Acer Unveils Compact SFF RTX Spark Desktop for Local Agentic AI

On September 2, 2026, Acer introduced the Acer SFF RTX Spark, a compact desktop designed to run agentic AI workloads locally. [1] The system, presented at the IFA 2026 conference in Berlin, is targeted at AI developers, creators, and gamers. [1] According to Acer, the small-form-factor design provides on-device data privacy while maintaining compute capacity. [1] The hardware was officially showcased during the next@Acer global press conference. [1]

The desktop is powered by NVIDIA's RTX Spark, a single superchip that links a 6,144-core Blackwell RTX GPU with a 20-core Grace CPU via the NVLink-C2C interconnect. [1] MediaTek collaborated with NVIDIA to design the custom Arm-based CPU. [1] NVIDIA stated the platform delivers up to 1 petaflop of AI compute and includes up to 128 GB of unified memory. [1] This hardware allows the system to run 120-billion-parameter large language models featuring up to 1 million tokens of context. [1]

The shift toward running 120-billion-parameter models locally marks a significant advancement in on-device AI capabilities, challenging cloud-dependent agentic architectures. By integrating Blackwell GPUs and custom Grace CPUs into small-form-factor desktops, manufacturers are bringing data center-grade compute to individual developers and creators. The collaboration with MediaTek on custom Arm silicon also highlights the continuing reinvention of PC architecture for specific AI demands, moving away from traditional app-centric workflows. What the sources don't address: The expected retail pricing and exact release date for the Acer SFF RTX Spark.