This AI Storage Stock is Up 560% so Far This Year. Is it Too Late to Buy? (Hint: It's Not Sandisk)

Kioxia's revenue rose 59% in fiscal 2025 and 37% in fiscal 2026, with net income turning profitable again in 2025 and more than doubling in 2026 as AI-driven data center demand lifted memory-chip pricing and margins. Analysts expect revenue and net income CAGR of 118% and 232%, respectively, from fiscal 2026 to fiscal 2028. The stock still trades at only 8.5x this year's earnings, and the company is preparing for a U.S. ADR listing in its upcoming IPO.

Analysis

The market is re-rating memory from a pure cyclical to a quasi-infrastructure asset, but the second-order winner is not just the chip seller — it’s the vertically exposed ecosystem that controls wafer access, packaging, and capacity allocation. If AI server builds keep pulling mix toward high-density enterprise storage, the pricing power shift should persist longer than typical memory upcycles because hyperscalers are optimizing for latency and endurance, not just unit cost. That supports a multi-quarter earnings uptrend even if consumer electronics remain soft.

The key risk is that this becomes a crowded “AI-adjacent” trade before supply normalization has been fully reflected in capex. Memory history says margin spikes invite capacity additions with a lag; if gross margins stay elevated for 2-3 quarters, the market will start discounting 2027 oversupply well before it shows up in reported numbers. Thin OTC liquidity also matters: any disappointment around the U.S. listing timetable, float expansion, or lockup mechanics could hit the stock harder than fundamentals would imply.

WDC is the cleaner way to express the same thesis with better liquidity and less single-name execution risk, especially if the JV economics keep transferring incremental value to the operating-side manufacturer. NVDA benefits indirectly through AI buildout intensity, but the trade here is more about storage attach rates than compute. The contrarian view is that the market may be over-anchoring on near-term AI demand while underpricing the speed at which memory supply responds; if data center capex pauses for even one budget cycle, the multiple can compress quickly despite still-strong earnings.

Watch the IPO/listing event as a catalyst: a formal U.S. listing can broaden the holder base and trigger passive demand, but it can also expose the name to a higher-quality debate on peak-cycle valuation. Near term, the better setup is to own quality exposure before liquidity improves, then reassess after the first earnings print post-listing when expectations are most vulnerable.

Key Decisions for Investors

• Long WDC vs. short a basket of weaker memory cyclicals for a 3-6 month relative-value expression; prefer WDC for liquidity and cleaner monetization of the AI storage upcycle.
• If willing to own single-name risk, accumulate KXIAY only on pullbacks ahead of the U.S. ADR listing; target a 6-12 month hold, but size smaller than normal because OTC liquidity can amplify downside on any guide miss.
• Buy NVDA on weakness only as a derivative beneficiary of sustained AI capex; this is a lower-conviction cross-pick versus direct memory exposure, so use it as a hedge for broader AI demand rather than the core trade.
• Consider a pairs trade: long KXIAY / short INTC over 6-9 months to isolate AI infrastructure winners versus legacy PC-cycle exposure; thesis breaks if memory ASPs normalize faster than expected or if PC demand surprises higher.
• Use call spreads rather than outright longs if entering into the IPO window; the setup has upside, but multiple compression risk after listing makes capped-risk structures more attractive than cash equity.