Tech Rivals Forge New AI Chip Alliances
Several strategic partnerships are reshaping the AI hardware landscape as companies seek to reduce reliance on single suppliers. ByteDance is partnering with Samsung to co-develop custom AI inference chips, while Samsung has also inked a research deal with Applied Materials. Separately, EPC and Renesas have signed a deal for high-efficiency GaN power technology.
- ByteDance's custom chip initiative, reportedly codenamed SeedChip, is focused on AI inference to power applications like video recommendations and large language models. The company aims to produce at least 100,000 of these chips this year, potentially scaling to 350,000 units. - A key motivation for ByteDance's partnership with Samsung is securing access to high-bandwidth memory (HBM) chips, which are in critically short supply globally due to the AI infrastructure build-out. - Samsung's collaboration with Applied Materials is centered on the EPIC Center, a $5 billion R&D hub in Silicon Valley, to jointly develop sub-2-nanometer process technology and advanced 3D stacking for future AI chips. - Applied Materials is introducing new manufacturing systems to enable Gate-All-Around (GAA) transistors for 2nm-and-beyond nodes, a critical architectural shift for improving the energy efficiency of AI chips. - The EPC and Renesas deal combines Renesas' high-voltage GaN portfolio, strengthened by its acquisition of Transphorm, with EPC's low-voltage eGaN technology. This allows them to offer a comprehensive GaN portfolio for applications ranging from 1V to 650V+. - The GaN technology alliance between EPC and Renesas aims to improve power efficiency in AI data centers, where GaN offers faster switching speeds and smaller form factors compared to silicon. - Samsung Foundry is aggressively pursuing AI chip orders for its 2nm GAA process, targeting a 130% increase in orders for 2026 and securing clients like Tesla for its next-generation autonomous driving chips. - The broader trend toward on-device AI is driving the development of specialized hardware like Neural Processing Units (NPUs) integrated into SoCs to handle AI inference locally, enhancing privacy and real-time performance.
