Storage Chips "In the Game Within the Game"

In October 2025, OpenAI signed an agreement with Samsung and SK Hynix under the name "Stargate," securing a supply of up to 900,000 DRAM wafers per month—approximately 40% of global DRAM production, all acquired by a single buyer, as part of its four-year, $500 billion infrastructure plan.

Everyone watching the storage market anticipates the upcoming trends.

Contract DRAM prices surged 171% following the announcement of the deal. The impact on the retail market was even more severe. A set of Team Delta RGB 64GB DDR5-6400 memory sticks, priced at $190 in August, is now listed online for $700. In less than three months, the increase reached 268%. Such a steep rise in the commodity market during peacetime is almost unprecedented.

Even the DDR4, which was expected to be in oversupply, has seen prices more than double. Storage devices were also affected: the Western Digital WD Blue SN5000 1TB SSD rose from $64 to $111, while the 2TB version increased from $115 to $154.

Demand is only part of the reason.

OpenAI's "Stargate" expansion requires an astonishing amount of high-bandwidth memory. Data centers are competing with consumer electronics manufacturers for the same basic resources, with the former bidding much higher—significantly higher. Factories are shifting advanced DRAM capacity towards AI demands, as the bids from data centers are unmatched by consumer device manufacturers.

If supply remained stable, the situation might be manageable. Unfortunately, supply has not stabilized.

Samsung, SK Hynix, and Micron have all shifted their production lines to HBM, which is specialized high-bandwidth memory for AI accelerator chips. The profits there are far higher than for consumer-grade DDR5. Micron's entire HBM capacity for 2026 has already been booked. The startup volume of wafers for traditional memory modules continues to decline, even though PC manufacturers are still purchasing.

Thus, we see: while demand is surging, the production structure is shifting. This indicates that the memory price surge, which appears to be a "conspiracy," is entering a short-term trend that lacks resolution.

What distinguishes this memory cycle from previous ones is that the upgrade path for AI infrastructure is entirely different from consumer hardware. Once these data centers are built, they will continue to consume memory as model sizes grow. The demand curve will only go up.

In the face of this storm, the responses of tech giants vary dramatically, creating two distinct camps.

Sony stocked up on sufficient memory for the PlayStation 5 before prices soared. According to industry insiders, its inventory is enough to maintain stable console prices for the coming months, and it may even weather the entire shortage period unscathed. Apple also locked in supplies early, and its profit margins are sufficient to absorb the cost increases that could crush low-margin businesses. Lenovo has gone even further. According to Bloomberg, its memory inventory is about 50% higher than normal levels, enough to last until 2026.

Microsoft's situation appears passive. Industry analysts believe that the Xbox console may have to raise prices again—after the company already increased prices earlier this year Valve's timing has only made matters worse. Its highly anticipated living room gaming PC, the "Steam Machine," had core component costs that were two to three times everyone's budget at launch. The company was unwilling to commit to a final price and directly blamed the issue on memory shortages. Any value proposition they had planned was completely devoured by memory costs. The modular laptop company Framework has removed all independent memory modules from its store.

This camp divided by "foresight" is leaving a profound impact on the competitive landscape— even if prices stabilize in the future, the effects will persist for a long time. Sony and Lenovo have maintained their market positions; Microsoft and Valve have not been spared. Just a few months of supply chain management have compressed what should have been a multi-year evolution of market share changes.

A natural question arises: Why aren't Samsung, SK Hynix, and Micron ramping up production to seize profits as prices have tripled?

Ask anyone who experienced the surplus in 2018, and they will give you a precise answer.

From late 2016 to early 2017: memory supply was tight, and prices soared. The three major manufacturers responded in the most direct way—breaking ground on new fabs in Korea and China and increasing capacity on existing production lines. On paper, chasing high prices and expanding capacity seemed completely reasonable.

However, as 2018 arrived, the algorithm failed. Demand stabilized, but all new capacities came online simultaneously. Chips piled up in warehouses. Samsung's semiconductor division—once a profit engine—delivered quarterly results that shocked investors. SK Hynix faced a similarly bleak situation. The shadow of oversupply lingered until 2019 before gradually dissipating.

That experience has profoundly shaped every capacity decision these companies make today. It takes several years to build a memory fab from construction to production. The wafer allocation choices made in November 2025 will determine market supply in 2027 or even 2028. And will AI infrastructure spending still be so hot by then? No one can answer. What if AGI development hits a bottleneck? What if hyperscale users realize their excessive investments and start canceling orders?

Manufacturers would rather maintain the current shortage than repeat the mistakes of oversupply. They call it "discipline." Meanwhile, consumers, watching the price of a memory module rise from $89 to $310, might use different words.

Industry forecasts indicate that the supply tightness of DRAM and NAND will continue until 2026; there is only partial optimism for a return to normalcy in 2027. Orders for large nearline hard drives are backlogged for two years. Distributors are beginning to bundle memory with motherboards just to control inventory distribution.

Zooming out, an uncomfortable picture gradually becomes clear:

PC builders and gamers are funding the development of AI infrastructure. Not directly, nor voluntarily, but that is how the mechanism operates. OpenAI, Microsoft, Google, Amazon—all hyperscale users—they have driven up memory prices to stockpile what is needed for training clusters. Manufacturers prioritize meeting the needs of these customers. Availability at the retail end continues to shrink. That teenager saving up for a gaming PC now has to pay an extra $400 for memory that only cost him $130 three months ago There has never been a public policy debate weighing whether "accelerating AI development" is worth the cost of "rising consumer computing costs." The market simply directs resources to the highest bidder. And when a single customer can lock in 40% of global supply, this process becomes rapid and fierce.

Three companies produce almost all of the world's DRAM. Such concentration brings systemic vulnerabilities that have not received sufficient regulatory attention. To be frank: it is time to examine this situation.

Source: JinDuan