The reallocation

Every wafer allocated to an HBM stack for an Nvidia GPU is a wafer that doesn't become the LPDDR5X in your next phone. This is the sentence that explains everything that follows.

The three companies that make virtually all of the world's memory — Samsung, SK Hynix, and Micron — have spent the last eighteen months converting their cleanroom capacity toward high-bandwidth memory for AI data centers.^[6] The economics are straightforward: HBM margins are five to ten times higher than standard DRAM. If you're a memory fab and Nvidia is offering to buy your entire output at premium pricing, you don't say no. You say "how much more can we make?"

The answer, as it turns out, is "not enough for everyone else."

HBM consumes roughly three times the wafer capacity of DDR5 per gigabyte. So the reallocation isn't one-for-one. Every gigabyte of HBM that Samsung produces costs the consumer DRAM market three gigabytes of capacity. That's the physics. The economics just make it worse: no rational manufacturer is going to shift capacity back toward the lower-margin product when the higher-margin product is sold out through 2027.

This is not a cyclical shortage. This is a structural reallocation of the world's silicon capacity from consumer products to enterprise AI infrastructure.^[11] And the consumer market is starting to feel it.

The numbers

Source: TrendForce DRAM contract price reports

Conventional DRAM contract prices rose 55 to 60 percent quarter-on-quarter in Q1 2026. Server DRAM prices are up 60 to 70 percent, with Samsung and SK Hynix quoting those numbers directly to Google and Microsoft.^[12]

For smartphones, the bill-of-materials impact varies by tier. DRAM price surges have increased BoM costs by roughly 25 percent for low-end devices, 15 percent for mid-range, and 10 percent for flagships.^[9] The asymmetry matters: memory is a larger share of the component cost in a $150 phone than in a $1,200 phone. The cheapest devices get hit hardest.

IDC projects global smartphone shipments will fall 12.9 percent in 2026 to 1.12 billion units — the steepest annual decline in the market's history.^[2] Counterpoint Research, which is calling this "RAMageddon" (we did not make that up), forecasts a 12.4 percent decline.^[3] Q1 alone is expected to be down 6.8 percent as vendors struggle to secure supply at prices they can pass through.

PC prices are expected to rise 15 to 20 percent in Q1.^[1] Smartphone ASPs will climb 3 to 8 percent depending on scenario.^[4]

The sub-$200 smartphone segment — the volume backbone of markets like India, Southeast Asia, and Africa — is forecast to contract by more than 20 percent.

Who survives

The memory shortage is a filter, and it selects for purchasing power.

Apple will be less affected, because Apple pre-negotiates supply contracts, buys in enormous volume, and charges enough per device to absorb component cost increases without destroying margins.^[5] Samsung the device maker is in an even more peculiar position: it is simultaneously the company causing the shortage (by diverting wafers to HBM) and the company best positioned to survive it (by guaranteeing its own supply).

Everyone else faces a hierarchy of pain. Companies with strong supplier relationships and scale — Xiaomi, Oppo, Vivo — will secure memory but at elevated prices. Smaller and regional OEMs will find it "increasingly difficult to compete for supply," which is analyst-speak for "some of them will exit the market."^[4]

The LPDDR4 supply — the memory that powers budget and entry-level devices — is shrinking faster than expected. Samsung and SK Hynix have limited incentive to keep producing it when the same fab capacity can make LPDDR5X at better margins (or HBM at dramatically better margins). So the bottom of the device market gets squeezed from both directions: the memory that goes into cheap phones is disappearing, and the memory that replaces it costs more.

The second-order effects

Here's where the memory shortage stops being a semiconductor story and starts being a consumer story.

Phones. Fewer units at higher prices means replacement cycles extend. If your $200 phone now costs $250 and the $150 option doesn't exist anymore, you keep your current phone longer. This is deflationary for unit volumes and inflationary for ASPs — the worst combination if you're a carrier subsidizing handsets or an emerging-market operator trying to grow data subscribers.

PCs. A 15 to 20 percent price increase lands squarely in the corporate refresh cycle that was supposed to drive volume in 2026. Enterprise buyers will defer purchases. Consumer buyers — already stretched by everything else in the economy — will do the same. The "AI PC" narrative that was supposed to drive upgrades now has to compete with the reality that the AI infrastructure buildout is making the PCs more expensive.

Consoles. Gaming consoles use GDDR6, which competes with server DRAM for some of the same inputs. The next console cycle — whenever it arrives — will face a component environment that is meaningfully worse than anything Sony or Microsoft planned for.

Cloud. Server DRAM at plus-60 percent feeds directly into cloud computing costs. AWS, Azure, and GCP will absorb some of it and pass the rest through. Startups that were building on the assumption of declining compute costs per unit are in for a surprise.

The paradox

The AI boom is creating a consumer recession in the hardware that consumers actually buy. The mechanism is simple: finite wafer capacity, allocated to the highest bidder, with the consumer market as the residual claimant on whatever's left.

Samsung warned about this explicitly.^[7] SK Hynix's advanced packaging lines are at capacity through 2026. Micron has sold out its HBM production for the year. The supply isn't coming back because the demand that's pulling it away — hyperscaler AI infrastructure — is not slowing down.

So we have a situation where the most celebrated technology trend in a generation (AI) is directly causing the worst smartphone downturn in history, the steepest PC price increases in years, and the likely exit of multiple device manufacturers from the market. The same chip shortage that makes SK Hynix's margins beautiful makes your next phone $50 more expensive and your next laptop $200 more expensive.

We suppose that's what a "supercycle" looks like from the other side.