In the scorching deserts of California, a gunshot rang out. The object being blasted to pieces was none other than an Apple iPod music player. The shooter was Scott Nelson, a sales executive at memory chip maker Kioxia. To this day, fragments of that broken iPod still sit on his desk.

Though this anecdote dates back over a decade, it resurfaced in a Wall Street Journal (WSJ) report on the 29th, laying bare the long-standing friction between Apple and its hardware suppliers. Embittered by what he viewed as heavy-handed and excessive management by the tech giant, Nelson has stubbornly stuck to Android smartphones ever since.

To consumers, Apple is synonymous with sleek innovation and design. But to the memory makers and component suppliers behind its devices, Apple has historically been the ultimate "super-superior" (super-gap), exercising immense pressure. However, the dawn of the Artificial Intelligence (AI) era is radically flipping this dynamic, as a new wave of buyers emerges—willing to buy more components, at much higher prices.

"Inspecting Supplier Books and Pushing for Rock-Bottom Prices"

According to the WSJ report, Apple’s negotiation tactics were ruthless. Suppliers were forced to accept an unyielding principle: they could never sell to anyone else cheaper than they sold to Apple. Apple would go so far as to inspect suppliers’ internal accounting books to verify compliance. Even after a contract was signed, if market prices dipped, Apple would refuse to take delivery, leaving suppliers vulnerable to heavy inventory losses unless they agreed to slash prices yet again.

Apple’s tactics extended to psychological pressure. Meeting schedules were deliberately set so that rival sales teams would cross paths in the waiting rooms, instilling a fear that if a supplier didn't yield on price, the contract would instantly go to a competitor sitting just outside the door. Late-night negotiations stretching past midnight, with executives unable to grab dinner, were commonplace.

In late 2015, when SanDisk refused an extra supply order because another customer was willing to pay a higher price, Apple’s procurement chief Tony Blevins reportedly threatened a SanDisk sales representative, vowing to "make the CEO crawl in on his knees." Coincidentally, SanDisk's CEO at the time was Sanjay Mehrotra—who now leads memory giant Micron. Had a domestic conglomerate treated a supplier this way, it would have faced immediate public outcry over corporate bullying.

Memory CEOs Find Their Voice in the AI Era

The friction between Apple and Mehrotra continues today, but the power balance has shifted dramatically. Back in June, then-Apple CEO Tim Cook openly admitted that soaring memory prices were becoming unsustainable and that product price hikes were inevitable—a rare public lament from a company accustomed to calling all the shots.

Shortly after, U.S. media reported that Apple was lobbying the Washington government for permission to source Chinese memory chips. According to the WSJ, Apple is projected to shell out a staggering $25 billion for NAND flash memory this year, compared to just $6 billion last year.

Mehrotra, however, fired back with little hesitation. He pointed out that past aggressive price-cutting by certain customers crashed memory prices to a third of their value by 2023, driving suppliers into the red and starving manufacturing facilities of necessary investment. The subtext of his counter-attack was clear: Apple must shoulder much of the blame for the current AI memory bottleneck.

The primary reason memory makers can now stand their ground is the rise of a new class of heavyweights: hyperscalers building massive AI data centers. According to TechInsights estimates, while Apple used to buy twice as much NAND flash as Microsoft, Amazon, Meta, and Google combined, those four tech giants now purchase double the volume that Apple does.

Dylan Patel, CEO of SemiAnalysis, remarked in a recent panel discussion on the 23rd: "While Apple squeezed memory companies during past downturns, memory companies are now squeezing Apple."

Apple's recent overtures to China have reportedly hit a wall as well. News broke recently that Apple sought long-term contracts with domestic Chinese memory maker CXMT, only to be turned down because the offered price was too low and CXMT's capacity is already stretched thin meeting internal Chinese demand.

Will LTAs Shape a New Era for Memory Supply?

However, the memory sector's unprecedented boom is not entirely without risks. Soaring memory prices are triggering contractions in adjacent markets. A prime example occurred when Intel’s stock plummeted over 20% following its earnings release on the evening of the 29th.

Intel warned that rising memory costs are choking PC demand, a market squeeze expected to stretch into the first half of next year. Investors panicked over the prospect that surging AI infrastructure investments are undercutting Intel's bread-and-butter core businesses. While Anthropic’s securities filing on the 26th temporarily injected optimism into semiconductor stocks by showcasing rapid AI growth and profitability, Intel’s earnings report served as a stark reminder of the hidden costs behind spiking memory prices.

As hyperscalers fiercely compete to secure scarce memory—the primary bottleneck in AI scaling—the question remains: Will the relationship between memory vendors and buyers perpetually swing between extremes?

Neither hyperscalers nor memory makers appear eager to push extremes this time around. Suppliers recognize that aggressively squeezing customers could ultimately break the market—a sentiment echoed by SK Group Chairman Chey Tae-won, who noted in July that memory prices had reached "abnormal levels."

To keep pace with the breakneck speed of AI advancement, hyperscalers desperately need stable memory supplies. Building a multi-billion-dollar data center only to leave it idle is far costlier than locking up multi-year supply volumes at sustainable prices.

This mutual need explains the surging popularity of Long-Term Agreements (LTAs), which allow suppliers to secure long-term demand even if they accept slightly lower margins, while buyers eliminate critical bottlenecks to bolster their AI competitiveness.

During its Q3 earnings call, SK hynix announced it had signed LTAs with 14 major global clients, securing long-term demand covering over half of its projected production capacity for 2027–2028. Samsung Electronics has similarly signed contracts with 10 clients and is actively negotiating with 10 more.

Skeptics in the stock market remain wary of LTAs, noting that long-term contracts forged during past booms were frequently broken when market tides turned. Much of the recent stock volatility reflects these ghosts of semiconductor history.

Yet, one thing is certain: both memory suppliers and their customers have begun searching for sustainable, long-term ways to do business. Whether memory’s elevated status in the AI era is here to stay, and whether LTAs can finally resolve the decades-old feud between tech giants and component makers, only time will tell.

[ⓒ Global Economic Times. 무단전재-재배포 금지]