Running a large AI model at home normally means one of two things: renting time on someone else's cloud server, or building a bulky tower around a power-hungry graphics card. ASUS is pushing a third route, and it has just made it easier to choose. The ASUS Ascent GX10, a desktop AI computer small enough to sit next to your monitor, is now sold in two memory sizes, 64GB and 128GB.
Contents
Quick facts
- The ASUS Ascent GX10 desktop AI computer now comes with 64GB or 128GB of unified memory
- Both run the NVIDIA GB10 Grace Blackwell Superchip, rated at 1 petaFLOP of AI performance
- 64GB handles AI models up to 70 billion parameters, 128GB up to 200 billion
- No Malaysian price confirmed yet; ASUS sells it directly
The ASUS Ascent GX10 now comes in two sizes
Both versions run on the same NVIDIA GB10 Grace Blackwell Superchip, pairing a 20-core NVIDIA Grace processor with a Blackwell graphics engine, and both deliver what ASUS rates at 1 petaFLOP of AI performance. The difference is memory, and memory is what decides how big a model you can load.
The 64GB model handles AI models of up to 70 billion parameters, which covers most of the popular agentic models in the 30 to 35 billion range, and it runs them entirely on the device at a lower price than the larger version. The 128GB model stretches to models of up to 200 billion parameters, aimed at people working with bigger or fine-tuned systems. The specifications come from the ASUS press release.
Why a desk-sized AI box matters
The appeal is keeping the work local. A model that runs on the machine in front of you does not send your data to a cloud provider, and it does not bill you for every query the way a subscription does. For comparison, Google's cloud Gemini plans in Malaysia run from RM23.99 to RM979.90 a month, so a one-time piece of hardware changes the sums for anyone running models all day.
The GX10 is also built to grow. Two 64GB units can be linked over their ConnectX-7 network ports to act as one 128GB machine, reaching up to 2 petaFLOPS, which ASUS says is up to 1.7 times a single 128GB unit, with no server rack involved. Two 128GB units combine to 256GB, and four can scale to 512GB and 4 petaFLOPS for the largest models.
It is a small part of a much bigger shift. AI hardware now accounts for more than 70 percent of Malaysia's export growth, and tools that put that kind of computing on a single desk are part of the same story.
Size, storage and heat
The Ascent GX10 measures 150 x 150 x 51mm, roughly the footprint of a small hardback book. Storage starts at a 1TB NVMe SSD for smaller datasets, with 2TB and 4TB options for heavier workloads. ASUS says the cooling uses seven-level fan control, five heat pipes and twin fans to hold performance steady through long jobs.
Price and Malaysian availability
ASUS sells the Ascent GX10 directly, with a contact-sales route for bulk orders, and both memory versions are available now. A Malaysian price and local availability were not confirmed at the time of writing, so treat it as reachable through ASUS rather than on local shelves for the moment. We will update this piece if ASUS Malaysia confirms pricing.
For developers and researchers who have been put off by cloud bills or by building a full workstation, a quiet box that holds a 200-billion-parameter model on a desk is a genuinely different way to work.
