微软AI GPU面临电力不足，OpenAI推动美国每年新增100吉瓦发电
Satya Nadella already said last year that the company doesn’t have enough electricity to power all its AI GPUs.
Satya Nadella 此前表示微软没有足够电力为全部 AI GPU 供电，芯片可能因无法接入电源而闲置。Rohan Paul 引述 Bg2 Pod 内容指出，瓶颈不在芯片供应，而在缺少靠近大型电源的“warm shells”即已完工的数据中心建筑，OpenAI 正推动美国政府每年新增 100 吉瓦发电作为 AI 战略资产。电力交付与建设需经历多年审批和施工周期，而 GPU 供货以季度为单位扩张，等待中的 GPU 会折旧形成闲置资本。
推荐理由
AI 评分 66
从电力交付与建设周期远慢于GPU供货的角度，解释算力扩张为何可能形成闲置资本。
主题
查看原帖埃隆·马斯克：算力将被困在半成品数据中心里。 > Nscale的IPO文件（2026年9月）：截至8月31日，约25,000块GPU处于活跃状态，而已活跃和已签约的GPU为461,000块。这家新型云服务商承诺的GPU中，约95%尚未投入运行。 > Wood Mackenzie，2026年7月报告：2026年第一季度，发电机升压变压器交货周期超过160周，高于2024年的143周；高压断路器升至125周，而2023年为77周。GPU几个月就能到货。而让它们通电的设备需要两到三年，因此任何在锁定电气设备之前就订购芯片的项目都面临风险。 > 自建燃气发电不再能让AI建设者跳过电力排队，因为涡轮机制造商的订单已排到数年后。 GE Vernova（2026年第二季度）：其燃气设备积压订单和槽位预订量在一个季度内从100吉瓦增至116吉瓦。今天订购的涡轮机要到2031年左右才能到货。现场发电——绕过电网排队的常用办法——也有自己的多年排队。 arkolith
查看原文
Elon Musk: compute will sit stranded in half-built data centers. > Nscale’s IPO filing (Sept 2026): As of Aug 31, about 25,000 active GPUs against 461,000 active and contracted. Roughly 95% of the GPUs this neocloud has committed to are not yet running. > Wood Mackenzie, July 2026 report: Generator step-up transformer lead times exceeded 160 weeks in Q1 2026, up from 143 weeks in 2024, and high-voltage circuit breakers rose to 125 weeks versus 77 weeks in 2023. GPUs ship in months. The gear that energizes them takes two to three years, so any project that ordered chips before locking in electrical equipment is exposed. > Building your own gas power no longer lets AI builders skip the power queue, because turbine makers are booked years ahead. GE Vernova (Q2 2026): Its gas equipment backlog and slot reservations grew from 100 GW to 116 GW in one quarter. A turbine ordered today won’t arrive until around 2031. On-site power, the usual way around grid queues, has its own multi-year queue. arkolith
正文
Satya Nadella already said last year that the company doesn’t have enough electricity to power all its AI GPUs. He said they might end up with a pile of chips just sitting around because there’s not enough power to plug them in.
The issue is not supply of chips but missing “warm shells” (empty but utility-ready building ), meaning finished data center buildings near large power feeds.
OpenAI is pushing the U.S. government to add 100 gigawatts per year of generation as a strategic asset for AI.
This power gap creates stranded capital since GPUs depreciate while waiting for buildings, substations, and transmission to be ready.
Even if training stays centralized, inference demand is the big driver for electricity, so any shift of queries to efficient edge devices would change grid and data center sizing assumptions.
The main problem is, power delivery and construction schedules run on multi-year permitting and build cycles, while GPU supply ramps in quarters.
---
From 'Bg2 Pod' YT channel