Samsung Electronics is planning to more than double production of its sixth-generation HBM4 and seventh-generation HBM4E memory chips next year, industry sources say. To support this, the company will raise outsourced cleaning of glass carriers—supports used to stabilize wafers during thinning—from 20,000 sheets monthly this year to 50,000 sheets monthly in 2025, continuing a trend that saw volumes double from 10,000 sheets last year.
PieceMakers, a DRAM designer majority-owned by Nanya Technology, began trading on Taiwan's Emerging Stock Board at a NT$740 reference price and closed its first session up 23.6% at NT$915, valuing the company near $1.4 billion. Rather than competing in HBM, the firm is pursuing custom memory designs that bond DRAM directly to processors via hybrid bonding, positioning the technology between SRAM-only chips like Nvidia's Groq LPU and traditional HBM for AI inference workloads.
SK Hynix is in talks with Intel about producing RAM chips in the United States, according to a Reuters report citing unnamed sources. Options under discussion include leasing space at Intel's planned Ohio plant or forming a joint venture, possibly with cloud-service providers involved, though SK Hynix says no final decisions have been made.
SK hynix is reportedly discussing options with Intel for U.S.-based chip production, including leasing space at Intel's under-construction Ohio One facility or forming a joint venture involving Intel and AI hyperscalers seeking HBM supply. The talks come as SK hynix expands its Indiana HBM packaging plant following its $26.5 billion Nasdaq listing, though neither company has confirmed the discussions.
SK Hynix is reportedly discussing options with Intel to produce memory chips in the US for the first time, possibly by leasing space at Intel's delayed Ohio fab or forming a joint venture involving cloud companies. SK Hynix has confirmed it is exploring new production sites but says no decisions have been made, and Intel declined to confirm details beyond ongoing investment in the Ohio project.
At IFA 2026, AMD introduced the Threadripper Halo Station, a liquid-cooled workstation combining a 96-core Threadripper PRO 9995WX processor with up to four Instinct MI350P accelerators. A fully configured system would offer 576GB of HBM3E memory, enough to run trillion-parameter AI models locally without cloud access. AMD showed a two-card version at the event but has not disclosed pricing, availability, or full specifications.
SK Hynix CEO Kwak Noh-jung said he sees no signs of oversupply in the memory chip market and expects the current shortage to persist until 2030, driven largely by AI-related demand for advanced memory like HBM. He made the comments as the company broke ground on a new chip packaging facility in Indiana, part of its broader US expansion tied to growing customer needs.
SK hynix held a groundbreaking ceremony this week for its first U.S.-based HBM packaging and testing facility, located in West Lafayette, Indiana. The $4 billion site will package HBM stacks using DRAM wafers made in South Korea, with volume production of next-generation HBM memory expected in the second half of 2029, roughly a year later than originally planned.
At Hot Chips 2026, SK hynix's Jaesik Lee said hybrid bonding won't be ready in time for HBM4E and will instead debut with HBM5, as current HBM4 packages are already constrained by a 775-micron thickness ceiling matching standard logic wafers. To keep scaling within that limit, 16-Hi HBM4 stacks now use dies thinned to about 50 microns with half the interlayer gap of 12-Hi designs, while SK hynix continues using its existing MR-MUF bonding method through upcoming Nvidia Rubin-generation products.
Samsung presented plans at Hot Chips 2026 to rethink High Bandwidth Memory design by fabricating the base die on a 4nm logic process instead of a DRAM node, starting with HBM4 and HBM4E. This logic-node shift reduces power consumption and frees up unused base die area, which Samsung is now exploring ways to use, including a first phase that would integrate the memory controller directly onto the base die.
At Hot Chips 2026, researchers Anurag Agarwal and Radhakrishna Giduthuri presented a tutorial on High Bandwidth Flash (HBF), a package-integrated flash memory technology built like HBM but functioning more like an SSD. Since no HBF hardware exists yet, the talk relied on simulations and projections, examining how software—especially inference frameworks like vLLM—would need to be redesigned to exploit HBF's large capacity despite its coarse-grained, storage-like access pattern.
Nvidia has informed top customers that server systems built around its Grace Blackwell and Vera Rubin AI chips will cost more than 15% extra when they ship early next year, according to a Bloomberg report. Contract manufacturers serving Microsoft, Google, and Oracle have already passed along the increases, which vary by chip generation and memory configuration.
GoKawiil