ChangXin Technology's memory chips have been rising in price continuously. GigaDevice (603986) uses its chips to make modules—will this cost first be reflected in the single-quarter gross margin of the 2026 annual report, or will it not be seen until the 2027 first-quarter report?

ChangXin Technology's memory chips have been rising in price continuously. GigaDevice (603986) uses its chips to make modules—will this cost first be reflected in the single-quarter gross margin of the 2026 annual report, or will it not be seen until the 2027 first-quarter report?

Let me start with the conclusion: this cost will only be partially visible in the 2026 annual report; it will fully hit single-quarter gross margin in the 2027 Q1 report and be fully absorbed in Q2. For the Q4 single quarter in the 2026 annual report, I estimate it will be flat at normal levels, dropping two or three points at most; what jumps first is inventory and accounts payable.

First, correct the premise: GigaDevice does not buy DRAM dies, nor does it make modules

GigaDevice is Fabless; it designs DRAM itself and has CXMT foundry it.

The CXMT Group "opens part of its capacity to provide foundry services for the company's niche DRAM business"; the transaction category is "procurement and foundry related to own-brand DRAM products."

This is the original wording of the September 29, 2026 related-party transaction announcement. It sells chips; it does not make memory modules. The 2026 half-year report's main business composition has only four lines: storage chips 98.27 yi, microcontrollers 14.30 yi, sensors 1.00 yi, analog products 1.91 yi; there is no "modules."

Those that buy dies to make modules are Longsys, Biwin, and Demingli. Module makers' cost is the die purchase price, which enters raw materials and follows orders; GigaDevice's cost is the CXMT foundry fee, which enters outsourced processing and inventory and follows the production cycle.

Where it comes from: the foundry price has already jumped once

The 2026 routine related-party transaction quota with CXMT was raised from 57.11 yi to 78.54 yi; the original reason was "DRAM product foundry prices rising with market conditions."

Actual January-August was 35.96 yi; the remaining four months imply 42.58 yi, or 10.65 yi per month; January-April 2027 is expected to be 44.11 yi, or 11.03 yi. The January-August monthly average was 4.50 yi.

The monthly average goes from 4.50 yi to 10.65 yi and then to 11.03 yi, 2.4 times. CXMT itself had first-half revenue of 1503.10 yi and net profit of 776.05 yi; for full-year 2025 it was only 617.99 yi and 18.75 yi.

Through what: 62.92 yi of inventory, 182 days

Inventory at the end of September was 62.92 yi, up 105% from 30.66 yi at the beginning of the year; calculated as quarterly average inventory divided by single-quarter operating cost, multiplied by 90 days, turnover days are 182 days.

Most of the goods sold in Q4 were put into inventory in Q2 and Q3 at the old foundry price. The high-priced foundry contracted in Q4 first sits in inventory, then is spread into operating cost on a weighted-average basis. At the end of September, contract liabilities rose from 2.18 yi to 12.36 yi; customers are paying in advance, so Q4 shipment volume will not be bad, and what is consumed is still cheap material.

Therefore, the single-quarter gross margin in the 2026 annual report is still at a high level and cannot fall. The step is in the 2027 Q1 report—when old material is exhausted and new material accounts for more than half, operating cost will fully absorb this round of price increases, and Q2 will be fully absorbed.

One more distractor to rule out: this round of gross margin is not produced by reversing inventory write-downs. Asset impairment losses in the first three quarters were 1.12 yi, versus 0.24 yi in the same period last year; these are provisions, not reversals.

Which line item it lands in: operating cost

The foundry fee appears only in operating cost; it touches neither expenses nor impairment. There is only one observation point: when the ratio of single-quarter operating cost to revenue starts to rise.

But one point: the cost side is already locked by contract; January-April 2027 is reported at 11.03 yi per month; the selling price side is still moving. If DRAM contract prices continue to rise in Q1, gross margin may only stay flat. The timing of cost entering the statements is certain; whether gross margin falls depends on when selling prices stop.

Peers have already played it out, one to two quarters earlier

Longsys and Biwin peaked in Q2 and fell 13 to 17 points in Q3; Demingli peaked in Q1. GigaDevice was still moving up in Q3, one to two quarters later.

The difference is not the thickness of inventory, but when input prices are repriced: module makers buy dies, and purchase prices were already rising in the first half, with costs following orders; GigaDevice's foundry price had its quota adjusted only at the end of September and was repriced only in Q4. Input prices are one to two quarters later, and 182 days of inventory push it back another quarter, so the inflection point falls in 2027 Q1.

When would I be wrong: if the Q4 single-quarter gross margin in the 2026 annual report falls below 65%, or if Q4 inventory days are compressed to within 120 days, that would mean high-priced foundry is carried over in the same month and transmission is faster than I calculated, and this judgment should be overturned.

Sources: GigaDevice 2026 Third Quarter Report (2026-10-30); GigaDevice 2026 Half-Year Report (2026-08-19); GigaDevice Announcement on Increasing the Estimated 2026 Routine Related-Party Transaction Quota and Estimating the January-April 2027 Routine Related-Party Transaction Quota (2026-09-29, Announcement No. 2026-087); Longsys 2026 Third Quarter Report (2026-10-30, Announcement No. 2026-119); Biwin Storage 2026 Third Quarter Report (2026-10-29); Demingli 2026 Third Quarter Report (2026-10-26, Announcement No. 2026-093).