AI infrastructure bottleneck shifts from chip shortage to power supply
集邦咨询：AI 基建瓶颈正从“缺芯”转向“缺电”
This report provides concrete data on the growing power demand for AI servers, which is a critical concern for anyone involved in building or managing data centers.
TrendForce (TrendForce) reports that the next bottleneck for AI development is shifting from a chip shortage to power and grid issues.
The company estimates that AI server power consumption in data centers will increase from about 25% in 2025 to about 33.
4% in 2026 and possibly exceed 40% in 2027.This means future cloud infrastructure will be more focused on AI.
The report estimates that global data center power demand capacity will reach about 161 GW in 2026, a year-on-year increase of about 31%, while the grid's access and transmission and distribution construction speed may not keep up.
Power access, transformers, transmission lines, and cooling capacity may delay new data center launches.TrendForce believes the supply-demand gap will widen after 2028;
according to its scenario estimates, data center demand capacity could reach 490.7 GW by 2030, while the grid's capacity for data centers is about 222.
6 GW, creating a gap of about 268 GW.Google announced at SEMICON Taiwan 2026 that its TPU system has expanded to a million chip-level scale, and power supply has become the core bottleneck for AI infrastructure expansion.
Google CTO Amin Vahdat pointed out that as AI models continue to grow in scale, the core bottleneck restricting expansion has shifted from the chip itself to power supply, and the ceiling for AI computing power is changing from 'can't make chips' to 'can't find power'.
集邦咨询：AI 基建瓶颈正从“缺芯”转向“缺电”
IT之家 9 月 19 日消息，集邦咨询（TrendForce）于 9 月 16 日发布 AI Server 产业研究， 指出 AI 发展的下一道瓶颈正从芯片转向电力与电网。 TrendForce 估计， AI 服务器在数据中心电力需求中的占比将从 2025 年约 25% 增至 2026 年约 33.4%，2027 年可能超过 40% ，意味着未来云计算基础设施将更多围绕 AI 重构。 报告估计 2026 年全球数据中心用电需求容量约 161 GW、同比增长约 31%，而电网的接入和输配电建设速度未必跟得上；电力接入、变压器、输电线路和散热能力都可能推迟新机房投产。 TrendForce 认为供需背离会在 2028 年后扩大；按其情境估计，2030 年数据中心需求容量可达 490.7 GW，而电网可供给数据中心的容量约 222.6 GW，账面缺口约 268 GW。 IT之家此前报道，在 SEMICON Taiwan 2026 活动中，谷歌宣布其 TPU 系统已扩展至 100 万芯片级规模， 而电力供应成为 AI 基础设施扩张的核心瓶颈。 谷歌首席技术官阿明 · 瓦赫达特（Amin Vahdat）指出，随着 AI 模型规模不断扩大，制约扩张的核心瓶颈已从芯片本身转移至 电力供给 ， AI 算力的天花板正在从“造不出芯片”变成“找不到电”。 参考 TrendForce 集邦咨询：预估 2026 年全球数据中心用电需求年增 31%，电网供应缺口自 2028 年扩大
This report is categorized under AITOP's "Industry" desk, focusing on China's AI industry and domestic foundation model ecosystem. Within our cross-regional analytical framework, it directly maps to AI regulation, compliance frameworks, and sovereign compute policy. Key entities and technical topics under continuous tracking include AI infrastructure, power supply, data center, TrendForce.
Source attribution: curated from IT之家.
