samsung and SK Hynix are unusually important test cases because they are among Korea’s largest listed companies and now have the cash flow to make very large distributions without necessarily compromising their balance sheets.

A credible combination of earnings growth and higher returns could therefore support valuations even if the pace of the AI share-price rally slows.

There is an obvious counterargument. Memory remains one of the most cyclical parts of the semiconductor industry. Periods of tight supply and extraordinary profitability have historically encouraged heavy investment, only for excess capacity to pressure prices later.

HBM adds another complication because falling behind technologically could be costly. SK Hynix is expanding production through new fabs and advanced packaging capacity, while Samsung is spending heavily to strengthen HBM, foundry, and next-generation semiconductor manufacturing. The companies are not simply deciding between dividends and cash sitting idle; they are deciding how much investment is required to protect future earnings.

That makes capex quality more important than capex size alone.

If billions of won in spending secure HBM leadership and long-term supply agreements with major AI customers, reinvesting may create more value than returning the same money immediately. If capacity growth starts running ahead of demand, stronger payouts become more attractive.

The best outcome for shareholders may therefore be neither maximum capex nor maximum dividends, but evidence that management can fund the strongest AI opportunities while returning surplus cash.

HBM pricing remains the clearest operational catalyst. Continued tight supply and strong AI infrastructure spending would support margins, while any sign of slower hyperscaler demand could quickly revive concerns about the memory cycle.

Customer qualification also matters. Samsung has been pushing HBM4 and HBM4E deeper into major AI platforms, while SK Hynix needs to protect the technological lead that has underpinned its recent earnings surge.

Capital spending will be watched just as closely. Higher capex could be positive if it reflects strong contracted demand, but negative if investors believe the companies are heading towards another capacity glut.

The shareholder-return decisions then sit on top of those fundamentals. Samsung still has further 2026 payout decisions to make, while SK Hynix has already raised expectations with its KRW 40 trillion buyback and cancellation program.

For traders, the important question is whether management can increase distributions without weakening the investment needed to sustain AI-memory leadership

Trading Korean semiconductor shares with Mitrade

The capital-allocation debate adds another source of volatility to Korean chip stocks beyond memory pricing and AI demand.

Eligible Australian clients can use Mitrade CFDs to take long or short positions on available Korean semiconductor shares, allowing traders to respond to earnings, capex guidance, dividend changes or buyback announcements without owning the underlying shares.

That flexibility can be useful when the market interprets the same announcement in different ways. A larger payout may support the share price if investors see surplus cash being returned efficiently, but the reaction could reverse if traders believe management is sacrificing investment required to defend future AI growth.

Stop-loss and take-profit orders can be used to define exit levels, while pending orders allow positions to open only if a selected price is reached.

Mitrade is regulated in Australia by ASIC. Share CFDs are leveraged products, so both potential gains and losses can be magnified.

Start trading semiconductor shares in three simple steps

You might be interested in…