AI Infrastructure Bottleneck Shifts from GPUs to Power
Key Questions
What is causing the shift in AI infrastructure bottlenecks?
The bottleneck is moving from GPU scarcity to power infrastructure challenges, including transformers and grid capacity. Half of planned 2026 US data centers are delayed, with transformer lead times reaching 4 years and prices up 77%.
How are companies responding to power infrastructure delays?
Siemens Energy is sold out through 2030, shifting investment focus toward grid hardware, utility deals, and data center operators with secured energization. This impacts planning for large-scale AI deployments.
What role does OpenAI's custom chip play in this trend?
OpenAI's Jalapeño chip reinforces the vertical integration trend as companies seek to control more of their AI hardware stack amid power constraints.
Which investment areas are gaining attention due to these changes?
Investors are focusing on grid hardware providers, utility partnerships, and data centers that have already secured power capacity rather than just GPU suppliers.
What is the timeline impact on US data center projects?
Half of the data centers planned for 2026 face delays because of extended transformer lead times up to 4 years and limited grid capacity.
AI buildout bottleneck is moving from GPU scarcity to power infrastructure (transformers, grid capacity). Half of planned 2026 US data centers delayed; transformer lead times up to 4 years, prices up 77%. Siemens Energy sold out through 2030. This transforms investment focus to grid hardware, utility deals, and data center operators with secured energization. OpenAI's Jalapeño custom chip reinforces vertical integration trend.