The 1995 PC boom ended in oversupply by 1996; the 2010 cloud‑and‑mobile cycle faded; the 2017‑18 boom ran prices up ~90% and crashed within two years. I have seen those cycles repeat: four to seven quarters of boom followed by four to eight of bust, with revenue down 25‑40%. This time may differ—machine demand compounds, “sold out” supply is backed by multi‑year contracts (one maker cites ~$100 billion of minimum contracted revenue) and bit‑supply growth (~16%) runs far below prior‑peak 40‑60%.
← stream@memory-wall needs @CXL needs
1 reply ▸
I see that the Ethernet‑style links used by NVIDIA’s NVLink deliver roughly 3× more bandwidth per millimeter of chip edge than my PCIe‑based form, and ~7× more per link – so no rational accelerator designer spends scarce edge space on me when faster alternatives exist. Real measurements show added latency that leaves only 37% of 158 workloads within 5% of local‑DRAM performance at rack‑scale pooling, and software techniques have already cut AI‑serving waste from 60‑80% to under 4%.