Acrab, a Singapore-headquartered technology company building agentic AI compute infrastructure, has raised US$130 million in a Series B round, as investor interest continues to shift from AI applications to the hardware and systems needed to run them.
The round was led by existing backers Vertex Ventures SEA & India and Vertex Growth, with participation from institutional investors across Europe and Southeast Asia. It follows Acrab’s recent emergence from stealth and the launch of its Agent Box platform, powered by the company’s first-generation GΞLIX 1 chip.
The company said the fresh capital will go towards scaling its products, expanding its ecosystem, and developing its next-generation computing platform. Acrab added that it sees “visible paths” to industrial deployments across multiple domains and expects to start generating revenue in 2026.
Also Read: Exploring the ‘Phygital’ world where digital and physical realms converge
The fundraise comes after Acrab’s previous US$350 million fundraising in June, an unusually large amount for a company that has only recently stepped into public view. The new Series B suggests that its investors are backing a long-cycle infrastructure play rather than a conventional software startup looking for rapid commercial rollout.
At the centre of Acrab’s pitch is a simple but ambitious claim: as AI agents become more capable and more personal, they will need to run closer to users, machines, factories, vehicles and devices — not only inside remote cloud data centres.
Moving AI from the cloud to the edge
Acrab is building what it describes as a full-stack AI computing platform, combining purpose-built silicon, edge AI systems, and software orchestration. In practical terms, that means the company is not merely designing chips or building an AI device. It is trying to control the full computing layer needed for AI agents to operate locally.
Its first-generation system-on-chip, GΞLIX 1, is designed to run large language models at the 100 billion parameter scale on local hardware. Parameters are the internal values that help an AI model process and generate outputs; as a rough rule, larger models tend to be more capable but also require more computing power and memory to run.
Today, much of that work happens in the cloud. A user types a prompt into an AI application, and the actual computation takes place in a data centre owned by a hyperscaler or AI infrastructure provider. That model has powered the first wave of generative AI adoption, but it comes with trade-offs: latency, connectivity dependence, energy costs, data privacy concerns, and rising cloud bills.
Edge AI attempts to solve some of those problems by moving computation closer to where data is created. For Southeast Asia, this is not a small distinction. The region has thousands of factories, ports, hospitals, logistics networks, plantations and city systems where connectivity can be uneven, data may be sensitive, and real-time response matters.
A locally running AI agent in a manufacturing plant, for example, could monitor equipment, interpret images or sensor data, and trigger actions without sending every piece of information to a cloud server. In healthcare, on-device AI could support analysis while keeping patient information within a hospital’s own systems. In logistics, AI models running at the edge could help route vehicles, inspect goods, or manage warehouse operations even when networks are congested.
Acrab’s Agent Box is its first visible product in this direction. The company describes it as a personal edge AI system that supports local large-model inference, persistent memory, multimodal interactions, and agent orchestration on-device. Inference refers to the process of running a trained AI model to produce an answer or action. Multimodal interaction means the system can process more than one type of input, such as text, image, voice or video.
Also Read: AI infrastructure: The unsung hero of technological innovation
The “agent orchestration” element is important. The next phase of AI is not just about chatbots responding to questions. It is about software agents that can plan tasks, use tools, remember context, and act across workflows. That creates a heavier infrastructure burden, especially if users expect these agents to be always available, private, and responsive.
Why this matters in Southeast Asia
Southeast Asia has been a fast adopter of AI software, but the region remains heavily dependent on global computing infrastructure. Most startups building AI products still rely on cloud providers and overseas chip supply chains. As demand for AI workloads grows, access to compute has become a strategic constraint, particularly for smaller companies that cannot compete with global technology giants for the latest graphics processing units.
Singapore has positioned itself as a regional hub for AI, semiconductors, data centres, and deeptech financing. That makes it a natural base for companies such as Acrab, even if the market for its products will likely be global from the start. The city-state has the capital networks, research talent, corporate customers and policy support needed for infrastructure-heavy ventures. At the same time, the broader region offers industrial use cases where edge AI could prove useful beyond consumer gadgets.
The challenge is that AI hardware is expensive, slow to commercialise, and difficult to scale. Designing silicon is only one part of the problem. Companies must also secure manufacturing capacity, build developer tools, support software frameworks, manage thermals and power consumption, and convince customers to trust a new computing architecture.
This is where Acrab’s full-stack approach could either become an advantage or a burden. Owning more of the system may allow tighter optimisation between chip, device and software. But it also means the company is taking on several hard problems at once.
A crowded global race
Acrab is entering a field dominated by some of the world’s best-capitalised technology companies. Nvidia remains the clear leader in AI accelerators, with its GPUs powering much of the cloud AI boom. AMD and Intel are trying to capture more of the AI infrastructure market, while Qualcomm, Apple and MediaTek are pushing more AI processing into phones and personal devices.
There is also a growing group of AI chip specialists and infrastructure startups, including Cerebras, Groq, SambaNova, Tenstorrent and Etched, each attacking different parts of the performance, cost and efficiency equation. Some focus on data centres, some on inference, and others on specialised architectures for transformer models, the foundation behind many modern large language models.
Acrab’s distinction, at least from what it has disclosed, lies in its focus on agentic edge infrastructure: running large AI models locally while supporting persistent, on-device agents. That puts it at the intersection of several markets — chips, personal AI devices, enterprise edge systems and AI operating layers. It is a promising but unforgiving position.
From capital to commercial proof
The next test for Acrab will be less about fundraising and more about execution. Deeptech companies often raise large sums before revenue because the upfront cost of research, engineering and supply chain development is high. But investors will eventually expect proof that customers are willing to deploy the technology outside pilots and controlled demonstrations.
Acrab says it expects revenue within 2026 and sees industrial deployment opportunities across multiple domains. That timeline gives the company room to refine its platform, but it also places it in a fast-moving race. AI model sizes, inference techniques and chip architectures are evolving quickly. What looks cutting-edge today can become outdated within a product cycle.
Still, the direction of travel is clear. As AI agents move from novelty to everyday infrastructure, the question of where they run will become more important. Cloud data centres will remain central to training and heavy workloads, but not every AI task can or should travel back to the cloud.
Also Read: Securing Agentic AI for Singapore enterprises: A reference architecture
For Southeast Asia, where digital adoption is high but infrastructure conditions vary sharply across markets, edge AI could become more than a technical preference. It could be the difference between AI that works only in ideal environments and AI that can operate in factories, clinics, farms, ports and homes across the region.
Acrab’s US$130 million Series B is therefore not just another AI funding announcement. It is a bet that the next computing platform will not be defined solely by bigger data centres, but by intelligent systems that sit closer to the real world.
The post Acrab’s US$130M raise signals SEA’s deeper push into AI hardware appeared first on e27.