Former Arm China Co-CEO Returns to Singapore to Build AI Chip Startup

A Homecoming with New Ambitions

Ken Phua is betting his next chapter on a simple premise: after 25 years building someone else's vision, it's time to build his own. The former Arm China co-CEO has returned to Singapore to launch Acrab, a startup developing specialized silicon for agentic AI systems.

The move marks a significant shift for an executive who spent more than two decades at Arm, much of it focused on expanding the British chip designer's footprint across Asia. Now, Phua is applying that accumulated expertise to a different challenge: creating processors tailored for AI agents that can operate autonomously across complex tasks.

Why Singapore, Why Now

Singapore's position as a regional tech hub has attracted a steady stream of semiconductor talent over the past three years, driven by government incentives and proximity to key manufacturing clusters in Taiwan and Malaysia. Acrab joins a growing cohort of chip startups choosing the city-state over traditional strongholds like Silicon Valley or Shenzhen.

The timing reflects broader industry shifts. Agentic AI, which refers to systems capable of independent decision-making and task execution without constant human oversight, represents one of the fastest-growing segments in enterprise computing. Unlike general-purpose AI accelerators, chips designed for agentic workloads prioritize inference efficiency, low-latency decision trees, and distributed processing across edge environments.

Phua's background positions him well for this niche. During his tenure at Arm China, he oversaw partnerships with dozens of Chinese chipmakers and cloud providers, many of whom were racing to build custom silicon for AI inference. That experience gave him a front-row view of where existing architectures fell short and where new opportunities might emerge.

The Technical Challenge

Agentic AI silicon faces distinct engineering trade-offs. Traditional AI chips optimize for training large models or running inference at scale in data centers. Agentic systems, by contrast, often operate in resource-constrained settings where power budgets are tight and latency tolerances are measured in single-digit milliseconds.

Acrab's approach, while not yet publicly detailed, likely involves custom instruction sets and memory hierarchies tuned for the iterative reasoning patterns common in agent-based workflows. These systems must balance on-device computation with selective offloading to cloud resources, a hybrid model that requires careful orchestration at the hardware level.

The startup enters a crowded field. Established players like Qualcomm and MediaTek are extending their edge AI portfolios, while venture-backed entrants from Israel to India pursue similar goals. Success will hinge on Acrab's ability to carve out defensible technical advantages and secure design wins with tier-one customers before capital runs dry.

A Familiar Playbook in New Territory

Phua's playbook at Arm revolved around ecosystem cultivation: working closely with software developers, system integrators, and end customers to ensure that silicon capabilities aligned with real-world application needs. Applying that model to a lean startup presents fresh challenges. Resources are finite, go-to-market timelines are compressed, and the margin for error is slim.

Yet the fundamentals remain. In semiconductors, technical elegance matters less than customer traction. Acrab will need to demonstrate not just that its chips perform well in lab benchmarks, but that they solve pressing problems for enterprises deploying agentic AI at scale - whether in robotics, autonomous logistics, or intelligent edge computing.

Singapore's government has signaled willingness to support such ventures through grants, tax incentives, and access to fabrication partners. Whether that proves sufficient to compete with the deeper pockets and faster iteration cycles of rivals in the United States and China remains an open question.

What Comes Next

Acrab's trajectory will test whether Singapore can nurture not just semiconductor manufacturing and design services, but also product-focused chip startups that compete globally. The city-state has built a reputation for operational excellence and talent density, but translating those strengths into breakout hardware companies has proven elusive.

For Phua, the stakes are personal as much as professional. After decades executing someone else's strategy, he is now accountable only to his own vision - and to the investors and engineers who have chosen to bet on it. The semiconductor industry has never been forgiving of missteps, and the agentic AI window may be narrower than it appears.

The next 18 months will be critical. Acrab must finalize its architecture, tape out initial designs, and begin sampling with early customers. If those milestones slip, or if technical performance disappoints, the startup will join the long list of ambitious chip ventures that ran out of runway before reaching the market.

But if Phua and his team deliver, Acrab could anchor a new generation of Singaporean semiconductor companies - proof that the region can compete not just on execution, but on innovation.