Samsung's HBM4 Price Hike Is the Real Cost of the AI Boom

The Double Price

Samsung is negotiating to more than double the supply price of HBM4 for next year. Not 20 percent. Not 50. More than 100 percent. That is not a rounding error in the memory business—that is a structural shift in who holds the whip hand.

The negotiations are in their final stage this month. Samsung has already locked in most of its key customers’ volumes for next year: NVIDIA, Broadcom, and the growing roster of AI accelerator designers betting their products on high-bandwidth memory. The price increase is being driven by a simple equation that general-purpose DRAM buyers are now feeling the other side of: there is not enough silicon to go around, and the silicon that exists earns more when it is made into HBM rather than the commodity chips that power your phone.

Why Your Phone Is on the Hook

HBM4 is not used in smartphones. It lives in AI servers, stacked vertically beside GPUs and custom accelerators. But HBM and general-purpose DRAM share the same wafers. When a foundry runs a 12-inch wafer through its lines, each die that becomes HBM is a die that cannot become LPDDR—the mobile memory your phone uses. That is the channel through which AI’s appetite for server memory becomes your next device’s price tag.

Goldman Sachs has flagged exactly this mechanism: general-purpose DRAM spot prices are running strong, and if a wafer earns more making standard DRAM than it would making HBM, the economics flip. Samsung is now in the opposite position—HBM is so much more profitable that every wafer is pulled toward it. The result is a constriction of general DRAM supply that keeps its prices elevated, which in turn gives Samsung further justification for HBM price increases. It is a circular structure, and it is working in Samsung’s favor.

TrendForce projects general-purpose DRAM contract prices rising 10 to 15 percent in Q4, and 15 to 20 percent when HBM is included. KB Securities and TrendForce both estimate HBM prices climbing above 100 percent next year, with TrendForce land­ing at 121 percent as an average across the segment.

The Wafer Tax

HBM4 carries a steeper manufacturing burden than its predecessors. The base die now requires a 4-nanometer logic process, which consumes more wafer area and more complex fabrication steps. Each generation of HBM also uses larger dies, meaning fewer dies per wafer and higher unit costs. Samsung began mass-producing HBM4 in February 2026, and volume shipments ramp next year—but the yield curve and customer qualification pace remain variables that could trim the profit boost.

One semiconductor industry source noted that while Samsung is likely to hold the pricing主动权 (upper hand) in negotiations due to tight supply across both HBM and general DRAM, the actual earnings impact will depend on how quickly HBM4 yields stabilize and how aggressively customers push back on the price jump.

Micron Paved the Way

Micron’s move should not be overlooked. Multiple reports indicate Micron has already locked in 2027 HBM volumes at significantly higher prices than this year. That sets a market benchmark. Samsung and SK Hynix now have a clear reference point: the price floor has moved up, and both Korean memory makers have room to align their own contracts to the new reality rather than compete downward.

This is the third-time-around lesson of the memory cycle. After years of brutal price wars and margin destruction, the major DRAM producers have learned that coordination through market scarcity is more profitable than coordination through explicit agreement. No one needs to pick up the phone. They just need to watch the same wafer shortage and raise prices together.

Samsung’s Share of the Pie

HBM is about to become a much larger slice of Samsung’s DRAM revenue. Hanmi Securities projects HBM will account for roughly 15 percent of Samsung’s DRAM sales next year, driven by HBM4 volume expansion. If the 100-percent price increase materializes, that share could swell to 20 or even 30 percent. For a company that has wrestled with Intel’s former memory business exit and its own yield struggles at the leading edge, that kind of concentration in a single product category is both an opportunity and a risk.

SK Hynix in the Crosshairs

SK Hynix remains the established leader in HBM, particularly with NVIDIA’s flagship GPUs. But Samsung’s pricing aggression signals a company that is no longer willing to cede the margin landscape. If Samsung can deliver HBM4 at competitive yields and qualification speeds, the price leadership it claims in these negotiations could translate into real market-share gains—not just revenue gains, but the kind of share shift that reorders the memory hierarchy.

The competitive dynamic matters because it determines whether the price increases are short-lived (driven by a temporary shortage that eases as capacity comes online) or structural (driven by a permanent reallocation of wafer capacity toward HBM that leaves general DRAM chronically tight). The circular supply dynamic described above suggests the latter is more likely—at least through 2027.

What Happens Next

NVIDIA and its peers have already committed volumes. They cannot easily switch memory suppliers mid-design cycle without requalifying their hardware, which takes months. That gives Samsung structural leverage in these negotiations that few other suppliers in any industry enjoy.

For the broader market, the takeaways are straightforward: AI server costs will rise, general DRAM prices will stay elevated as wafer capacity tilts toward HBM, and the companies best positioned to absorb or pass through these costs will be the ones with the strongest pricing power downstream. Smartphone makers are on the weaker side of that equation.

Samsung started 2026 trying to prove it could compete with SK Hynix in HBM. By negotiating a price increase of this magnitude, it is doing something more consequential: it is proving it can dictate terms.