Whoever Wins AI Will Count Transformers, Not Nvidia Chips - Forbes

As artificial intelligence (AI) continues to expand rapidly, the bottleneck in AI development has shifted from semiconductor chips to power infrastructure. The article highlights that nearly half of t... Read original →

The AI race has shifted from chips to the grid’s physical backbone. Transformer and switchgear lead times now exceed two years, revealing a critical asymmetry: unlike semiconductors, power infrastructure can’t be scaled rapidly with capital alone. Technically, brute-force GPU scaling hits a hard energy ceiling, pushing model training toward efficiency-driven architectures. Regulatory risk intensifies as FERC and PJM gatekeep interconnection queues, forcing hyperscalers into costly long-term PPAs and stranded asset exposure. Microsoft and Amazon’s rush to secure nuclear and renewable off-takers isn’t just about ESG—it’s energy sovereignty as AI strategy. Within 18 months, data center siting in North America will prioritize substation capacity over fiber or talent. The era of 'compute-first' AI is over; without integrated power access, even the most advanced chips become inert silicon.