Winbond Electronics: The Complete Guide

The 2026 numbers are a memory upcycle at full strength, and they are worth reading in sequence. Full year 2025: consolidated revenue NT$89.406 billion, up 9.55%, gross margin 35%, EPS NT$0.88 1. First quarter 2026: NT$38.253 billion, up 91.3% year on year, gross margin 53.4%, EPS NT$2.25 10. Second quarter 2026: NT$59.843 billion, up 184.7%, gross margin 66.2%, EPS NT$5.40 11. Through August, revenue for the first eight months of 2026 was NT$152,178,225 thousand against NT$54,860,887 thousand a year earlier, up 177.39% 4 12.

A single quarter’s EPS of NT$5.40 against a full-year NT$0.88 the year before is the cleanest measure of how violent this cycle has been 11 1. Winbond notes the monthly figures are "internal data and has not been audited by CPA" 4.

The employee figure deserves the same caution as the revenue, and the report is explicit about why. It records that "As of the end of 2024, Winbond Group’s total number of employees reached 8,097", and states its own boundary as Winbond Group — Winbond, its subsidiaries, Nuvoton and Nuvoton Japan 15. The breakdown is 5,165 in Taiwan, 2,593 elsewhere in Asia, 200 in the Middle East, 135 in North America and 4 in Europe 15. It is a group number, not a Winbond memory number.

Two structural facts to hold together: Winbond is 52.06% owner of Nuvoton Technology, described in its own consolidated statements as engaged in "Research, design, development, manufacture and marketing of Logic IC, 6-inch wafer product, test, and OEM" 8; and Nuvoton exists because Winbond carved its logic business out in 2008, when a shareholders’ meeting approved the split and Winbond said "Nuvoton, a wholly owned subsidiary of Winbond, will be devoted to Logic IC business" 17.

Winbond’s milestone page carries a year filter running back to 1987, so entries for the founding, the listing, the fab openings and the Qimonda-era technology transfers exist 18. This check could retrieve only the most recent page of that timeline, so the earlier entries are not quoted here. The 2026 filter returns nothing — as of this check Winbond had not added a 2026 milestone, and the Infineon agreement is not on the timeline 18.

One older item is retrievable as a standalone release and is worth noting because it connects Winbond to the same lineage as Nanya: in June 2007 it announced a 75nm and 58nm technology transfer agreement with Qimonda AG, Infineon’s spun-out memory business 20. Nineteen years later Winbond is buying a different Infineon memory business outright 14.

The mix has moved sharply, and the movement is the story. In full-year 2025 the split was "Flash memory product line accounted for 35% of 2025 revenue…Logic IC product line accounted for 34%…Customized Memory Solution (CMS) product line accounted for 29%" 1. By the second quarter of 2026 it was CMS 53%, Flash 33%, Logic IC 13%, other 1% 11.

Logic did not shrink — Nuvoton posted net sales of NT$7,941 million in the quarter 22, and NT$7,941 million against consolidated NT$59,843 million is 13.3%, which is where the 13% comes from 11. What happened is that memory revenue roughly tripled around it. CMS revenue "increased by 78% QoQ and over 400% YoY, supported by strong AI demand"; Flash "increased by 65% QoQ and 176% YoY" 22.

Winbond’s roadmap chart labels its own territory "Winbond Focus", spanning "45nm 40nm" on flash and "24nm 19/16nm" on the memory side 22. That is a deliberate position two decades behind the leading edge, in nodes where the tooling is depreciated and the customers need parts that stay available for ten years.

Three further products are in development on 16nm besides the launched DDR4: CUBE, an 8Gb LPDDR4 and a 16Gb DDR4 24. And the earlier mid-2025 language shows how recently this was still uncertain — the shareholder handbook described "next-generation 16nm technologies currently under development as planned" 13. Anyone reading a claim that Winbond’s 16nm reaches mass production in 2026 should check it against the August 2026 deck, which says contribution comes from 2027 22.

Everything specific that circulates about the Kaohsiung expansion is press, not Winbond. The Taipei Times reported construction from January 2027, pilot production at the end of 2029, Module B starting at 10,000 twelve-inch wafers a month and scaling toward 50,000, Module A rising from 15,000 to about 24,000 by the end of 2026, and EUV in later phases — quoting Winbond’s president James Chen on the sidelines of an earnings conference: "I never thought we would need EUV equipment. Now we want to buy" a machine 26. None of those numbers appears in any Winbond document found here 22. Read them as a newspaper’s reporting of an executive briefing, which is what they are.

One correction worth making explicitly, because it circulates widely: this check found no first-party source for a TSMC–Winbond collaboration — nothing from TSMC and nothing from Winbond, and the word does not appear in Winbond’s 2026 investor decks 22. The verifiable wafer-stacking partnership in Winbond’s history is with UMC, which named Winbond in a 2023 wafer-on-wafer 3D IC project alongside Faraday, ASE and Cadence 27.

The second thing Winbond makes its NOR do is security. Its TrustME line puts "a hardware-based Root of Trust (RoT) to protect systems from code tampering and cyberattacks" into the same part that stores the boot code, so the thing that validates the firmware is the thing that holds it 21. The W77Q is "a secure drop-in replacement for standard NOR Flash" carrying NIST SP 800-193 firmware resiliency and post-quantum cryptography; the W77T carries ISO 21434, ISO 26262 and Common Criteria EAL2+; the W75F reaches EAL5+ for "more stringent security requirements" 21.

That is a defensible position in a commodity segment. A boot flash certified to automotive functional-safety and cybersecurity standards is not interchangeable with a cheaper part, and Winbond notes it is "the first memory manufacturer in Taiwan to obtain the ISO 26262 certification" 5.

The CUBE specifications are the most detailed technical numbers Winbond publishes, and they come from its own white paper: "CUBE’s I/O interface supports a data rate of 2Gbps with 1K I/O, providing total bandwidth ranging from 16GB/s to 256GB/s per die", and "CUBE can be designed from 1-8Gb/die based on the D20 process, or 16Gb/die on D16. Both non-TSV and TSV stacking are available" 28. D20 and D16 are its 20nm and 16nm DRAM processes.

The architecture also inverts the usual stack: "The innovative 3D architecture of CUBE strategically places the System on Chip (SoC) on the top die close to the heatsink, effectively mitigating heat dissipation concerns associated with AI computing" 28. In HBM the memory stack sits beside the processor; here the processor sits on top of the memory, where it can be cooled.

The margin series tells the same story from the other end. Gross margin was 35% for full-year 2025, 53.4% in the first quarter of 2026 and 66.2% in the second 1 10 11. Operating income in the second quarter was NT$28,979 million on NT$59,843 million of revenue 22.

Winbond’s stated response is to trade some of that peak for stability: "Increasing LTA penetration enhances long-term supply resilience and order visibility" 22 — long-term agreements, the same instrument Micron, Kioxia, SanDisk and Nanya each describe in their own words. A supplier signing multi-year contracts during a shortage is giving up upside for visibility.

CUBE is what Winbond offers in that gap. Its white paper describes technology "designed to revolutionize AI computing on edge platforms", delivering "the massive increase in memory-interface bandwidth needed for systems to service the rapidly escalating demands for AI applications on edge computing platforms" 28, with less than 1pJ per bit of energy on the 20nm process 28. The product page adds that CUBE addresses conventional memory’s limits "through novel approaches to increasing I/O count and raising data speed, support for Through-Silicon Via (TSV) technology as an option, and its 3D architecture that reduces thermal dissipation issues", and is "suitable for use with power-conscious high-bandwidth edge/endpoint AI devices" 6. Winbond expects it to matter commercially later rather than now: "CUBE and Si-Cap are emerging as the next wave of major growth drivers beyond 2027" 22.

Two absences are worth stating. Neither the CUBE product page nor the CUBE white paper mentions HBM at all 6 28, and the word does not appear in Winbond’s 2026 investor decks 22. And this check found no Winbond statement, anywhere, on whether it will or will not produce standard JEDEC HBM — so this page asserts neither 22 7.

Winbond names no customers. This check found no end customer identified in any Winbond source — not in the results releases, the investor decks, the shareholder handbook or the product pages 1 22. Claims that particular phone or car makers buy Winbond NOR circulate in the trade press sourced to unnamed people, and none is repeated here.

The one structural thing that can be said about the customer base is that it is long-lived. A supplier certified to ISO 26262 and ISO 21434, shipping automotive Grade 2+ parts and 32nm NAND alongside 24nm, is serving designs that were locked years ago and will ship for years more 5 21 11.

It gets sharper. Nuvoton is not merely non-memory, it was losing money while the memory business boomed: in the second quarter of 2026 Winbond’s own deck reports Nuvoton with net sales of NT$7,941 million, gross margin 38.0% and operating margin −4.2% 22. Winbond’s consolidated gross margin of 66.2% in the same quarter is therefore a blend of a spectacular memory business and a loss-making logic one 11 22.

It also shows up directly in Winbond’s own 2025 reporting, and the company states the reconciliation itself: "We achieved a consolidated net income of NT$ 3.18 billion after tax (of which NT$ 3.96 billion is attributed to the parent company)" 9, against the results release’s "net profit attributable to the parent company was NT$3.962 billion" 1. The roughly NT$0.78 billion gap between the two is the share of losses borne by the 47.94% of Nuvoton that Winbond does not own. The group earned less than its memory business did, and the difference has a name.

Winbond’s formal risk disclosures could not be read in this check. Its 2025 annual report lists a section "VI. Risk Analysis and Assessment" in its contents, but that section’s text was not retrievable 9. So this page carries no Winbond statement on customer concentration, supplier concentration or currency exposure, because none was obtained — the risks described above are derived from its published financial and operational figures.

The market-share position is worth being exact about, because Winbond’s leadership is real but measured by others. Its results releases assert it in words — "worldwide No.1 NOR Flash supplier" 1 — and the numbers come from a market tracker: a slide headed "2025 Flash Market Share" in its 2026 investor decks gives Winbond 23% of a US$3.4 billion NOR Flash market, ahead of GigaDevice and Macronix at 17% each, Infineon at 11%, Micron at 9% and Puya at 7%; and 15% of a US$1.5 billion SLC NAND market, behind Micron at 21% and Kioxia at 20%. The attribution line reads "Source: Omdia Mobile and Embedded Memory Market Tracker – 1Q26" 22.

The last figure Winbond generated under its own name is much older: 22.8% of NOR revenue in calendar 2019, attributed to WebFeet Research in a May 2020 release 32. Post-deal projections in circulation are third-party too — "over 30%" is Economic Daily News and Commercial Times reporting relayed by TrendForce, and "approximately 34%" is TrendForce’s own estimate 33. Neither is Winbond’s.

That makes Winbond the ninth maker documented in this section without India manufacturing. Micron remains the only one with a plant there, and that plant does assembly and test rather than wafer fabrication. Winbond’s position is nonetheless a step ahead of most: it lists two India-based distributors, in Pune and Delhi, rather than covering the country only through a Singapore-registered regional one 31.

What the deal documents do not say is as important as what they do. Winbond’s release is silent on fabs, wafer supply and manufacturing transfer entirely; it names no employee count; it itemises no intellectual property beyond the Spansion name and trademark; and it discloses neither the divested business’s revenue nor how the US$1.12 billion is financed 14. It mentions "customary adjustments prior to closing" and no earn-out 14. Infineon’s own release exists but could not be read in this check, so no wording is attributed to it here 34; counsel to Infineon described the business as "a portfolio of NOR Flash memory as well as F-RAM products serving customers across automotive, industrial and infrastructure markets" 35, and the Taipei Times reported the transaction as "marking the return of a historical brand to the memorychip industry" 36.

Beyond 2027 Winbond points at CUBE and silicon capacitors as "the next wave of major growth drivers" 22. The question those three plans pose together is whether a company at 100% utilisation, funding a US$1.12 billion acquisition, a Kaohsiung module and a node transition at once, is doing so at the top of a cycle whose peak it cannot see. Its own EPS history — NT$0.14 in 2024, NT$5.40 in one quarter of 2026 — is the reason to ask 13 11.