Nvidia Wants to Sell AI Factories, Not Chips - The Globe and Mail

Nvidia is shifting from being a traditional chipmaker to a provider of complete 'AI factory' infrastructure solutions. While historically focused on selling GPUs, the company now emphasizes building a... Read original →

Nvidia’s pivot from GPUs to 'AI factories' leverages its CUDA moat to lock in customers at the system level, shifting profit pools from chips to integrated infrastructure. This forces upstream EUV and 3nm foundries into tighter architectural alignment and raises software migration costs for cloud providers. U.S. export controls accelerate China’s push for domestic alternatives, yet CUDA’s million-developer ecosystem remains unassailable short-term. AMD and Intel may rally around open stacks like ROCm or oneAPI, but fragmentation limits scale; Cisco could carve out a niche via high-speed interconnects. Over the next 18 months, two parallel AI infrastructure paths will emerge: full-stack integration dominating global commercial deployments, while policy-driven ‘de-Nvidification’ takes root in mainland China, Taiwan, China, and Hong Kong, China—albeit with widening performance gaps.