Apple Asked China's CXMT for a Discount and Paid More Instead

The Phrase That Outlived a Farewell Call

Apple's earnings call on July 30, 2026 was the last one Tim Cook joined as chief executive. Under the succession plan announced in April, he hands the job to John Ternus and becomes executive chairman. The line that survived that call was not about a product. It was about a component: "We're in what I would characterize as a hundred-year flood on memory pricing, with exponential increases in memory prices."

A second sentence from the same call drew less attention. Cook noted that the DRAM market has essentially three suppliers, and added that "if there were more suppliers, that would be good, and it would help us on the supply side and perhaps the pricing side." That was the world's largest component buyer publicly conceding that its own bargaining power was no longer enough.

The next ten days showed how serious the concession was.

Four numbers, four different points in the same story. The first says Apple's cost structure has changed. The second says it isn't finished. The third says how narrow the exit Apple is looking for really is. The fourth says what the company is doing until it finds one: stockpiling.

The End of a Twenty-Year Assumption

Apple's leverage over its supply chain rests on a single assumption. An order of hundreds of millions of units a year fills a supplier's fab, and in exchange for that fill rate the supplier discounts. No component maker has a better customer than Apple: the volume is enormous, the payment is reliable, the demand is forecastable.

The assumption holds only as long as no better buyer shows up.

In 2026 one did. AI data centers do not treat memory as a cost line; they treat it as a revenue gate. For a hyperscaler, an extra dollar spent on server memory converts into rentable compute, which makes willingness to pay close to unbounded. Apple puts memory into a consumer device whose price it cannot raise at will. When two buyers queue for the same wafer, the winner is not the one with the deeper pocket but the one with the greater willingness to pay.

Two Products, One Fab

The critical point is that HBM — high-bandwidth memory, the stacked DRAM that sits next to an AI accelerator — is not made in a separate factory. It comes off the same lines, from the same wafers. Producers decide each month which product gets the capacity, and that decision is not easily reversed: converting a line takes months.

Every Gigabyte That Goes to AI Takes Three Gigabytes From Everyone Else

The arithmetic here explains the severity of the price move on its own.

With demand unchanged, removing a third of available supply does not move price by ten percent; it moves it by sixty. TrendForce estimates conventional DRAM contract prices rose 58–63% in the second quarter of 2026, with a further 13–18% expected in the third. The rate of increase is slowing, but the level is not falling — the third-quarter rise stacks on top of the second-quarter jump.

There is also a persistence problem. For the producer, the same wafer earns far more in HBM. The price signal is not pulling capacity back toward consumer parts; it is holding capacity where it is.

Apple's Attempt at a Third Supplier

Cook's "more suppliers" remark was not theoretical. Apple had been talking to China's CXMT for months.

In early August, reports sourced to Korean and Chinese trade press described how those talks went. Apple asked for price cuts on LPDDR5X, the mobile DRAM used in iPhones. CXMT declined. According to those reports, the unit prices the Chinese producer quoted were above — or at best equal to — those of Samsung and SK Hynix. It was the first time a Chinese maker had quoted above its Korean rivals in this market.

There are two reasons, and both are mechanical.

The first reason is the cost floor. US export controls bar CXMT from buying EUV machines. According to DigiTimes, that forces CXMT to start roughly 30% more wafers to reach the same output. If cost per wafer is broadly comparable, cost per bit sits about 30% higher. This is not a strategic choice; it is what physics and accounting jointly produce. A producer whose cost is above a rival's loses money by pricing below that rival.

The second reason is utilization. CXMT has no idle capacity to sell. In June it signed a server-memory agreement with Tencent worth roughly $2.94 billion; in July, a five-year contract with ByteDance worth up to $7 billion. Huawei and Xiaomi are already long-term customers. A producer running full has no reason to cut price.

How the Market Is Split Today

To see how little room Apple has, look at how the market divides.

On Counterpoint Research's numbers, three producers hold nine-tenths of the market. CXMT's 7% looks small — but that share was close to nothing a year ago, and the company has stated a 30% target for 2030. The problem is that Apple needs supply this quarter, not in 2030.

Where the Bill Landed

Tripling memory's share of an iPhone's bill of materials feeds straight into gross margin. Apple's product margin was 49.3% in the March quarter; stripping out the effect of tariff refunds, it fell to 48.1% in June. CFO Kevan Parekh said "more than 100%" of that 120-basis-point decline came from memory pricing, partly offset by other lines. The 47–48% range guided for September includes roughly one point of tariff benefit — take that out and the picture is harsher still.

Meanwhile the company is stockpiling. Memory and storage inventory rose to $11.09 billion, nearly double the prior period. That delays the price increase by a few quarters. It does not stop it.

The chart is here not as proof of anything but to mark the ground the argument stands on: this is the stretch over which margin pressure became the subject.

April to August

That last entry matters. The supplier that failed as a bargaining chip is now being tested not for leverage but for plain supply. According to the report, the target is devices sold in China first. HP and Acer have already begun using CXMT memory in machines sold outside the US, which suggests the path is not unique to Apple. Reuters noted it could not independently verify the report, and that neither Apple nor CXMT commented.

The Other Side

The August 21 date in the letter means this argument produces an answer within the next two weeks.

What Is Left

What Apple lost is not a supplier. It is a method. For twenty years, volume meant discount. In memory it no longer does, because the buyer across the table draws its power not from volume but from willingness to pay.

At the other end of the same equation sit the producers. When pricing power moves from buyer to seller, margin moves in the same direction.

This piece is based on reporting available as of the morning of August 10, 2026 from The Wall Street Journal, Reuters, DigiTimes, TrendForce, Counterpoint Research, the letter published by US senators, and the transcript of Apple's July 30, 2026 earnings call. Details of the CXMT talks rest on unnamed sources; neither Apple nor CXMT has commented. Reuters stated it could not independently verify the Wall Street Journal report. Price-increase estimates vary between research firms.