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.
SK hynix pushes hybrid bonding to HBM5 after hitting a 775-micron packaging limit, Everpure tops the Gartner MQ again, Kioxia and Sandisk's $31B NAND bet, and this week's Uplevel Highlight on CXL memory pooling.
Read Issue 03 →DDR5-8800 memory (up from DDR5-5600) and 5x larger L3 cache directly target the memory-bandwidth bottleneck in LLM inference and real-time analytics, showing hyperscalers are now tuning general-purpose compute specifically for AI memory pressure.
Tape remains the low-power archival tier under AI storage hierarchies; the dip and rebound (tied to trade uncertainty and a new 40TB cartridge) matters for anyone modeling the cost curve below all-flash and disk.
Vdura's Flash Volatility Index shows 30TB enterprise SSDs at $22,600 vs $3,460 a year ago; a mixed SSD/HDD 25PB system now costs $12.86M versus $51.6M all-flash, making storage architecture the second-biggest cost decision after GPUs.
The closest thing to an apples-to-apples benchmark for the storage workloads AI teams actually run today, not just checkpointing; Everpure's FlashBlade hit 877.52 GiB/s on a 1.25-trillion-parameter training workload.
Another object storage vendor positioning itself as the data layer under a sovereign AI buildout, following the same pattern as DDN's Australia deal.
Small benchmarks hide size-dependent performance cliffs; this gives teams evaluating vector databases a much more realistic production-scale stress test.
Checked twice daily. The full story gets the proper treatment in the next weekly issue.
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.
AI DevOps engineers, infra leads, and the executives they report to — the people who actually buy and operate storage and memory infrastructure. Sponsor slots are open. Get in touch to inquire.
One email a week on AI storage and memory infrastructure. No fluff, no spam.
One email a week. Unsubscribe link in every issue.