Home » Blog
How data actually moves through an AI server, why HBM became the industry’s tightest bottleneck, and where projected technologies like High Bandwidth Flash and CXL fit into the picture. The common mental model of AI hardware goes something like this: a GPU loads the model into its memory, then starts computing. Simple. It is also…
Read MoreTL;DR: Korea’s June 29 plan puts over $550B toward new memory fabs — a slice of a much larger $3.1T multi-decade Samsung/SK spending plan. Most new capacity targets high-margin server DRAM and HBM, not everyday PC/laptop memory. Micron is running its own parallel expansion; China’s CXMT is growing fast but not yet server/HBM-ready. Real relief…
Read MoreIf you run AI models in production, the bill has a shape you’ve probably noticed: a surprising amount of it traces back to memory. Every token a model generates has to be served out of fast memory sitting right next to the GPU, and that memory — High Bandwidth Memory, or HBM — is scarce,…
Read MoreMicron just turned in the memory market’s headline quarter. In its fiscal third quarter, reported after the close on June 24, revenue reached a record ~$41.5 billion on a record gross margin near 85%, with DRAM alone accounting for about $31.3 billion, roughly three-quarters of the total. Both figures are company records, and both sit…
Read MoreThe hardest part of selling decommissioned servers, memory, or storage in 2026 isn’t finding a buyer. It’s knowing what the gear is actually worth this week — because the number has been moving fast enough that any price more than a few weeks old is probably wrong. That accuracy is the part we’ve built our…
Read MoreOn June 15, AMD announced it had acquired MEXT, a small software company with an unusual pitch: make cheap NAND flash look like DRAM to the operating system, so a server can run with far less of the expensive stuff. Terms weren’t disclosed, and AMD picked up MEXT’s engineering team along with the technology. For…
Read More