newsfilter.io
Conference Presentation

Keynote by Sunghyun Park, CEO & Co-Founder of Rebellions | RAISE Summit 2026

  • Korea is projected to surpass the United States in semiconductor manpower, talent, and manufacturing ecosystem maturity, establishing the region as the primary hub for the next generation of AI chip innovation.
  • Hardware design direction is expected to shift toward "token per seconds per watt and per dollar" metrics as inference becomes commoditized, rendering specific chipset selection less relevant to users.
  • The company anticipates that an abstract software layer (e.g., PyTorch 2.0, VLLM) will further commoditize inference, causing production success to depend on predictable latency, 24/7 reliability, and the ability to handle variable traffic loads up to 32 or 64, rather than raw benchmark scores.
  • Operational goals include delivering 5 to 10 megawatts of NPU capacity within weeks rather than months to support sovereign AI, on-premise, and private cloud deployments.
  • Scalable chiplet architecture enables the rapid addition of functional components like CPUs or I/O dies without requiring new wafer tape-outs, while current chips operating at approximately 600 TDP are expected to eliminate the need for customer liquid cooling and significant CapEx upgrades.
  • A complete AI infrastructure platform will be delivered utilizing custom silicon, servers, racks, and pathing, with a strategic focus on optimizing inference and simultaneously scaling both inference and training solutions.
  • Native support for open-source software stacks (VLM, PyTorch, Hugging Face, Red Hat) is planned, accepting a performance compromise of less than 5% to ensure alignment.
  • Collaborations with ARM are underway to support "Gentic AI" workloads through a hybrid CPU and NPU orchestrated layer, which is deemed critical for seamlessly supporting AGI via combined CPU and NPU architectures.
  • The company proceeds with a "rebellious" strategy driven by conviction in Korea's unique positioning to lead in AI chips, despite acknowledging uncertainty regarding the strategy's ultimate success.
  • Efficient AI deployment and serving at scale is identified as the core definition of winning in the AI market, forming the foundational direction of the company's operations.