A field notebook on the platters, cells, and file systems holding up the AI build-out, for the people who build and buy that infrastructure, and the leaders they report to.
One email a week. Unsubscribe link in every issue.
Solidigm reportedly weighs a US IPO that could value it anywhere from $100 billion to $150 billion, still without a US fab, Samsung plans to nearly double its HBM4/HBM4E output next year, CXMT enters mass production on a fifth-generation DRAM node using quadruple patterning, VAST Data pitches confidential computing for AI storage, and confidential computing explained.
Read Issue 09 →A running log of smaller developments between issues, each with a source. Updated a few times a week.
Ahead of the SNIA Developer Conference (SDC'26) in Santa Clara, PEAK:AIO said its pNFS Lattice development team, including Los Alamos National Laboratory's Gary Grider, will host a Birds of a Feather session on the open-source parallel NFS scalable metadata service. Source ↗
At the AI Infra Summit in Santa Clara, Seagate showed a physical AI prototype using more than 60TB of storage to operate a robotic arm, alongside its 44TB Exos Mozaic HAMR hard drives. Source ↗
China's CXMT, currently the country's largest DRAM maker, is reportedly setting up an R&D production line for NAND flash memory at a new plant in Beijing, a move that would put it in competition with Samsung, SK hynix, and YMTC in NAND alongside its existing DRAM business. Source ↗
Huawei introduced the OceanStor M900 Context Memory Storage at Huawei Connect 2026 in Shanghai, saying a single cluster pools 64 petabytes of capacity and 40 terabytes per second of bandwidth to hold shared KV caches for inference SuperPoDs, with SSDs connected to NPUs over a one-hop UnifiedBus link. Source ↗
Micron said it has demonstrated a 512GB DDR5 RDIMM for servers, enabling up to 12TB of DDR5 in a single 24-slot dual-socket system at speeds up to 9,200 MT/s and over 60% less power than four 128GB modules, with volume production targeted for the second half of 2027. Source ↗
ADATA's TRUSTA brand showed its AI Scaler Extended Memory solution at the AI Infra Summit 2026, an open-source toolkit that pools GPU memory, DRAM, and SSD capacity so larger AI models can run on systems with limited GPU memory. Source ↗
SanDisk told investors at the Goldman Sachs Communacopia conference that data centers now make up more than half of total NAND demand, and said it has signed eight long-term supply agreements with data center customers over the past nine months covering roughly two-thirds of its fiscal 2028 bit supply. Source ↗
Kioxia CEO Hiroo Ota said he has told sales teams to stop pushing data center customers for further NAND price increases, after the company's average NAND price rose 70% in the June quarter, and ruled out deeper ties with SK hynix beyond their existing joint ventures. Source ↗
Kioxia and SanDisk said they plan to invest more than ¥5 trillion ($33 billion) to expand production at their jointly-owned Japanese NAND fabs, while SK hynix separately earmarked 54 trillion won ($40 billion) for chipmaking expansion in Korea. Source ↗
KB Securities said combined memory inventories at Samsung and SK hynix fell below 10 days of supply in the third quarter, with analyst Kim Dong-won warning that sellable DRAM and NAND volumes could run out next year. Source ↗
Samsung introduced V10 BV-NAND at FMS 2026 — the industry's first 400+ layer NAND built with wafer bonding, about 58% denser than its V9 generation. Source ↗
Samsung says a next-gen interface built on its zHBM architecture could deliver roughly 8x the performance of HBM5, extending the roadmap covered in Issue 04's Deep Cut. Source ↗
Most weeks are a fast read. Occasionally, a topic gets the long treatment instead.
This week's numbers, curated news with a "why it matters" take, one rotating deep-dive story, a quick storage/memory term explained, and what's coming up.
Occasional full-market issues covering every enterprise storage vendor, hyperscaler storage, the HDD capacity race, memory and flash, field reliability data, and an optimization playbook.
A visual walkthrough of the data flow from a stored checkpoint to a generated token, and why the memory footprint keeps growing after the model has already loaded.
Read the guide →One email a week on AI storage and memory infrastructure. No fluff, no spam.
One email a week. Unsubscribe link in every issue.