News
Innodisk’s margin surge signals AI memory tightness beyond HBM
AI-related demand has lifted Innodisk’s revenue and margins to record levels, providing evidence that the memory upcycle is spreading into conventional DRAM and enterprise flash.
Industrial memory supplier Innodisk said AI-related products accounted for more than 40% of its first-half 2026 revenue for the first time, as demand from GPU servers, networking equipment, industrial control and enterprise AI continued to exceed supply.
According to DIGITIMES, management expects memory supply constraints to persist from the second half of 2026 through the end of 2027. Innodisk plans to begin construction of its third Yilan production facility by the end of 2026, with operations scheduled to start in 2028.
Second-quarter revenue reached NT$22.44 billion, increasing 70.3% sequentially and 641.1% year over year. Gross margin climbed to 65%, while net income attributable to the parent company reached NT$10.37 billion and earnings per share rose to NT$108.93.
DRAM generated 61% of quarterly revenue, while flash products accounted for 34%. Although Innodisk reports only a small AI systems business, much of its AI exposure comes from the DRAM modules and storage products installed in servers, switches, and edge-computing systems.
Why it matters for memory
The results suggest that AI-related memory demand is expanding beyond HBM. Accelerated-computing systems also require conventional DDR5 and server DRAM, while switches, storage platforms and edge systems create additional demand for industrial and enterprise flash.
This is directionally positive for Samsung Electronics, SK hynix, and Micron. HBM is absorbing leading-edge DRAM capacity at the same time that AI infrastructure is increasing demand for conventional server memory. Innodisk’s flash growth also indicates that NAND is beginning to play a more meaningful role in the AI infrastructure cycle.
MemoryCapex view:
Innodisk’s 65% gross margin is a strong signal of pricing power, but it should not yet be treated as a sustainable long-term margin. Part of the increase may reflect a favorable product mix and the sale of inventory acquired before memory prices moved sharply. The central message is that AI-driven tightness is spreading through a broader share of the memory market. For the investment thesis, the next confirmation would be continued DRAM and flash revenue, accompanied by stable margins and stronger cash conversion—not revenue growth alone.
Original reporting: DIGITIMES
Financial data: Innodisk investor presentation
Independent analysis by Majid Asli. MemoryCapex provides research, not investment advice.