Nearline HDD Capacity Up 31% YoY, Enterprise SSD Up 139%
Claire Weston
May storage data show nearline HDD capacity up 31% and enterprise SSD capacity up 139% year-on-year — AI data-center procurement has shifted from unit count to capacity density, with demand spreading from compute chips into storage and passive components.
HDD and SSD are both surging — aren't they supposed to be rivals?
Nearline HDD — large-capacity drives built for data-center cold storage — saw capacity output rise 31% YoY, with average drive capacity reaching 23.0 TB, up 14% YoY.
Enterprise / data-center SSD capacity output jumped 139% YoY and 10% QoQ, with average drive capacity hitting 7.1 TB, up 64% YoY.
This means → cloud buyers are no longer adding drives; they want more data per drive. The core procurement variable has shifted from unit count to capacity density.
Why are both growing at once?
In plain terms = AI training and inference split data into two tiers: hot data (frequently accessed) needs SSD for high-throughput, low-latency reads, while cold data (logs, old model versions, compliance archives) relies on HDD for low-cost bulk storage.
As the total storage pool in data centers expands, both tiers benefit — this is not a zero-sum replacement.
The report notes demand for 61.44 TB, 122.88 TB, and even 245.76 TB SSDs is at "unprecedented" levels, with QLC designs — storing four bits per cell for higher capacity at slightly lower endurance — increasingly common at the top end.
This reflects a shift in how NAND flash is valued: pricing is moving from consumer-electronics cycles to an AI-infrastructure capacity cycle. If mix improvement and price increases arrive together, NAND makers' margin upside could prove more durable than a simple price-hike cycle.
Capacitor shipments hit a new high against the trend — what does that signal?
April data from major Japanese component makers showed total shipments weakening MoM, yet capacitor revenue reached ¥167.6 billion, up 20% YoY. Dollar-denominated revenue also grew, ruling out a pure currency effect.
This means → the squeeze from AI servers on high-spec passive components is already visible. As GPU and ASIC power draws rise, commodity MLCCs — multilayer ceramic capacitors — cannot substitute for high-capacitance, high-reliability part numbers. The upward product-mix shift is reshaping MLCC pricing.
Ceramic electrode material output is still down 15% YoY, but recovered 8% MoM — the direction has turned from destocking toward testing whether supply tightness will widen.
Why watch connector data separately?
Connector shipments improved both YoY and MoM, and dollar-denominated figures also recovered, but the slope remains gentle — no sustained uptrend yet.
In plain terms = capacitors are already "hot"; connectors are still "warm." Whether connectors shift from mild recovery to sustained growth is a key thermometer for AI-server diffusion.
Which companies did J.P. Morgan flag?
TDK is named the top pick among HDD component makers. The logic: HDD makers are shifting internal capacity toward HAMR — heat-assisted magnetic recording, which uses a laser to heat the disk surface for higher storage density — and external supply-chain share should consolidate toward TDK.
TDK also benefits from HDD heads/suspensions, AI-server aluminum capacitors, and mid-size batteries. The diversified mix lowers its near-term leverage versus pure MLCC plays, but gives it stronger resilience against single-product swings.
Murata Manufacturing and Taiyo Yuden map more directly to the supply-demand tightness in high-value MLCC part numbers for AI servers.
What data points should we track next?
Four verification checkpoints: ① Can capacitor revenue keep growing in dollar terms? ② Can connectors shift from mild recovery to a sustained uptrend? ③ Can nearline HDD average capacity hold above 23 TB? ④ Can enterprise/data-center SSD's share of total capacity keep rising from 20.3%?
This means → if all four weaken at the same time, the premise behind the AI hardware second-wave diffusion trade — the investment thesis that demand spreads from GPU chips into downstream storage and components — faces a material challenge.
Content is for reference only, not financial advice.