CXMT Filed an 84.84% Margin and a $1.78 Price Trough
An 84.84% gross margin is a remarkable thing to report. It is a stranger thing to report alongside a paragraph reminding readers that DRAM once traded at $1.78 per gigabyte.
CXMT did both, in the same H1 2026 filing. This post takes the risk factors section seriously and works out what actually breaks first if that price returns.
Part 2 of a series on CXMT's H1 2026 interim report. All figures in Chinese yuan unless marked otherwise.
KEY TAKEAWAYS
1. The company's own risk section cites a 2015–2025 DRAM range of $7.89 down to $1.78 per GB — a 77.4% fall — and states that current pricing may not be sustainable.
2. An 84.84% gross margin means cost of sales is 15.16% of revenue. Holding volume and unit cost constant, price would have to fall 84.8% to erase gross profit. A repeat of the historical trough still leaves roughly 32.9%.
3. The real exposure is one line down. Depreciation of 14.51B yuan is booked regardless of price, taking it from 9.7% of revenue to 42.7% in that scenario.
What the company wrote about itself
The risk section lists four items: uncertainty in AI demand, price cycle volatility, fixed-asset and depreciation burden, and geopolitics. A fifth, structural one covers the absence of a controlling shareholder.
The second is the subject here. CXMT recorded that DRAM moved between $7.89 and $1.78 per GB across 2015 to 2025, and stated that current elevated pricing may not be sustainable. This from a company that grew revenue 873.6% and posted 77.61B yuan of net profit in the half.
There is no need to read modesty into it. No memory maker files a prospectus without a price-cycle warning. What is worth reading is which numbers it chose. A 4.4x gap between peak and trough is close to an admission that results in this industry are set more by price than by execution.
Turn 84.84% around
Start with where the number comes from. Memory prices are not indexed to cost. Die per wafer and yield set the cost; the day's supply and demand set the price. The shortage that began in the second half of 2025 pulled those two apart, and 84.84% is the gap. The company's own explanation — high utilization and improved product mix — points the same way. It sold dear rather than built cheap.
An 84.84% gross margin means cost of sales is 15.16 for every 100 of revenue. Hold that cost fixed and move only price, and the margin curve falls out.
A 20% price decline leaves 81.1%. A 40% decline leaves 74.7%. Even a halving leaves 69.7%. The cost base is so low that ordinary declines get absorbed. Gross profit reaches zero at an 84.8% price decline, and a full repeat of the trough the company cited (−77.4%) still leaves about 32.9%.
This holds volume and unit cost constant, so it is a concept calculation rather than a forecast. The direction is clear enough: to break this company at the gross profit line, price has to go below its own historical trough.
One line below sits 14.51 billion
Judged on gross margin alone, this looks like a fortress. The exposure is underneath it.
Depreciation for the half was 14.51B yuan, up 27.85%. Property, plant and equipment carries at 181.56B yuan, 38.79% of total assets. Depreciation runs at its own schedule regardless of price, because it is the cost of having built the plant.
Right now revenue is large enough that depreciation is only 9.7% of it. If price returned to trough levels and revenue fell to roughly a quarter, the same 14.51B would be 42.7% of revenue. A 32.9% gross margin does not cover that.
This is usually where memory makers break in a downcycle. Fixed costs outlast the price move. And CXMT has not started its build yet — as covered in Part 1, the 66.6B yuan raised in July has still to flow into fabs, which will make both the asset base and the depreciation line larger than they are today.
Where earnings quality shows
Set the year-over-year multiples side by side: revenue 9.7x, operating cash flow 30.9x, accounts receivable 4.4x, depreciation 1.3x.
Cash flow outrunning revenue is a good sign — the profit is converting. Receivables went from 1.52B to 6.65B yuan. That is slower than revenue growth, but the absolute step-up is large enough that collection is the first thing to check when the cycle turns.
One more: other income of 2.18B yuan (+139%) came mainly from an expanded VAT input-credit deduction. State tax support contributed to the result, and that line moves with policy rather than with the market.
What I actually watch
| Order | Line | Why here |
|---|---|---|
| 1 | Gross margin | Reacts first. How many points come off 84.84% sizes the price move |
| 2 | Depreciation ÷ revenue | Once 9.7% goes double digit, the fixed-cost squeeze has actually begun |
| 3 | Receivables turnover | Revenue falling while receivables hold is a channel-stuffing signal |
| 4 | Inventory | The route by which falling prices become write-downs |
Risks to this view
▶ The calculation is optimistic. Holding unit cost constant is generous. In a real downcycle utilization falls, unit cost rises, and margin compresses faster than the curve shows.
▶ Mix. 84.84% is a blended half-year figure. DDR5 and the LPDDR family price on different paths, and a mix shift breaks the average.
▶ The other direction. Prices can also keep rising. A company documenting a past trough in its risk section is not forecasting one.
The short version
84.84% is partly what CXMT built and partly what the price handed it. Which share is which becomes visible only when price falls and you watch how fast the margin follows. The line to watch then is depreciation.
Next in this series: the geography of the same filing. Overseas revenue went from 36.9% to 63.8% in a year, while essentially all of the production sits inside one country.
Sources: ChangXin Technology Group H1 2026 interim report (unaudited, 688825), risk factors section and financial statements. Everything here is from public sources.
Disclaimer: This post is for informational and educational purposes only. It does not constitute investment advice or a recommendation to buy or sell any security. All investment decisions are your own responsibility.
Comments
Post a Comment