GIGABYTE unveils AI supercomputer as Africa’s AI race gains pace

GIGABYTE is taking aim at a growing barrier to artificial intelligence adoption in Africa: access to powerful computing without having to depend entirely on expensive cloud infrastructure. 

The computer maker will showcase two NVIDIA-powered AI systems at Ai Everything in Abu Dhabi on Oct. 6-7, including a compact personal AI supercomputer capable of 1 PFLOPS of FP4 performance and a desk-side workstation delivering up to 20 PFLOPS. 

The products arrive as African businesses, developers and smaller technology teams look for more ways to develop AI locally, particularly where data security, proprietary models and computing costs are major considerations.

Compact AI computing 

GIGABYTE’s AI TOP ATOM is built on NVIDIA’s DGX Spark platform and is designed for local AI development, model fine-tuning, data science and workloads involving sensitive information. 

The plug-and-play system offers between 1TB and 4TB of storage and can be connected in a four-node cluster through a high-speed QSFP switch, with 200GbE connectivity per node. 

Using NVIDIA ConnectX-7 networking, the four-node setup can provide up to 512GB of aggregate unified system memory, allowing developers and small and medium-sized businesses to handle larger AI workloads locally before moving them to enterprise data centers or cloud platforms. 

That model could be particularly relevant to African businesses seeking greater control over proprietary and sensitive data while building AI capabilities.

20 PFLOPS for local development 

GIGABYTE is also displaying the W775-V10, a high-performance desk-side AI workstation powered by NVIDIA’s GB300 Grace Blackwell Ultra Desktop Superchip. 

The workstation delivers up to 20 PFLOPS of FP4 performance and 748GB of coherent memory. It can support as many as 400 concurrent requests, positioning it for secure local AI development involving regulated data and proprietary models. 

The systems reflect a broader change in how AI infrastructure can be deployed: development and testing can happen closer to where data is generated, while larger workloads can later be scaled to data centers or the cloud. 

“AI development has traditionally required significant cloud investment or shared clusters, putting smaller teams at a disadvantage,” Jay Lee, general manager for META at Giga Computing, said. 

The company will demonstrate the systems at Booth H3-04 during Ai Everything Abu Dhabi, highlighting a path for developers and businesses to move from local AI experimentation to larger-scale computing infrastructure.