What happens to AI if the US can't build data centers?
If the US halts new data center projects due to power and policy constraints, the growth of computing power will face a massive roadblock. While current models will run fine, training the next generation of AI will slow down significantly, forcing the industry to prioritize energy efficiency and token limits.
Original image temporarily unavailable
As multiple US states freeze data center approvals due to mounting electricity and infrastructure pressures, up to 41% of planned capacity faces cancellation risks. If this extreme scenario unfolds, American data center growth would flatline by 2028 at around 120 to 150 GW, cutting expected expansion by half.
While existing massive clusters can still handle current model iterations, building the multi-gigawatt facilities needed for future frontier models by 2030 becomes nearly impossible. This shortage forces a dramatic shift: effective computing growth drops sharply, and labs must choose between training larger models or serving user queries under fixed power limits.
Consequently, the industry will pivot toward efficiency-first strategies, such as optimizing tokens per watt, exploring distributed training, and leaning into post-training or edge computing. This constraint could inadvertently narrow the technology gap for efficient model developers and open new windows for alternative players.
View selection rationale
Importance 88/100
About this brief
- Content type
- News summary
- Source type
- Publisher report
- Original source
- GeekPark 极客公园 ↗
- Source published
- Process
- AI-assisted discovery, translation, summarisation, classification and scoring; standards and corrections are maintained by the editorial desk.
- Verification
- The original URL and publication metadata are retained. This brief is not independently fact-checked unless explicitly stated.
- Published by
- Polyemergence Editorial Desk · PolyEmergence Pte Ltd.
- Conflict disclosure
- No financial or commercial conflict declared.
