Wire · technology
8 GPUs to 2… Nota Lowers Enterprise AX Costs by Lightening Ultra-Massive AI
◆ Sectors
◆ Geography
◆ Source
◆ Verified
Fusion42 · 27 July 2026 · Fusion42 review
Nota has optimised Solar Open 2, a 250-billion-parameter language model, reducing GPU memory requirements from 8 H100s to 2 whilst maintaining tool-calling performance through quantization and pruning techniques. The model weight shrunk by 76.5% (500.6GB to 117.8GB), lowering the infrastructure cost threshold for enterprise AI deployment.
This Wire brief sits within Fusion42's coverage of AI Infrastructure and Generative AI, and 2 sources have reported it.
◆ ◆ The Wire takeaway
If you're selling inference infrastructure to enterprises running large language models, your unit economics just got four times worse: Solar Open 2 now runs on 2 GPUs instead of 8, and this technique scales to any 250B+ model. You need to compete on something other than raw compute—or become the inference optimizer yourself.
◆ Coverage
2 sources · 27 Jul 2026
◆ Related on Wire
◆ Topics