South Korea’s leading AI chip startups are sharply expanding their workforces as the sector enters a critical phase: commercializing next-generation processors designed to run artificial intelligence services.
Industry sources and data from Korean startup investment database TheVC show that four prominent Korean AI fabless firms — Rebellions, DeepX, Mobilint and FuriosaAI — have significantly increased headcount over the past year as their latest chips move toward initial mass production or customer validation.
Fabless companies design semiconductors but outsource manufacturing to external foundries.
The hiring surge comes as these companies push new neural processing units aimed at AI inference, the stage where trained AI models are deployed to deliver real-time services. Inference chips are typically optimized for lower power consumption and operating cost than processors used to train AI models.
Edge AI chip developer DeepX roughly doubled its workforce from around 70 employees early last year to more than 140 as of early 2026, according to industry data. Mobilint, which develops AI processors for industrial and on-device applications, also nearly doubled its staff to about 105 employees.
Data center-focused startup Rebellions expanded from roughly 200 to nearly 300 employees, while FuriosaAI, another inference chip developer targeting enterprise servers, grew from about 100 to around 130 and aims to reach 200 this year.
Next-generation chips move into production and testing
The staffing increases coincide with key product milestones across the companies.
Rebellions has begun mass production of its second-generation AI processor, REBEL-Quad, and is conducting proof-of-concept testing with global partners. PoC refers to pilot testing with potential customers before wider commercial deployment.
FuriosaAI said in January it received an initial production batch of about 4,000 units of its second-generation Renegade AI inference chip from manufacturing partner TSMC. The chips are being assembled into accelerator cards for enterprise systems.
Mobilint plans to begin mass production of its Regulus chip later this year after completing PoC work with potential customers. DeepX is already producing its first-generation DX-M1 chip while developing its next-generation DX-M2 processor.
The four companies are pursuing different segments of the AI inference market. Rebellions and FuriosaAI are largely focused on data center and enterprise workloads, while DeepX and Mobilint target edge and on-device AI applications such as robots, cameras, vehicles and industrial systems, where low power consumption and compact deployment are critical.
Their progress is closely watched because South Korea, despite global leadership in memory semiconductors through companies such as Samsung Electronics and SK hynix, has yet to produce globally competitive firms in AI logic and accelerator chips.
Government policy has also begun to focus more directly on the sector. In February, the Ministry of Trade, Industry and Energy said it would invest 1 trillion won ($678 million) over five years in a joint development program for on-device AI semiconductors aimed at accelerating commercialization and expanding domestic adoption.
mjh@heraldcorp.com
