The setup
Puya Semiconductor (Shanghai) is a fabless designer of non-volatile memory: NOR Flash and EEPROM at the core, built on a dual SONOS + ETOX process platform, with a "storage-plus" adjacency in MCUs and analog drivers (VCM, and now Iris) that ride the same customer sockets. It is a niche-memory house, not a fab — which matters, because in a shortage the fabless niche player captures price without carrying the capex.
Two things happened to that business at once. The first is industry-wide and well understood: as HBM and high-layer 3D NAND absorbed capital and capacity, the majors pulled 55/65nm capacity away from small-density NOR, and the resulting squeeze pushed NOR Flash and SLC NAND contract prices up more than 100% across the first half of 2026 — with AI server BIOS, optical modules, DDR5 SPD, and on-device AI wearables all adding new demand into a shrinking supply base. HBM4 alone is credited with raising NOR content per AI server roughly 50%.
The second is company-specific. In , Zhuhai Noah Changnan bought 100% of SkyHigh Memory Limited — the Hong Kong-domiciled high-performance 2D NAND franchise, SLC NAND and eMMC and MCP, automotive- and industrial-qualified — from an SK hynix subsidiary for US$69.392 million. Three months later Puya took its Noah Changnan stake from 20% to 51% for RMB 144 million and began consolidating. In it moved to buy the remaining 49%.
Read the H1 print without that second fact and you conclude Puya had a very good cycle. Read it with that fact and you conclude something narrower and more durable.
The numbers
Revenue of RMB 3.959 billion, up 336.67% against RMB 907 million a year ago. Attributable net profit of RMB 827 million, up 1,929.65%. Ex-items net profit of RMB 825 million, up 2,994.08% — and the fact that the ex-items number is essentially the headline number, with non-recurring gains of only RMB 2.19 million, is the first quality tell. Nothing here is a disposal gain or a subsidy. Basic EPS RMB 5.58 against RMB 0.35. Weighted ROE 28.73%, against 1.82%.
The sequential shape is steeper than the year-on-year shape, which is what you want in a cycle you think is still inflecting: Q1 attributable net profit RMB 251 million, Q2 RMB 576 million — up 129% quarter on quarter. Gross margin reached 51.15%, the best in the company's history, and for a fabless niche-flash designer that number is not a normal-world number.
Cash confirmed it. Operating cash flow swung to +RMB 945 million from −RMB 42 million, on cash collected from sales of RMB 3.447 billion (+322%) — collection growing faster than revenue. That is the line that separates a real upcycle from a shipped-and-hoped one, and it is the single most reassuring figure in the release.
| Segment | H1 2026 revenue | YoY | What it is |
|---|---|---|---|
| "Storage" — NOR + EEPROM | RMB 1.284B | +90.79% | The legacy franchise. >3.7B NOR units shipped (record); >870M EEPROM units. |
| "Storage X" — SHM 2D NAND | RMB 2.346B | n/a (new) | SLC NAND, eMMC, MCP. 59.3% of group revenue. Consolidated from Q4 2025. |
| "Storage +" — MCU, drivers | RMB 0.329B | +41.20% | Cortex-M0+/M4 MCUs, VCM and Iris drivers. >3.3B MCUs cumulative. |
Strip SHM out and the legacy business still did RMB 1.613 billion of revenue, up 77.9%, with roughly RMB 319 million of net profit against RMB 41 million — a genuinely excellent cyclical result that would have been the story in any other year. But the acquired business contributed roughly 59% of revenue and about 61% of profit, and no amount of segment enthusiasm makes that a second-order detail.
The rest of this dive is for paid subscribers.
The headline numbers are above. The 7 sections below carry the mechanism, the peer read, the valuation work, and the dated tests that decide it.
- What's actually driving it
- The thing that should bother you
- How peers read this
- Valuation reality check
- What we'd watch from here
- The trade
- Bottom line
Already a subscriber? Sign in.