Jensen Huang 将 AI 竞争力分为五层：能源、晶片、基础设施、模型、应用，并称在最底层「能源」上，中国的能源规模约为美国的 2 倍。BNEF 预测 2024–2030 年间，美国电力需求增长中 38% 来自数据中心，而中国仅 6%；到 2030 年，数据中心将占美国总用电需求近 7%，中国约 2%。中国一面以 solar 与 wind 等可再生能源快速扩张，一面同步新增 coal、nuclear 与 gas；Bloomberg 指出中国 solar 装机容量在 2026 年有望首次超过 coal，但新增可再生能源利用率低于化石资产。
美国发电侧扩张更慢；在经历约两个十年的用电增长平坦后，需求近期回升，AI 需求引发 gas-fired 新项目加速，但更严格的监管与供应链瓶颈使电厂投运往往需要多年；在 Donald Trump 反对可再生能源的背景下，部分清洁能源项目被取消或延宕。电网侧已出现约束：部分公用事业公司告知数据中心接网需数年，北弗吉尼亚等枢纽曾因故障出现断连；BNEF 警告到 2030 年某些市场可能出现实质性电力短缺。对比之下，顾问 David Fishman 认为中国新数据中心接网「不是问题」；Goldman Sachs 分析称，到 2030 年中国可能拥有相当于全球数据中心需求 3 倍以上的富余电力容量。但能源并非唯一变量：Google DeepMind CEO Demis Hassabis 称中国 AI 公司仍落后最前沿约 6 个月，Gartner 分析师 Chirag Dekate 则认为美国在晶片层与模型层仍具创新优势。
Since 2021, China has added more electricity generation capacity across all technologies than the US has added in its entire history; National Energy Administration figures show 543 gigawatts (GW) added in 2025 alone. BloombergNEF projects China will add more than 3.4 terawatts (TW) of capacity over the next five years, almost six times the US, expanding headroom for rising AI data-center load. Elon Musk argues the binding constraint for AI deployment is electrical power, warning chip output may soon exceed the ability to energize it, “except for China.”
Jensen Huang frames AI competitiveness as five layers—energy, chips, infrastructure, models, and applications—and says at the base layer China has about twice the energy the US has. BNEF projects data centers drive 38% of US electricity-demand growth from 2024 to 2030, versus 6% in China; by 2030 they reach almost 7% of total US power demand, compared with 2% in China. China is rapidly building solar and wind alongside coal, nuclear, and gas; Bloomberg reports solar capacity is on track to surpass coal in 2026, though new renewables run at lower utilization rates than fossil assets.
US capacity growth is slower; after roughly two decades of flat electricity demand, demand has begun rising, and AI needs are accelerating gas-fired development, but tighter regulation and supply-chain bottlenecks mean plants can take years to come online; under Donald Trump’s opposition to renewables, some clean projects are canceled or delayed. Grid constraints are already visible: some utilities tell data centers hookups will take years, and northern Virginia has seen disconnections tied to grid faults; BNEF warns effective shortfalls in some US markets by 2030. By contrast, consultant David Fishman calls grid connection for new Chinese data centers a “non-issue”; Goldman Sachs analysts say China may have spare capacity exceeding three times total global data-center demand by 2030. Energy is not decisive alone: Google DeepMind CEO Demis Hassabis says China’s AI firms remain about six months behind the frontier, while Gartner analyst Chirag Dekate argues the US still holds an innovation edge in the chip and model layers.