数据中心供需Data center supply & demand
CBRE 北美 8 个主要市场：空置率、机房租金、在建和预租容量。CBRE’s 8 North American primary markets: vacancy, asking rents, construction and preleasing.
每月3日09:05（北京时间）检查；按上游报告期更新。当前报告期：Check on the 3rd monthly at 09:05 (Asia/Shanghai); update with new source reports. Current period: H1 2026
主要市场空置率Primary-market vacancy
1.4%
H1 2026 · 期末存量指标Period-end stock measure
在建容量Under construction
7,481.1
MW · H1 2026
在建中已预租比例Share of construction preleased
80.4%
已预租MW ÷ 在建MWPreleased MW / construction MW
当期供需快照Current supply and demand
| 指标Metric | 观测值Value | 单位与期间Unit and period |
|---|---|---|
| 已建成存量Existing inventory | 10,902.9 | MW · H1 2026 |
| 净吸纳量Net absorption | 1,456.2 | MW · H1 2026 · 报告期内流量Flow during reporting period |
| 250–500 kW 平均要价250–500 kW average asking rent | $204.69 | USD/kW/month · H1 2026 |
容量以供电规模MW计，不能直接换算为GPU数量。CBRE样本包含AI和传统工作负载，不代表全球AI机房总量。Capacity is measured in MW of power and cannot be directly converted to GPU counts. CBRE’s sample includes AI and conventional workloads; it is not a global AI data-center total.
机房租金：按部署规模Data-center asking rents by deployment size
这是机房供电容量的月度要价，单位不同于GPU小时租金。旧年份与最新半年期保留原报告标签。These are monthly asking rents per kW, a different unit from GPU-hour rental prices. Annual observations and the latest half-year retain their original reporting labels.
正在读取图表…Loading chart…
在建与已预租容量Construction and preleased capacity
预租容量是在建容量的子集，两条线不能相加；两者均为期末存量。Preleased construction is a subset of total construction; do not add the lines. Both are period-end stocks.
正在读取图表…Loading chart…
在建容量预租比例Share of construction preleased
由官方图表的预租与在建MW计算。高预租比例不代表这些容量已投运。Calculated from the publisher’s preleased and construction MW. High preleasing does not mean capacity is operational.
正在读取图表…Loading chart…
主要市场对照Primary markets
| 市场Market | 存量 MWInventory MW | 空置率Vacancy | 在建 MWConstruction MW | 期内吸纳 MWPeriod absorption MW | 租金区间 $/kW/月Rent range $/kW/month |
|---|---|---|---|---|---|
| Northern Virginia | 4,496.5 | 0.2% | 2,420.2 | 467.6 | $190-$235 |
| Atlanta | 1,803.2 | 1.6% | 2,882.0 | 343.0 | $185-$200 |
| Dallas-Ft. Worth | 1,432.8 | 2.8% | 767.7 | 352.0 | $160-$185 |
| Phoenix | 1,069.2 | 1.1% | 248.0 | 260.4 | $175-$215 |
| Chicago | 910.6 | 2.2% | 706.8 | -3.3 | $195-$225 |
| Silicon Valley | 509.2 | 4.2% | 144.0 | 22.0 | $200-$275 |
| Hillsboro, OR | 491.4 | 0.2% | 170.3 | 15.9 | $185-$225 |
| New York Tri-State | 190.0 | 7.2% | 142.1 | -1.4 | $190-$235 |
地区租金区间为250+ kW、N+1/Tier III要求的要价，不能作为250–500 kW平均值的上下限。Regional asking ranges cover 250+ kW at N+1/Tier III requirements; they are not bounds for the 250–500 kW average.
历史容量与期间Historical capacity and periods
| 原报告标签Source period | 在建 MWConstruction MW | 预租 MWPreleased MW | 预租比例Preleased share | 净吸纳 MWNet absorption MW | 吸纳量周期Absorption period |
|---|---|---|---|---|---|
| 2020 | 457.8 | 238.5 | 52.1% | 329.6 | 全年Full year |
| 2021 | 727.6 | 321.7 | 44.2% | 493.4 | 全年Full year |
| 2022 | 2,709.2 | 1,297.8 | 47.9% | 751.3 | 全年Full year |
| 2023 | 3,245.2 | 2,687.1 | 82.8% | 1,006.3 | 全年Full year |
| 2024 | 6,350.1 | 4,840.2 | 76.2% | 1,809.5 | 全年Full year |
| 2025 | 5,994.4 | 4,799.9 | 80.1% | 2,497.6 | 全年Full year |
| H1 2026 | 7,481.1 | 6,012.0 | 80.4% | 1,456.2 | 半年Half-year |
净吸纳量是流量：全年和半年不能直接比较。图表中年份标签映射到12月31日，H1映射到6月30日，仅用于定位报告期；未制造日频历史。Net absorption is a flow: full-year and half-year totals are not directly comparable. Annual chart labels map to December 31 and H1 to June 30 solely to locate reporting periods; no daily history is implied.
数据来源与差异Source and discrepancies
数值直接读取官方报告内嵌图表的数据，并核对地区合计和报告期。租金历史采用当前报告图表。H1 2026图表的2025年值为196.25美元，而H2 2025报告摘要曾出现195.94美元。本站保留当前图表口径，不对两个数值取平均。Values come directly from the official report’s embedded chart data, with regional totals and periods checked. Rental history follows the current report chart. The H1 2026 chart gave $196.25 for 2025, while the H2 2025 summary also quoted $195.94. We retain the current chart definition and do not average the conflicting figures.
当前只展示可复核的空置率快照，未据此构造连续空置率历史。上游可能调整主要市场范围或修订历史，引用时应保留报告来源。Vacancy is shown as a verified snapshot; no continuous vacancy history is inferred from it. The publisher may change market coverage or revise history, so retain the report source when citing.