The global NAND flash leader, Kioxia Holdings Corp., based in Japan, released a relatively disappointing profit outlook on Friday, with its core performance metrics for the same period also falling short of market consensus. This indicates that the increasingly fierce competition among Samsung, SK Hynix, Kioxia, and Sandisk in the NAND flash market, along with the unprecedented surge in NAND flash prices driven by strong AI training and inference computing demand, may be showing signs of easing. Following the announcement, Kioxia shares traded on the PTS system at approximately 44,100 yen, down about 5.2% from the official closing price, suggesting the market interprets the results as strong growth that still missed extremely optimistic expectations.
For the April-June quarter, Kioxia reported total revenue of approximately 1.7671 trillion yen, a massive year-on-year increase of 415.5%. SSD-led NAND storage revenue reached about 1.1747 trillion yen, up 440% year-on-year, while smart device revenue was around 525.7 billion yen, a 565% increase. Operating profit for the quarter was roughly 1.27 trillion yen, surging about 28.3 times compared to the same period last year, with an operating margin of approximately 71.9%. Sequentially, revenue grew by about 76.2%, operating profit by 112.8%, and net profit attributable to shareholders by 106.6%. The company's management clearly stated that the primary driver was a significant rise in NAND average selling prices (ASPs) fueled by demand from generative AI data center customers.
Where to begin
The latest performance data from Kioxia underscores that the AI data center construction boom continues to significantly boost NAND demand, and with the BiCS-10 technology entering the sampling stage, the structural demand for high-capacity, high-bandwidth, low-power NAND flash memory from AI data centers remains strong. However, both the current quarter's results and the subsequent guidance fell short of the market's highly optimistic expectations, signaling that the slope of NAND price increases may be slowing. This report can be characterized as one with "extremely strong fundamentals but cooling marginal expectations" – a major positive for Kioxia's medium-to-long-term competitiveness, but not a simple clear-cut positive for short-term stock prices. Nevertheless, it provides a fundamental floor and confirms industry prosperity for the semiconductor rebound.
Why only 10 ASX 200 shares?
In the April-June quarter, AI data center demand, NAND ASPs, and the SSD business all exploded simultaneously, leading to a quantum leap in profitability and cash flow. However, the market had already priced in a near-perfect storage super-cycle, and the next quarter's operating profit guidance of 1.89 trillion yen slightly missed the consensus estimate of 1.95 trillion yen. Therefore, the core contradiction in trading has shifted from "whether AI-driven demand exists" to "whether prices and profits can continue to exceed expectations." As NAND flash memory evolves from its traditional role as a "cold data/capacity storage" asset to an "extended quasi-memory layer" in the AI inference era, bullish sentiment on Wall Street for Kioxia remains strong. Nomura Securities previously maintained a "Buy" rating on Kioxia, raising its target price from 115,000 yen to 126,000 yen, based on the core judgment that NAND bit prices have consistently exceeded expectations and supply-demand tightness has not substantially eased. However, this target price was released before the latest earnings report and does not reflect the new information that the 1.89 trillion yen guidance fell short of consensus.
Despite the 800 billion yen buyback, competition anxiety is palpable: Samsung, SK Hynix, and Sandisk are all ramping up NAND capacity expansion, and the battle for NAND market share is escalating. Global investors had high hopes for Kioxia as a major beneficiary of the record-breaking wave of AI data center construction. The company, a former chip business within Toshiba, supplies high-performance NAND storage chips for data center servers, and its growth is closely tied to the expanding capital expenditures of big tech companies like Meta Platforms and Alphabet. However, Kioxia faces larger and better-funded competitors. Samsung Electronics and SK Hynix are expected to launch next-generation NAND storage chips next year and plan to accelerate NAND capacity expansion. The major challenge for Kioxia is to catch up with its Korean rivals and US-based NAND giant Sandisk in both production volume and NAND flash performance. The company has previously stated it plans to expand capacity only slightly above the industry growth rate to avoid flooding the market. But if Samsung and SK Hynix, currently focused on expanding DRAM and HBM capacity, shift their capital expenditures toward NAND, Kioxia could face market share loss.
Absolute growth still impressive, but marginal growth under fierce competition becomes a risk
Kioxia management forecasts revenue of 2.39 trillion yen for the July-September quarter, a sequential increase of 35.2%; operating profit is expected to be 1.89 trillion yen, up 48.8% quarter-on-quarter; and net profit attributable to shareholders is projected at 1.27 trillion yen, a sequential rise of 50.8%. This is not a recessionary signal for AI computing demand, but rather a continuation of the storage chip super-cycle. What truly makes the market cautious is that the 1.89 trillion yen profit forecast is about 3.1% below the average estimate of 1.95 trillion yen from eight analysts surveyed by LSEG, implying that while NAND prices continue to rise, the pace of increase may not sustain the extremely optimistic expectations. The stock price reaction must be analyzed carefully: Kioxia closed at 46,500 yen in Tokyo on July 31, up 17.72% and hitting the daily limit, but the earnings were released after the market close at 3:30 PM Japan time. Therefore, the limit-up move mainly reflected the global semiconductor rebound and pre-earnings short covering, not the earnings themselves. However, after the earnings and outlook, shares traded on the PTS at around 44,100 yen, down about 5.2% from the official close. This indicates that the market views the report as "strong results but not enough to surprise," rather than an undisputed major positive. Meanwhile, the buyback of up to 800 billion yen (up to 30 million shares) and the 1-for-3 stock split effective October 1 will improve per-share supply and liquidity but cannot replace the market's updated judgment on the NAND price cycle.
Kioxia – the quintessential beneficiary of the storage chip super-cycle
Large AI training sets, model weights, checkpoints, vector databases, RAG corpora, multimodal data, logs, and inference results all require long-term residence in high-capacity storage. During training, massive data needs to be continuously fed into GPU clusters from object storage and local NVMe SSDs. The inference era further generates vast amounts of KV Cache, long contexts, agent states, and retrieval data. HBM handles the highest bandwidth "hot data layer," DRAM serves as system working memory, while enterprise NAND SSDs take on the much larger capacity, lower-cost "warm data and persistent layer." Therefore, NAND does not replace HBM but expands alongside HBM and DRAM in the AI server storage hierarchy. Notably, NVIDIA is driving the expansion of some ultra-large, short-lived KV Cache from expensive GPU memory to rack- or cluster-level flash layers. Kioxia's latest CM10 series is specifically designed for AI inference, KV Cache, and NVIDIA's CMX context storage solutions, featuring PCIe 6.0, NVMe 2.1, and BiCS-10 technology. It offers capacities from 1.6TB to 61.44TB, with sequential read performance up to 92% higher than the previous generation and random read performance up to 85% higher, while supporting cold plate liquid cooling. This means enterprise SSDs are evolving from traditional back-end storage into computing infrastructure that directly impacts GPU utilization, first-token latency, and inference throughput.
Kioxia's tenth-generation BiCS FLASH, BiCS-10, uses 332-layer 3D NAND, 1Tb TLC cells, CBA wafer bonding, and an OPS structure, achieving a NAND interface speed of 4.8Gb/s, a 33% improvement over the eighth generation. Bit density is increased by 59%, and write and read energy efficiency are improved by 18% and 30%, respectively. The economic significance of these improvements is not simply "more layers," but the ability to produce more bits per wafer while reducing power consumption, cooling costs, and server rack space per unit capacity, thereby improving Kioxia's bit cost and customer total cost of ownership. BiCS-10 has entered the sampling stage and is planned for production at the Kitanihon Fab2 facility, initially targeting enterprise and data center SSDs. The CM10 is a PCIe 6.0 enterprise SSD series for AI data centers, representing a downstream product that utilizes BiCS-10 flash memory.
More notably, NAND flash memory is being upgraded from its traditional position as a "cold data/capacity storage" asset to an "extended quasi-memory layer" in the AI inference era. It is likely to become one of the most important frontier technology trends in the storage chip industry, following the HBM super storage system. The emergence of the HBF (High Bandwidth Flash) technology route, built from NAND flash, further reinforces this view. Sandisk, SK Hynix, and Samsung have explicitly defined HBF as a new NAND form factor targeting the AI "memory wall," aiming to provide larger capacity for AI inference. They claim HBF can achieve performance close to "unlimited capacity HBM" in relevant inference tests while significantly increasing usable memory capacity.
Overall, this earnings report and outlook are a major positive for Kioxia's fundamentals – revenue structure, earnings elasticity, cash flow, and technology roadmap are all validated. However, for the short-term stock price, it is a positive that provides a "floor" rather than an immediate catalyst for re-rating. For the stock to truly re-enter an uptrend, the next quarter's actual operating profit needs to exceed the consensus estimate of around 1.95 trillion yen, and the concurrent realization of BiCS-10 production volume, yield rates, and hyperscaler customer orders must be confirmed. Nomura previously stated that for its 126,000 yen bull case to materialize, three conditions are needed: continued increases in NAND contract prices, Samsung and SK Hynix continuing to allocate most new capital to HBM and DRAM rather than aggressively expanding NAND, and BiCS-10 and CM10 successfully winning market share in the next-generation NAND competition against Samsung and Hynix, along with passing certification from US hyperscalers and converting that into multi-year orders. Therefore, the 126,000 yen target represents a bull scenario where the storage super-cycle persists and Kioxia captures AI enterprise SSD market share, not an unconditional base case valuation.
Comments