Chipmakers Renew Nerdy Performance Tussle That Nvidia’s Dominance Had Quashed - Bloomberg.com

Following NVIDIA's prolonged dominance in AI and graphics processing that had suppressed competition, global chip manufacturers are reigniting performance battles. This shift indicates that while NVID... Read original →

NVIDIA’s prolonged AI chip dominance had stifled performance rivalry—until now. The maturation of TSMC’s 3nm node and high-volume EUV adoption has reignited competition, enabling AMD, Intel, and select Taiwanese and Korean designers to challenge with chiplet-based architectures and domain-specific accelerators. This technical cascade is accelerating upgrades across HBM memory stacks, UCIe interconnect standards, and next-gen EDA flows. Geopolitical pressures are inflating compliance costs but also driving regionalized foundry investments. NVIDIA will likely double down on Grace-Hopper iterations and lock in TSMC’s CoWoS capacity, while rivals pursue CUDA-alternative software stacks via heterogeneous integration. Over the next 12–24 months, yield curves at sub-3nm nodes and CoWoS packaging bottlenecks—not raw TFLOPS—will dictate competitive advantage, shifting the battleground from peak compute to system-level efficiency and full-stack optimization.