SK Hynix options trading has launched—should we jump into memory stocks?

牛牛課堂

joined discussion · ·

Hunting Japanese Stock Market Winners | Surpassing Toyota Motor to Become Japan's Most Valuable Company: How AI Is Reshaping Kioxia's NAND Super Cycle?

A soaring myth in the memory/storage sector is sweeping the globe. Yet, while all eyes are fixed on the wealth-generating rallies in U.S. and Korean equities, Japan’s capital market is quietly nurturing an unprecedented 'storage revolution.'

On June 12, 2026, this undervalued super stock— $Kioxia Holdings (285A.JP)$ —surged past Toyota Motor with unstoppable momentum to become Japan’s most valuable company by market capitalization.

This isn’t just ordinary sector rotation—it’s a historic shift in market weightings: tech semiconductors are replacing traditional manufacturing as the new pricing anchor in Japan’s capital markets.

Goldman Sachs upgraded Kioxia in early June from 'Neutral' directly to 'Buy,' raising its price target sharply from ¥48,000 to ¥93,000; Morgan Stanley took an even more aggressive stance, hiking its target from ¥80,000 to ¥155,000 on June 19 and naming it a top pick among Japanese equities;JPMorgan has now gone even further, nearly doubling its price target for Kioxia from ¥80,000 to ¥155,000.It’s rare for three top-tier investment banks to simultaneously turn bullish on a single NAND manufacturer—the core thesis behind this move boils down to just one word:AI.

In this article, we take a deep dive into this company, which has surged more than7500%The 'King of the NAND Cycle'—is it truly AI's chosen one, or merely the final dance of a cyclical frenzy?

Core moat: Why AI inference—the 'hard drive' of AI—can't do without Kioxia

One sentence to understand Kioxia’s business

Understanding AI data center architecture through the simplest analogy:

The 'ultra-high-speed compute layer' in training and inference relies on HBM (High Bandwidth Memory), similar to a computer’s RAM

Meanwhile, the storage layer—which holds massive datasets, model weights, and KV Cache inference context—relies on NAND (SSDs), equivalent to a 'hard drive'

Within this entire AI data flow system, Kioxia is one of the world’s core suppliers of NAND flash memory. Its technology originates from Toshiba’s memory division and has been deeply involved in the evolution of 3D NAND technology for over thirty years.

Its core value lies not in 'compute power,' but in being an irreplaceable storage layer that absorbs the deluge of AI data.

Moat #1: AI inference is redefining NAND’s role

Traditionally, NAND served only as 'cold storage' (for low-access-frequency data archiving). However, with the explosion of AI inference workloads—especially the rise of RAG (Retrieval-Augmented Generation) and Agentic AI—NAND is evolving from cold storage into 'warm/hot storage.' Large AI models now require frequent access to vast external knowledge bases, making enterprise SSDs (eSSDs) a critical I/O performance component in GPU servers.

Kioxia has built a differentiated SSD product portfolio centered around AI data centers:

CM Series (TLC technology): High-bandwidth read/write performance, optimized for rapid AI model loading and high-frequency RAG retrieval.

GP Series (XL-FLASH): Ultra-low latency architecture that effectively bridges the latency gap between conventional SSDs and DRAM, compatible with next-generation AI server storage architectures.

LC Series (QLC technology): Ultra-high capacity, offering industry-leading 245TB storage to efficiently address the cost challenges of storing massive AI training datasets.

According to Tech Insights, demand for data center flash memory is projected to surge from 295 exabytes (EB) in 2025 to 1,807 EB in 2028, with AI-related demand accounting for the core portion of this data center growth and delivering a compound annual growth rate (CAGR) of 46%.

Moat #2: Historic supply-demand mismatch ushers in a profit 'golden era'

This NAND cycle is dubbed a 'super cycle' not only because of explosive demand, but more critically due toextreme supply-side discipline。

In 2022, Kioxia invested heavily—1 trillion yen—to expand its fabrication capacity, only to face a sharp post-pandemic demand downturn, resulting in five consecutive quarters of losses. This painful experience directly shaped its current conservative strategy:Capital expenditures over the next three years will remain 10% below historical peak levels, with average annual capital expenditures of approximately JPY 470 billion, primarily focused on improving the efficiency of existing cleanrooms rather than building new fabs from scratch.

According to TrendForce data, NAND contract prices in Q2 2026 are expected to rise 70%–75% quarter-over-quarter; SLC NAND average prices have already increased by 130%–150% cumulatively in the first half of the year. However, large-scale new capacity additions won’t come online untillate 2027 to 2028.

Moat #3: Long-term agreements (LTAs) lock in revenue visibility

Amid tight supply, hyperscale cloud service providers are rushing to sign long-term supply agreements. Kioxia President Yuuki Ota has explicitly set a target:to lock in 50% of its shipments through multi-year LTAs by 2028. This provides exceptionally high certainty for the company’s revenue and cash flow over the next several years.

Meanwhile, the company is raising its target for data center revenue share toover 60% by FY28, eSSD (enterprise SSD) sales are expected to rise from 34% in 2025 to67%, significantly improving the product mix and ASP.

Moat #4: Technology remains firmly in the top tier

Kioxia's BiCS FLASH 3D NAND technology has already reached its eighth generation in mass production,with tenth-generation samples scheduled for delivery this summer. Its proprietary CBA (CMOS directly Bonded to Array) technology is reportedly about four years ahead of competitors, enabling more cost-efficient production of high-performance chips.

In terms of capital allocation, Kioxia plans average annual capital expenditures of approximately JPY 470 billion from FY26 to FY28, a significant acceleration compared to FY25’s actual capital spending of JPY 283.7 billion (equivalent to roughly JPY 45 billion in FY26 terms, representing a ~60% year-over-year increase). Spending will focus on ramping up Gen.10 production, building infrastructure for next-generation processes, and strengthening back-end processes.

Regarding R&D investment, the company plans an average annual R&D budget of approximately JPY 23 billion over the same period, a ~63% increase from FY25, with a focus on BiCS FLASH Gen.10/Gen.11 development, ultra-high IOPS SSDs, and novel memory devices such as OCTRAM.

Financial Fundamentals and Capital Strategy: Data-Backed 'Explosive Growth'

For the fiscal year ending March 31, 2026, Kioxia reported revenue of JPY 2.337 trillion, up 37% year-over-year, and operating profit of JPY 870.4 billion, a ~93% year-over-year increase—both record highs for the company and exceeding analyst consensus estimates. Fourth-quarter operating profit reached JPY 596.8 billion, surpassing Bloomberg’s expectation of JPY 519.3 billion, primarily driven by a significant increase in average selling prices—despite a decline in shipment volume (measured by storage capacity) during the quarter.

Kioxia simultaneously released its business outlook for the first quarter of fiscal year 2027 (April to June 2026), forecasting operating profit of JPY 1.3 trillion (approximately USD 8.2 billion), a 117% increase from the previous quarter and nearly 50% above Bloomberg’s consensus estimate of JPY 874.1 billion. Net profit is expected to reach JPY 869 billion, significantly exceeding the market expectation of JPY 612.7 billion. The company stated that robust data center demand is expected to persist, supporting continued sequential growth in both revenue and profits.

Notably, Kioxia’s rapid expansion in NAND market share has been partly driven by structural shifts in the competitive landscape.In recent years, Korean manufacturers such as Samsung Electronics and SK Hynix have redirected more resources toward high-bandwidth memory (HBM)—a critical component for advanced AI processors—enabling Kioxia to capture additional orders in the NAND market.

In addition,Kioxia announced it has officially commenced preparations for a U.S. listing, planning to issue American Depositary Shares (ADS) based on common stock and list on a U.S. securities exchange.Formerly Toshiba’s chip business, Kioxia completed its initial public offering on the Tokyo Stock Exchange in December 2024. This move to prepare for a U.S. listing is seen as a key step to broaden its international financing channels during the current AI-driven storage demand cycle.

How are Kioxia and SanDisk leveraging 'coopetition' to stage a comeback in the AI storage race?

In the first half of the year, $Kioxia Holdings (285A.JP)$ And, $SanDisk (SNDK.US)$ became the standout 'top gainer' in global capital markets, completely upending traditional valuation logic.

These two companies are the leading giants in the global NAND flash memory sector and have maintained a deep strategic partnership for over 25 years.The core of their joint venture model lies in sharing substantial equipment and R&D expenditures, leveraging economies of scale to maintain technological and cost advantages in the highly competitive memory market. Specifically, according to Huachuang Securities, the joint factory structure between these two companies is as follows:

Joint development activities:Kioxia receives compensation for manufacturing services and secures continued supply to SanDisk; SanDisk pays for these manufacturing services and ensures continued supply from Kioxia. As compensation for the company’s provision of manufacturing services and ongoing product supply to SanDisk, the company will receive total cash payments of $1.165 billion (approximately JPY 178.2 billion) over a four-year period from 2026 to 2029, averaging approximately $290 million (approximately JPY 45 billion) per year.

The company plans to recognize approximately $131 million (approximately JPY 20 billion) in additional annual revenue over a roughly nine-year period from February 2026 to December 2034, all of which will be directly recorded as operating profit. Previously, foundry revenue from SanDisk followed a cost-plus model, with annual revenue totaling approximately JPY 200 billion; this has now shifted to a model featuring explicit compensation for manufacturing services.

Capacity allocation: Kioxia owns approximately 20% of its own capacity, and within joint venture capacity, about 40% is allocated to Kioxia and about 40% to SanDisk, resulting in an overall wafer production capacity split of approximately Kioxia:SanDisk = 6:4. Kioxia maintains 100% control over wafer manufacturing at both the Yokkaichi and Kitakami fabs and independently manages all elements required for fab operations and production, including procurement of raw and indirect materials and equipment, optimization of automation systems, reduction of wafer processing time and production cycle times, and production and yield management.

Shareholding Structure:Kioxia holds a 50.1% stake in each of the three NAND flash joint ventures, with SanDisk entities holding the remaining 49.9%.

Extension of joint venture contract term:The joint venture contract term has been extended from February 2026 (current) to December 2029, and further extended to December 2034. The original expiration date for the Yokkaichi fab was December 31, 2029; the new expiration date is December 31, 2034. The current expiration date for the Kitakami fab is December 31, 2034.

Summary

Kioxia’s current capital narrative is undoubtedly highly compelling: the AI inference boom has created unprecedented structural demand growth for NAND,叠加 extremely disciplined supply-side behavior has enhanced pricing elasticity, long-term agreements (LTAs) provide a performance safety net, and the initiation of dividends has opened the door for institutional capital allocation. In just one year, Kioxia surged to become the largest company by market capitalization on the Japanese stock market—a textbook example of 'value re-rating.'

However, amid widespread market enthusiasm, Bernstein has maintained its 'underperform' rating, assigning a target price of only JPY 40,000. Its core bearish thesis points directly to underlying risks in the memory market:

Cycle peak warning: NAND prices are expected to peak in the first half of 2027, and by the end of 2028, excess profits will be erased, with gross margins inevitably reverting to their historical mean of 35%.

Long-term agreements offer no real defense: The much-hoped-for long-term contracts cannot truly insulate against cyclical downturns. Current floor prices in these agreements are only 10% below prevailing market quotes; if NAND prices fall by more than 37%, customers would find it economically rational to default rather than fulfill their contracts.

Competitive landscape shifting: Rivals such as Yangtze Memory Technologies (YMTC) are accelerating capacity expansion at a pace exceeding market expectations and are projected to become the world’s third-largest NAND supplier by 2028, inevitably reshaping the pricing structure.

Undeniably, strong cyclicality is deeply ingrained in the DNA of the memory chip industry. Historical evidence shows that gross margins exceeding 65% have never been sustained over the long term—Bernstein’s warning is far from alarmist.

For investors, the most prudent strategy right now is to strike a balance during the market consensus shift—from the myth of a 'perpetual supercycle' to 'peak-cycle panic.' Stay clear-headed amid the perfect AI narrative and maintain a healthy dose of caution when the market turns greedy.

Lastly, NiuNiu previously“Memory sector faces correction! Jensen Huang urgently ‘recharges’ market confidence—what’s next?”compiled an overview of the memory supply chain—fellow investors interested can click to follow:

5.3M Views