Korea's Chip Exports Jumped 270%. Prices Didn't.
Korea publishes customs data every ten days, which makes it the highest-frequency read on global memory demand available anywhere. The latest release, covering September 1-10, shows semiconductor exports up 270.1% year over year. That number is real, and it is not a calendar artifact — but it is also not what most headlines took it to mean.
Memory contract prices did not rise 270% in September. The export figure is a cumulative result, and the underlying price momentum is decelerating while the headline accelerates.
KEY TAKEAWAYS
1. Korea exported $34.97bn in the first 10 days of September, up 82.6%. Semiconductors were $16.48bn of that, up 270.1%, or 47.1% of all exports versus 23.3% a year earlier.
2. Working days were 8.5 in both periods, so the daily-average growth rate is also 82.6%. This is not a working-day distortion — a rare case where the headline survives the usual first check.
3. Server DRAM contract price growth has slowed from 90-95% QoQ in Q1 2026 to a forecast 13-18% in Q3 2026. Value growth is peaking while price momentum fades.
What the data actually says
The Korea Customs Service released the September 1-10 figures on September 11. Total exports were $34.97bn, up 82.6% year over year, the largest ever for that window and ahead of the previous record of roughly $30bn set in early July this year. Imports rose 20.7% to $24.6bn, leaving a $10.4bn surplus.
Semiconductors came in at $16.48bn. Of the $15.8bn increase in total exports, $12.0bn came from chips alone. Strip semiconductors out and the rest of Korea's export basket still grew 25.8% — concentration is extreme, but the rest of the economy is not shrinking.
The first thing to check on any 10-day print is the working-day count, because weekend and holiday placement can swing these numbers badly. Here it was 8.5 days in both years. Daily-average exports were $4.11bn, growing at the same 82.6%. The surge is genuine.
Why 270% is a twelve-month number
A year-over-year rate compares against early September 2025, when Korean semiconductor exports were roughly $4.46bn. Every step of the memory price increase over the past twelve months is compressed into that single comparison.
Korea's August trade release cited fixed contract prices directly: DDR5 16Gb went from $35.00 in April to $46.50 in August, up 32.9% in four months, while 128Gb NAND went from $24.20 to $30.50, up 26.0%. Those are substantial increases. They are also an order of magnitude away from 270%.
Price momentum is already fading
TrendForce, in a July 9, 2026 release, put server DRAM contract price growth at 90-95% QoQ in Q1 2026, 58-63% in Q2, and a forecast 13-18% in Q3. Prices are still rising. The rate of increase has collapsed by roughly a factor of six in two quarters.
TrendForce attributes part of this to long-term agreements signed by several US cloud providers. LTAs cap what suppliers can charge those customers, so increases concentrate on buyers without contracts. The same DRAM part realizes different prices depending on who is buying it — which is why a supplier's blended ASP can lag the spot and contract headlines that get quoted.
Three ways this data gets misread
Dollars per kilogram as a chip price. Export value divided by weight is widely quoted as a price proxy. HSK 8542 covers wafers, packages and modules together, and the mix of high-value parts like HBM shifts constantly. Product mix alone moves dollars-per-kg with chip prices unchanged. It is not a like-for-like price series.
Comparing a 10-day print to a full month. Month-over-month comparisons only work against the same window — September 1-10 against August 1-10. Mismatched-period growth rates show up in published coverage more often than you would expect.
Treating preliminary as final. These are preliminary customs-clearance figures. Confirmed monthly data arrives around the 15th of the following month and gets revised.
What I actually watch
| Indicator | Timing | What it answers |
|---|---|---|
| MSIT ICT trade report | Around the 14th monthly | Memory split out from logic |
| Customs confirmed monthly data | Around the 15th monthly | Size of revision from preliminary |
| TrendForce quarterly contract prices | Start of quarter | Whether Q4 holds a double-digit rate |
| Memory maker quarterly results | 3-4 weeks after quarter end | Bit shipment growth reported separately from ASP |
Value chain read-through
| Segment | What the value surge implies | Caveat |
|---|---|---|
| Memory suppliers | Price and volume improving together | Heavier LTA exposure caps realized price gains |
| Materials and components | Tracks volume, not price | Almost none of the price effect passes through |
| Equipment | Lags capacity decisions | Higher prices alone do not trigger capex |
Risks to this view
• Base effects turn against the series from Q4 2025 comparisons onward. The same absolute dollar figure produces a far lower growth rate.
• Preliminary 10-day data is revised. Treat the 270.1% as provisional until the confirmed monthly print.
• The 13-18% Q3 figure is a forecast, not a settled outcome. If realized contract prices land below it, export value growth follows.
• With 47.1% of exports in one product category, there is no cushion if that cycle turns.
Export value is price times volume. Judging a cycle from the value line alone means missing the turn whenever those two variables move in opposite directions — which is exactly what is starting to happen now. Next, once confirmed September data is published, I will separate memory from logic and isolate the volume effect on its own.
Sources: Korea Customs Service preliminary trade data released September 11, 2026; Korea Ministry of Trade, Industry and Energy August 2026 trade report; TrendForce press release July 9, 2026. Everything here is from public sources.
Comments
Post a Comment