The archive · Hardware & Devices · Product decision · 2024–2026
Acrab bets agentic AI runs at the edge; $480M+ raised since 2024 stealth start
Singapore startup founded 2024 by an ex-Arm China co-CEO raises $130M Series B for edge AI silicon; cumulative funding tops $480M.
Acrab
What the business is
Singapore startup building a full-stack AI computing platform — custom chips, local LLM inference, OS and agent orchestration — to run AI agents at the edge.
Starting capital：Over 480M USD cumulative funding through August 2026, including a $130M Series B backed by Vertex Ventures SEA & India and Vertex Growth; over $350M was raised before exiting stealth.
How it started
Founded in 2024 in Singapore by Ken Phua, former co-CEO of Arm China and earlier a leader at Arm UK; backed from the earliest days by Vertex (Temasek's venture platform) on the conviction that the next wave of AI will run at the edge rather than in the cloud.
What happened
Built GΞLIX, a first-generation compute platform for local LLMs spanning AI silicon, local inference, operating systems, multimodal interfaces and agent orchestration. Emerged from stealth in June 2026 revealing over $350M cumulative financing; launched the GELIX 1-powered Agent Box. In August 2026 closed a $130M Series B, bringing cumulative funding past $480M.
How it ended up
Entering commercialization: expects to begin generating revenue within the year through industrial deployments across multiple sectors; scaling production of Agent Box and the next-generation platform.
Background
Acrab is a Singapore AI infrastructure startup, founded in 2024, betting that the next wave of AI will run at the edge rather than in the cloud. Led by Ken Phua, former co-CEO of Arm China, it is building a full-stack compute architecture — custom AI silicon, local LLM inference, an operating system, multimodal human-machine interfaces and agent orchestration — so AI agents can execute in real time on local hardware.
The company stayed in stealth through early development, backed by Vertex, the global venture platform backed by Temasek, which invested through Vertex Ventures SEA & India and Vertex Growth on the conviction that agents would move to edge computing. Its first-generation platform, GΞLIX, supports local LLMs for agentic AI workloads and was validated in demanding real-world deployments.
Acrab emerged from stealth in June 2026 disclosing over $350 million in cumulative financing, then closed a $130 million Series B in August 2026 that brought total funding past $480 million. It has launched the GELIX 1-powered Agent Box platform, expects to begin generating revenue within the year through industrial deployments, and plans to scale production and its technology ecosystem.
What has to be true
- The thesis is directional and testable: agents need low-latency, always-on compute that the cloud alone can't deliver.
- A founder with Arm pedigree gave investors a credible path to silicon — the hardest part of the stack.
- Staying in stealth let the team validate the platform before facing public scrutiny or revenue expectations.
- Raising enough to build silicon and software together avoids the classic 'great demo, no supply chain' trap.
What can be applied
Bet on where the industry is heading — edge compute — and raise enough to build hardware and software together before revenue; conviction capital plus a validated platform beat a product demo.
Aftermath
As of 2026-09-01, Acrab is entering commercialization: the GELIX 1-powered Agent Box is launched, industrial deployments are targeted to start generating revenue within the year, and the $130M Series B is earmarked for scaling production and expanding the ecosystem. No revenue figures or further rounds have been publicly reported.
Sources
- AI startup Acrab raises $130m in Vertex-backed Series B funding
- Singapore's Acrab raises $350M to build agentic AI compute infrastructure
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card