Source: SiliconANGLE
The economics of generative AI are reversing a decade-long cloud migration. Companies are building private data centers to run large language models because per-token costs at cloud providers have become unsustainable at scale, while regulatory and competitive pressure push them to keep sensitive training data offline. This shifts enterprise infrastructure spending from SaaS consumption toward capital-intensive hardware procurement—a move that erodes the margin-expansion model cloud vendors (AWS, Azure, Google Cloud) have depended on and benefits on-prem infrastructure vendors and specialized silicon makers like NVIDIA.