When all investment groups are talking about storage, what should you consider

This wave of gains in the storage sector is unreasonably strong.

A month ago, SK Hynix's share price was still at 1.04 million Korean won, but today it surged to a record high of 1.97 million during trading, only to be pushed back down to 1.835 million before closing; yesterday alone it jumped by 11.5%, with nearly a 30% increase over the past week. Samsung followed behind with a slightly lower magnitude, but the upward rhythm was roughly the same. The frenzied rise brought an extremely high level of market enthusiasm, with almost all investment groups discussing these two companies, and most questionsfocused on whether it’s still possible to buy Samsung, which has lagged in gains.。

Over the past two years, AI has exploded in popularity, with the compute side seeing massive growth: NVIDIA, Taiwan Semiconductor, and Broadcom. Meanwhile, memory stocks have consistently been left behind. However, as large-scale model applications began advancing into productivity phases starting in mid-last year, token consumption skyrocketed, and AI's appetite for memory expanded accordingly. The market suddenly realized that memory-related stocks, which had been neglected over the past two years, had instead become the bottleneck in computational expansion. This consensus expectation thus formed.

For institutional investors, this situation is nothing new—just buy. However, for individual investors, before investing in these two companies now, especially considering buying Samsung, there are three questions to think through:

What do I currently hold in my portfolio? What kind of return am I aiming for by purchasing it? Should I adjust my holdings accordingly?

Among these three questions, the second one is the most fundamental. After reviewing analyses from various institutions and companies regarding this memory cycle, a relatively moderate consensus is that the peak may occur around Q4 2027 to Q1 2028. This expectation also forms the basis for our subsequent analysis. If you don’t agree with this, feel free to stop reading here.

The facts supporting this premise aren't hard to find. Upstream players are putting their money where their mouth is: SK Hynix committed 19 trillion Korean won in April to expand its packaging plant. Downstream, order flow is keeping up as well—not only is NVIDIA a well-known customer, but Microsoft and Google have long-term contracts secured, with HBM orders extending to 2028.

More crucially, the limitation lies within the underlying technical architecture. NVIDIA’s own tech blog spells it out clearly: For LLM inference, the bottleneck at the decode stage is memory bandwidth, not compute power. A 70-billion-parameter model at a 128k context consumes 40GB just for KV cache, scaling linearly with concurrency.

Memory itself is highly cyclical, a fact no one would dispute. However, the intensity of this cycle far surpasses previous ones and will likely last longer. As for whether to buy and which one to choose, the decision seems clearer for SK Hynix: the leader among the memory duopoly and the top player in HBM, with valuation and growth potential written all over its face. Samsung, however, is different. Its rise has been much lower than SK Hynix’s, making it look like an undervalued opportunity. But Samsung’s business structure is more complex than SK Hynix’s, and its lower valuation conceals certain factors.

01 Can't compare with SK Hynix.

For a long time, this publicly listed company Samsung has been perceived as the perennial runner-up: its mobile phones can't compete with Apple, its memory business is no match for Hynix, its foundry business is being crushed by Taiwan Semiconductor, and its chip design falls short against Qualcomm; additionally, due to the historical burden of Korea's old chaebol engaging in diverse businesses, low-profit segments like home appliances and displays must be included in the company, suppressing their valuation.

Yes, there are historical reasons for Samsung’s poor stock price elasticity, directly reflected in their revenue structure. The semiconductor business accounts for 61%, while the remaining 39% comes from mobile phones, displays, and home appliances—businesses that barely generate profit; however, in terms of operating profit, the semiconductor segment alone contributes 94%.In the semiconductor business' sales, memory accounts for 92%. Those miscellaneous businesses contribute nearly 40% of the revenue but only 6% of the profit.。

Here, there is a somewhat counter-intuitive situation that needs to be understood: although Samsung's stock price volatility is not as high as Hynix's, from a valuation perspective, the market has not discounted Samsung for its diversified businesses recently; in other words,in the eyes of the capital market, Samsung is essentially a pure-play memory semiconductor company.。

Looking at the P/E ratio, analysts' consensus forecast after the Q1 2026 earnings report for the 2026 fiscal year shows Samsung with a forward P/E of 6.87x, compared to Hynix’s 6.55x, making Samsung relatively more expensive; in terms of price-to-book ratio, Samsung’s is 3.85, while Hynix’s is 11.04, making Samsung appear much cheaper.However, Hynix’s return on capital (nearly 60%) is more than three times that of Samsung’s (less than 20%).This is fundamentally because single-quarter profits have not fully entered equity yet; by the end of the year, when accumulated equity is high enough, the price-to-book ratio will naturally decline.

Looking at both the P/E ratio and the price-to-book ratio together: the market values Samsung and Hynix as homogenous memory companies, with almost equal P/E ratios. Currently, Hynix's share price is close to analysts' consensus target price, while Samsung still has about 13% upside potential. This is why most people see Samsung as a catch-up investment opportunity.

From a business perspective, traditional DRAM price increases are driven by HBM siphoning off advanced production capacity, and Hynix, being the leader in HBM, benefits more directly. For AI inference demand in NAND and enterprise SSDs, SK Group’s Solidigm unit steps in. Regarding multi-year long-term contracts, Hynix has already signed a $10 billion order with Microsoft and a five-year contract with Google; shareholder returns also show room for improvement at Hynix.

Therefore, the key gap between Hynix and Samsung currently appears to bethe lead in HBM。

Where is Samsung in catching up with HBM? According to TrendForce, an authoritative research firm in the memory industry, by Q3 2025, Hynix will hold 53% of the HBM3E market share by shipment volume, Samsung 35%, and Micron 11%. However, by revenue share, Hynix holds 57%, while Samsung only accounts for 22%. The 12 percentage point revenue gap indicates that although Samsung's HBM3E has entered NVIDIA’s supply chain, it mainly secures lower-spec orders, unable to command higher prices.

The outlook for the HBM4 phase is also not optimistic. On April 23, Hynix announced a deep collaboration with Taiwan Semiconductor at the TSMC Technology Summit. The entire suite of HBM4 technologies, including the base die, 3D Stacked DRAM, and advanced packaging, are tied closely with Taiwan Semiconductor. Counterpoint’s forecast for HBM4 market shares directly adjusted Samsung back to around 28%.

However, Samsung's own guidance is quite aggressive. In its Q1 earnings call on April 30, the company gave a full-year guidance statinga threefold increase in HBM revenue year-on-year, with clear indications in the Q2 outlook that HBM4 base-die supply will increase; in February, TrendForce also cited industry sources saying that NVIDIA might relax HBM4 specifications to give Samsung some supply opportunities.

Looking at market share, revenue, technology stack, and Samsung’s own guidance together, whether Samsung can truly catch up this time in HBMthere are significant bearish and bullish factors at the factual level, and market divergence itself is substantial. Analysts took the triple growth guidance as the basis for the 'catch-up curve,' projecting a steeper acceleration curve for Samsung in the second half compared to Hynix; however, this acceleration runs in the same direction as Hynix: As the HBM cycle rises and storage realization increases, both companies are growing together. Samsung just started late, making its acceleration appear more pronounced.

At its core, the points of competition between Samsung and Hynix are quite similar. Whether viewed from a fundamental perspective or through industry analysis, if one holds Hynix, there isn't sufficient reason to switch to Samsung; if starting without any semiconductor holdings and forced to choose one, given Samsung's current price, a better option would be to wait for a pullback.

02 Differentiate between primary and secondary

Beyond memory, part of Samsung's valuation also hinges on their unique business bets: foundry services. Samsung is the only global foundry besides Taiwan Semiconductor that has mastered the most advanced processes. In this area, Samsung has been trying to emulate Taiwan Semiconductor for many years, but so far, they haven't secured any truly significant clients, except Tesla's AI6 with a cornerstone $17 billion long-term contract (signed for 2025-2033).

The 2nm order for AMD's next-generation Venice server CPU was reported by Korean media on May 11, citing industry insiders as being awarded to Samsung, though neither party has officially confirmed it; as for the manufacturing归属 of Qualcomm’s Snapdragon 8 Elite Gen 6,CEO Cristiano Amon’s statement at CES 2026 remains “in discussions.”On April 16, TrendForce also cited Korean industry sources saying that Tesla's AI6.5 (an upgraded version of AI6) had shifted to Taiwan Semiconductor, indicating even Tesla’s long-term contracts are beginning to loosen, though AI6 itself remains a cornerstone client.

During the Q1 earnings call, Samsung’s wording regarding foundry operations included polite statements devoid of numbers: 'Profit decline due to seasonal factors,' 'Continuously securing design orders for high-performance computing chips,' 'Advancing 2nm expansion with major clients,' etc.The official response avoided addressing 2nm yield rates; TrendForce's mid-April figure stood in the mid-50% range, still short of the 60% threshold required for mass production.。

The gap between Samsung and Taiwan Semiconductor in foundry services isn’t something that can be easily bridged in the short term, nor is it easier than HBM4. Having multiple business lines might seem like an additional path, but against the backdrop of surging AI demand today,It also implies the need for larger-scale R&D and production capacity investments., the return on these investments will ultimately be factored into the valuation framework of Samsung. As for the final outcome, it depends on whether they have the ability and luck to pull it off.

The following conclusion is what this article hopes readers will take away most:For the vast majority of readers who already hold any assets related to AI capital expenditure, adding Samsung to their portfolio does not essentially diversify into a new source of excess returns but rather doubles down on their storage bet.It adds another bet on whether Samsung's foundry business can succeed.

Looking across the entire market in 2026, semiconductors are almost an unavoidable track. Names like Hynix, NVIDIA, Taiwan Semiconductor, Broadcom, AMD, ASML, Marvell, and Micron all rest on the same underlying assumption: AI capital expenditure will not recede, HBM mass production will proceed smoothly, and the memory supercycle will extend to 2028. Buying all these names may sound like diversification,but it is essentially a single bet packaged into multiple positions.。

Before adding Samsung, the real question isn’t 'Is Samsung expensive?' or 'Can I get in?', but rather a few more fundamental issues: What specific assumption does this position represent? Is this assumption aligned with other holdings? Adding another position—does it amplify existing bets or open up a genuinely new risk dimension?

On the same AI capital expenditure theme, sensitivity varies noticeably: Hynix is the purest and most sensitive play, capturing the most upside from the HBM cycle; NVIDIA’s bet centers on demand for computational power, particularly relevant to inference expansion; Taiwan Semiconductor covers advanced process technology, including HBM4 base dies, 3D Stacked DRAM, and advanced packaging—all within its domain; while the seven giants ETF dilutes exposure at the group level, offering the least volatility but also the most diversified risk.

Once this layer becomes clear, Samsung is no longer just a simple option as a value play. Whether it’s worth adding depends less on how much more expensive Samsung is compared to Hynix or how much catch-up potential remains, and more on which layer of the track you're currently standing on.

03Conclusion

The current memory supercycle is set to last at least until Q4 2027, forming the foundational premise of all conclusions in this article. But determining direction is one thing; timing operations is another matter altogether.

The direction of this cycle has become clear with almost no divergence: AI capital expenditure remains strong, HBM production is proceeding smoothly, and storage giants are benefiting from elasticity on both supply and demand sides. The rhythm, however, is different—rhythm pertains to position structure, leverage, and the specific choice of entry points. Direction tells you this is a forest where you can hunt; rhythm determines whether your shot hits the target and whether you can withstand the recoil after hitting it.

As Hynix's stock price continues to rise, people around me are gradually taking profits: enjoying the benefits of the cycle while leaving room to handle potential reversals in rhythm. This is especially true for those using leveraged instruments—they need to be extra cautious. The daily compounding characteristic of leveraged ETFs amplifies returns during accelerating trends but reverses just as sharply once the peak is reached. By the time the market confirms a reversal in direction and you decide to sell, the compounding effect will have already eroded part of your gains.

Judging that this trend is still intact and determining whether the current position is still appropriate have always been two separate issues; if confused, even being correct on direction can lead to losses. $CSOP SK Hynix Daily (2x) Leveraged Product (07709.HK)$$CSOP Samsung Electronics Daily (2x) Leveraged Product (07747.HK)$$Micron Technology (MU.US)$$SanDisk (SNDK.US)$

Disclaimer: This article is intended solely for learning and communication purposes and does not constitute investment advice.

520K Views