Category
TrendForce has raised its 2027 capex forecast for nine major cloud providers. Comparing it with its August estimate and Microsoft's accounting change shows how unit growth, higher prices and prepayments each affect spending totals.
Samsung is reportedly in early development of zNAND-T and zNAND-I for AI. We look at how they differ from the officially unveiled zNAND-O, when moving KV cache pays off, and what SK hynix's AIN strategy says about purpose-built NAND.
Elon Musk praised Micron's approach to future memory pricing. Micron's earnings materials show price caps and floors alongside purchase obligations, and the $32 billion commitment, remaining contract value and cash received measure different things.
Pat Gelsinger called HBM terrible yet the best available. We look at the interface specs and thermal analysis behind that view, separate Nanya's 2028 supply outlook from its fab plans, and examine AI memory demand beyond HBM.
Kioxia has added direct liquid-cooling models to its CM9 enterprise SSD line. They keep PCIe 5.0 and also work in air-cooled systems. We look at how capacity and DWPD differ, and what the SNIA specification says about cooling surfaces and pressure.
Elon Musk says China is quickly learning to make LPDDR5 and LPDDR6, and can scale output massively. We compare CXMT's official speed figures with SoC-linked operation and look at HBM packaging and cooling challenges.
Apple has reportedly asked the court to overturn a jury verdict of more than $5.7 billion in the Taptic Engine patent case. How did the court treat the vibration-testing evidence versus the damages evidence? Court documents shed light on the test conditions, the patents' contribution to value, and next steps.
Getty Images plans to restructure through Chapter 11 and go private, cancelling existing shares while continuing normal operations. SEC filings show the funding commitments and how debt payments, the Shutterstock merger and litigation drained liquidity.
Apple has described Minions, a system that pools idle Mac GPUs to shorten the wait before AI inference jobs start. Drawing on the paper and Apple's MLX materials, we look at how it distributes jobs rather than models, how it reruns interrupted work, how priorities and data isolation work, and where its cost claims fall short.
A user reports that Intel's ICPS software disrupts IPv6 connections and stops apps like OneNote from syncing. We review the comparison tests, the RFC specification, and device-specific distribution limits to see what has improved and what remains unresolved.
