Acrab Raises $130M to Commercialize Agentic AI Compute
Related Stories
AI agents can reason in the cloud. Acrab is betting that the next commercial opportunity lies in helping them remember, coordinate and act directly on local devices.
Singapore-based Acrab has raised a $130 million Series B, with continued backing from Vertex Ventures Southeast Asia & India and Vertex Growth. The company said other institutional investors from Europe and Southeast Asia also participated, but did not disclose their names.
The round follows Acrab’s emergence from stealth with more than $350 million in cumulative financing. The new capital will support product scaling, ecosystem growth and development of its next-generation computing platform as the company moves toward industrial deployments and expects revenue this year.
Founded in 2024 and led by CEO Dr. Ken Phua, Acrab is building a full-stack system that combines purpose-designed silicon, local AI models, software and agent orchestration. Phua previously held senior roles at Arm, including co-CEO of Arm China, according to Vertex.
Its first commercial platform is built around GΞLIX 1, a 5-nanometer edge AI system-on-chip designed to support open-source models in the 100-billion-parameter class. The chip integrates a 20-core Arm CPU, GPU and multicore NPU with 273 GB/s of unified memory bandwidth. These specifications and performance figures are company-reported.
GΞLIX 1 powers Agent Box, a desktop edge AI system designed for local inference, persistent memory, multimodal interaction and coordination across tools and connected devices. Keeping more processing on the device can reduce cloud dependence, retain data locally and maintain core functions when connectivity is limited.
Acrab plans to extend the platform into AI PCs, network-attached storage systems, smart vehicles, industrial robots and service robots through partnerships with manufacturers and developers.
The company is entering a crowded race to define the compute layer beneath agentic AI. Its advantage will depend on whether it can turn an ambitious chip-and-software stack into repeatable deployments across very different device categories.