Nvidia has started shipping its $4,999 DGX Spark desktop AI computer with half the RAM and storage originally advertised, a quiet downgrade amid an industry-wide memory crunch. The compact personal AI supercomputer is based on the DGX Spark platform delivering 1 PFLOPS of FP4 performance, aimed at developers who want to build and test AI models locally before scaling to data centres. At the Ai Everything event in Abu Dhabi on October 6-7, GIGABYTE showcased its AI TOP ATOM, a DGX Spark-based system with 1TB to 4TB storage options and four-node clustering support. Also on display was the W775-V10 desk-side AI workstation, powered by the Nvidia GB300 Grace Blackwell Ultra chip, delivering up to 20 PFLOPS of FP4 performance and 748GB of coherent memory for up to 400 concurrent requests. The downgrade highlights how tight memory supply has become as AI demand surges across the industry. Analysts say the memory squeeze could persist into 2027 as AI data centres keep absorbing supply.