Samsung Electronics is seeking more than three times the price of its current flagship memory for next-generation HBM4, and has already begun negotiations with major customers, according to an October 2 report from South Korean financial daily Maeil Business (MK).During annual supply talks for next year's HBM4 contracts, Samsung proposed prices in the mid-to-high $4 per gigabit range — a steep jump from the roughly $1.50 per gigabit that HBM3E, its current flagship, commands.
The timing is deliberate. HBM4 is set to replace HBM3E as the flagship product next year, marking a full generational transition in the high-bandwidth memory stacked into AI accelerators.Unlike conventional DRAM, HBM is supplied almost entirely under annual long-term contracts, with manufacturers and customers negotiating the following year's volumes and prices well in advance.Most of next year's major HBM volumes are understood to be largely negotiated already, with price talks expected to reach their final stages this month.

A premium built on industry-leading speed

Samsung's confidence comes down to performance. According to the report, the company's HBM4 has reliably achieved data-transfer speeds of 11.7 gigabits per second and can reach up to 13 Gbps — the highest level in the industry.The strategy is straightforward: convert the ability to stably supply higher-specification products than competitors into a direct price premium.
The company is also investing in the generation after that. Separately, Samsung is reportedly working with equipment manufacturers and backend partners on new cooling technology at the package and server-system levels for HBM4E, a sign that the performance race is only intensifying.

Chip stocks rally on the pricing signal

Markets wasted no time pricing in the windfall. On Friday, European chip stocks jumped after the report: ASML, Europe's most valuable company, rose 2.2%, ASM International gained 4.7%, and analog chipmakers Infineon and STMicroelectronics climbed 6% and 3.3% respectively, according to Dow Jones Newswires market talk.Stocks tied to the broader AI buildout, including Siemens Energy and Prysmian, also traded higher.
Independent forecasts support the trajectory.Taiwan's DigiTimes has projected HBM4 prices climbing from about $2 per gigabit in the second half of this year to more than $4 to $5 next year, driven by exploding demand ahead of Nvidia's next-generation AI platform, Vera Rubin, and by large customers preemptively locking in supply through three-to-five-year long-term agreements.The analysis suggests roughly half of total production capacity could be prioritized for those contract customers starting next year.
The stakes are high for Samsung itself. Analysts estimate profits from its non-memory divisions could shrink from about 6.74 trillion won ($5.1 billion) in 2025 to roughly 3.92 trillion won ($2.95 billion) in 2026 — a decline that makes a fat-margin HBM4 cycle look like the perfect hedge.

What the hike means for the AI supply chain

For AI accelerator makers, the bill is real: HBM is bundled into every high-end AI chip, and a threefold price increase directly raises the cost of building GPU and AI systems.Whether customers accept the full $4-plus asking price will depend on qualification results, supply availability, and how competing HBM4 products from SK Hynix and Micron are priced, according to industry watchers.
Samsung is also said to be targeting a quadrupling of its HBM customer base, betting that the bandwidth leap — and the supply commitments signed well ahead of mass production — will carry the premium through.With negotiations nearing their final stage this month, the terms agreed now will effectively set the cost structure of AI hardware in 2027.