AI factories enter the execution era as Cisco and NVIDIA push rack-scale systems into production - SiliconANGLE

The artificial intelligence infrastructure market is entering a new phase, shifting from merely acquiring GPUs to building complete AI factories capable of generating tokens reliably and at scale. Thi... Read original →

The AI infrastructure sector is transitioning from isolated GPU procurement to full-scale AI factory construction, marking a shift toward execution. Cisco and NVIDIA’s collaboration integrates liquid-cooled computing, AI-optimized networking, and unified OS (NX-OS, SONiC), dramatically reducing time from deployment to token generation. This evolution intensifies demands on upstream semiconductor fabrication, especially at 3nm and EUV nodes, where TSMC faces capacity and yield pressures. Downstream, data centers are rapidly adopting high-efficiency, low-power solutions, accelerating liquid cooling adoption. Geopolitical tensions, particularly around Taiwan and Hong Kong, heighten supply chain risks, compelling firms to diversify and localize sourcing. Competitors like AMD and Intel are racing to establish AI-native architectures to capture standard-setting influence. Over the next 12–24 months, AI factories will evolve into scalable, replicable business platforms, with token/watt efficiency becoming the key performance metric, reshaping the industry toward platform-based, operationally driven models.