Key Bullet Points:
- TrendForce forecasts conventional DRAM contract prices rising 13–18% this quarter and NAND up 10–15% — down from 58–63% and 70–75% in the second quarter
- The stated reason is not weak AI demand. It is that PC and smartphone buyers "are reaching their affordability limit"
- SK Hynix answered that slowdown Friday by committing 54 trillion won ($38.1 billion) to two new fabs that will not produce a wafer until 2028 and 2029
- Markets cut September Fed hike odds to 42% from 58% in a single session after payrolls fell 23,000. A hike is still close to a coin flip
- The 2-year Treasury yield fell to 4.193%, its lowest since July 17. The 10-year dropped about 7 basis points to roughly 4.6%; the 30-year eased to 5.192%
- July CPI lands Wednesday. A Reuters poll puts headline at 3.4% year over year and core at 2.5%. RBC models core at +0.2% m/m, taking the annual rate to 2.4%
- Core PPI is forecast at +0.3% m/m, and RBC expects nominal retail sales to fall 0.3% — though up 0.3% excluding autos and gas - The S&P 500 closed Friday at a record 7,757.51, capping a 3.58% week. The Nasdaq gained 5.19%
The Steepest Deceleration Nobody Is Discussing
Three months ago, memory was the most violent pricing story in technology. Conventional DRAM contract prices rose 58% to 63% in a single quarter. NAND flash rose 70% to 75%. Those are not annual figures. That was one quarter.
TrendForce now forecasts DRAM contract prices rising 13% to 18% this quarter, and NAND 10% to 15%.
Prices are still climbing. But the rate of climb has fallen by roughly three quarters, and the reason TrendForce gives is worth reading closely: contract prices have reached record highs, and customers in consumer markets such as PCs and smartphones "are reaching their affordability limit." Price tolerance, in the firm's words, has hit its limit.
That is a demand ceiling, discovered by the market rather than announced by anyone. And it arrives in the same week the industry committed tens of billions of dollars to building more supply.
The Capacity Answer to a Demand Question
SK Hynix said Friday it will spend 54 trillion Korean won — about $38.1 billion — on two new plants: a DRAM fab called Y2 in Yongin and a NAND facility, M17, in Cheongju. Groundbreaking runs from February 2027 to July 2027. First cleanrooms arrive December 2028 and June 2029.
The company's reasoning was explicit. "In the AI era, technological competitiveness alone is not enough," it said, "and the ability to supply the required volume at the exact moment customers need it is the ultimate competitive advantage."
Read that against the TrendForce data and the tension becomes obvious. The binding constraint on memory pricing is no longer how many wafers exist. It is what buyers will pay. Capacity decided today lands in 2029, and the semiconductor industry's oldest and most reliable pattern is that shortage-era capital expenditure arrives precisely when the shortage ends.
Micron sits in the middle of this. It has been among the largest beneficiaries of the memory repricing, and its margin structure has been rebuilt on contract prices that TrendForce now says are decelerating hard. The current quarter is still an up quarter. The second derivative is what changed.
There is a version of this story the market has already lived through once this summer, when more than a trillion dollars came out of AI chip names and Nvidia ended up trading at a multiple it last saw in 2015. Enthusiasm and pricing power are not the same variable, and they do not peak on the same day.
Bad News Bought a Rally
The macro backdrop shifted just as sharply on Friday.
July payrolls fell 23,000 against forecasts of an 83,000 gain. May and June were revised lower by a combined 103,000 jobs. The unemployment rate fell to 4.1% from 4.2%, but only because participation slid to 61.4%, its lowest in more than five years.
The bond market repriced immediately. The 2-year Treasury yield, the maturity most sensitive to Fed expectations, fell more than five basis points to 4.193% — its lowest since July 17. The 10-year dropped roughly seven basis points to about 4.6%. The 30-year eased two basis points to 5.192%. Odds of a September rate hike collapsed to 42% from 58% the previous day, and expected increases by December narrowed to 28 basis points from 32.
Equities took it as relief. The S&P 500 closed at a record 7,757.51, up 3.58% for the week; the Nasdaq gained 5.19%.
But notice what 42% actually means. The market did not conclude the Fed is finished. It concluded a September hike is slightly less likely than not. That is a very thin cushion under a record high, and the disagreement running through the largest names in the index has been visible for weeks.
Wednesday Settles It
July CPI arrives Wednesday, and it will do more to shape the rest of August than Friday's jobs report did.
A Reuters poll of economists puts headline CPI at 3.4% year over year, with core at 2.5%. RBC's forecast is a shade friendlier: core up 0.2% month over month, pulling the annual rate to 2.4%, and headline up just 0.1%, helped by retreating gasoline prices. Cooler shelter costs and cheaper airfares should assist. Elevated wage growth remains the floor under core services.
Two other releases matter. Core PPI is expected to rise 0.3%, with pressure from transportation and warehousing, and the ISM services prices-paid index jumped unexpectedly in July — a signal that wholesale and retail margins may be absorbing less than assumed. Retail sales close the week, with RBC modeling a 0.3% nominal decline on cheaper gasoline and softer vehicle sales, but a 0.3% gain excluding autos and gas.
Inflation has run above the Fed's 2% target for several years, and three of twelve policymakers dissented in July in favor of hiking. A hot print does not merely delay relief. It revives the hike case that Friday's payroll number appeared to bury, and the bond market has already shown this year that it can move against equities for a long time before stocks acknowledge it.
What Lands Next
Wednesday's CPI, Thursday's PPI, Friday's retail sales — into an index at an all-time high that is priced for an inflation outcome it has not seen yet.
The memory data is the quieter signal. When the fastest-rising prices in technology decelerate from 63% to 15% because customers cannot pay more, that is disinflation arriving through the demand side rather than the supply side. It is good news for CPI and considerably more complicated news for the companies whose earnings were rebuilt on those prices — a distinction the market has been slow to draw all year, as it was when the AI trade quietly narrowed to a single name.
Found this helpful? Share it with others.