# AI数据中心扩张的真正瓶颈：电网接入

- 来源：Chubby♨️ (@kimmonismus)
- 发布时间：2026-06-25 20:23
- AIHOT 分数：54
- AIHOT 链接：https://aihot.virxact.com/items/cmqthjf9w03mvsl0e29bfn4h0
- 原文链接：https://x.com/kimmonismus/status/2070120644050657317

## AI 摘要

AI算力需求激增推动数据中心扩张，但真正的瓶颈可能并非芯片或能源生产，而是电网接入。OpenAI与SoftBank在德州的Stargate园区耗资超400亿美元，峰值负载约1.2吉瓦。然而美国电网并网等待时间中位数从2005年的不到20个月增至2023年的55个月。现行先到先得的审批机制导致严肃项目被投机项目阻塞。未来赢家可能不是拥有最佳模型或最多芯片的国家，而是能快速接入电网的国家。

## 正文

We are still not building enough data centers.

That sounds almost absurd， given the scale of the current AI infrastructure boom. OpenAI and SoftBank's Stargate campus in Texas alone is expected to cost well over $40 billion and draw around 1.2 gigawatts at peak load. Such an interesting article by @ChrisGillett

tl；dr：

AI labs need more compute. Compute needs more data centers. Data centers need enormous amounts of electricity. And the real bottleneck may not be chips， GPUs， or even energy generation itself.

It may be the grid！

Before a new data center or power plant can connect， grid operators have to study whether it will overload transmission infrastructure. In the US， the median wait for power plant interconnection reportedly increased from less than 20 months in 2005 to 55 months by 2023.

That is a brutal constraint for an industry trying to scale in months， not decades.

The current system often works on a first-come， first-served basis， which means serious projects can get stuck behind speculative or lower-value ones. The result is a growing mismatch between the speed of AI infrastructure demand and the speed of Western grid bureaucracy.

America may not have an energy shortage. It has a grid connection problem.

And if AI becomes one of the defining infrastructure races of the century， the winners may not just be the countries with the best models or the most chips， but the ones that can actually plug them in.

Highly recommend you read his whole article