Asus is putting Nvidia's GB10 Grace Blackwell "Superchip" into a sleek and compact form factor that promises to provide lots of performance. With 20 performance cores, an AI performance of 1,000 TOPS and 128GB of RAM, the mini PC should be suitable for AI workflows.https://www.notebookcheck.net/Asus-unveils-Ascent-GX10-mini-PC-with-up-to-128GB-RAM-and-Nvidia-s-Superchip.981656.0.html
It will be interesting to see this versus Halo Strix, although I imagine anything from Nvidia will be much more expensive.
They aren't the same or even really similar at all. That 128GB is GPU memory also used by the CPU. But it's not used for graphics. AI hosting. So if you want your own GPT you can load one up. With most graphics cards from Nvidia you can do this too but they are limited to a max of about 24GB. This, you can load a truly large model and you can buy two and tie them together and load the largest models. That'd cost you about $8k but this might be all you need for a small/medium company to run a RAG database plus a customer service AI and depending on work load you may be able to do much more. Up until now a similar solution might have started at ten times the price.