Nvidia Unveils 64GB Desktop AI System to Help Firms Run Models In-House

Nvidia has expanded its small-footprint artificial intelligence hardware line, revealing a 64GB configuration of its DGX Spark workstation designed to let companies run capable AI models directly in their own offices.

The system is scheduled to reach the market on 23 October through hardware partners including Acer, Asus, Dell, Gigabyte, HP, and MSI, with pricing starting at $4,999. The machine is engineered to support open-weight artificial intelligence models of up to 100 billion parameters straight out of the box, without requiring access to external cloud providers.

Bringing AI off the cloud and behind your doors

For many Australian small and medium businesses, running advanced generative systems has historically meant piping sensitive commercial files, intellectual property, and client records to third-party offshore cloud providers. That arrangement regularly introduces data sovereignty challenges, recurrent monthly subscription fees, and worries about privacy.

The compact, desktop-sized system operates on Nvidia's Grace Blackwell architecture and runs DGX OS alongside the company's full enterprise software stack. By running workloads locally, businesses can execute complex automated workflows and keep proprietary customer databases and commercial records firmly on-premises.

Nvidia has also bundled a multi-node linking feature called Sync Cluster Assistant. If a business needs more power down the track, two of the 64GB machines can be tethered directly to pool their memory into a 128GB environment, providing a stepped upgrade path as computing demands grow.

What this means for Australian businesses

Local commercial adoption of AI is rapidly shifting from basic conversational chatbots to autonomous agents that handle core business processes. For regional firms, accounting practices, healthcare clinics, and engineering outfits handling sensitive records, keeping that automation strictly in-house solves several major headaches:

• Stronger data governance: Client records and commercial workflows never leave the physical office network, simplifying compliance under Australian privacy rules.
• Predictable computing costs: Capital expenditure replaces runaway monthly API queries and cloud hosting fees.
• Lower latency: Internal teams can deploy specialised agent tools without relying on offshore server latency or internet connectivity interruptions.

While hardware investment still requires upfront capital and technical know-how to set up frameworks like Ollama or vLLM, hardware developments like this show that running enterprise-grade AI privately is moving within reach of everyday businesses.