Silicon Valley CEO Says Nvidia Has One Weakness — and He Plans to Exploit It - Yahoo Finance

NVIDIA has become the backbone of modern AI infrastructure, with its data center revenue soaring from $15 billion in fiscal 2023 to nearly $194 billion in 2026. However, a Silicon Valley CEO argues th... Read original →

NVIDIA’s GPU architecture, heavily optimized for AI training, reveals an energy-efficiency gap in inference workloads—triggering a cascade across the tech stack. Cloud providers will fast-track dedicated inference ASICs, while EDA and EUV toolmakers must adapt to wafer-scale design paradigms. U.S. export controls on advanced packaging could raise Cerebras’ manufacturing costs, yet its monolithic wafer approach may bypass CoWoS bottlenecks, enhancing supply-chain resilience. Intel and IBM are likely to push hybrid-precision IP licensing, while Cisco could embed inference accelerators into data-center switches. Within 18 months, the AI hardware landscape will shift from ‘GPU hegemony’ to heterogeneous training-inference co-design. If NVIDIA fails to decouple these workloads post-Blackwell, its >90% data-center dominance faces structural erosion.