Rebellions CEO: AI Compute Power Has Shifted to Inference

Why It Matters

The shift to inference‑centric, heterogeneous AI hardware reshapes competitive dynamics, creates supply‑chain risks, and opens new investment opportunities across data‑center and consumer devices.

Key Takeaways

• AI compute focus moves from training to inference workloads.
• Hyperscalers pursue heterogeneous custom chips, challenging Nvidia dominance.
• Memory shortages strain HBM, DRAM, and supply chain resilience.
• On‑device AI chips will dominate consumer electronics by 2025.
• Strategic investors target AI coding assistants as next market frontier.

Summary

The conversation with Rebellions CEO Seyun Park centered on a fundamental shift in AI infrastructure: compute power is moving from the training phase to inference, prompting hyperscalers to explore custom, heterogeneous chip solutions beyond Nvidia’s traditional stronghold. Park highlighted that today’s inference workloads often combine memory‑bound and compute‑intensive demands, driving firms like Google, Samsung, and emerging AI labs to diversify their silicon portfolios. Key insights included the growing importance of memory bandwidth, with shortages of HBM, DRAM, and even ancillary components such as PMICs creating a multi‑year supply‑chain bottleneck that may persist until 2027‑2028. The market is also seeing a proliferation of off‑the‑shelf custom chips, exemplified by Google’s TPU availability, and a move toward multiple architectures rather than a single de‑facto standard. Park cited concrete examples: Samsung’s high‑nickel HBM, Apple’s on‑device AI chips, and strategic investors like SK Telecom backing AI‑first ventures such as Anthropic. He also referenced SpaceX’s interest in coding‑assistant chips, underscoring the broader race to capture the next AI application wave. The implications are clear: Nvidia’s ecosystem faces credible challenges, supply‑chain diversification will become a competitive advantage, and on‑device AI is poised to reshape consumer electronics within the next two years, while investors eye AI‑driven software tools as the next trillion‑dollar market.

Original Description

Sunghyun Park, CEO of South Korean chipmaker Rebellions, discusses the latest trends in artificial intelligence and the infrastructure powering major hyperscalers. Speaking to Arjun Kharpal and Chery Kang at CNBC’s Converge Live summit in Singapore, Park weighs in on the AI “super‑cycle,” the global chip crunch and Apple’s decision to name chip veteran John Ternus as its next CEO.