INTENSIVE READING · INTERACTIVE

财经外刊精读 · 互动版 US AI data centers face a 34% power shortfall by 2028

2026 年 10 月 8 日 · 星期四 本期：AI 基建 · 电力瓶颈 · 供应链传导

30 秒速读

先建立框架，再进原文。默认只给英文，点摘要段落（或右上角「译」）出中文对译，点数字卡展开中文。

The boom in US AI data centers has run into a wall: Morgan Stanley estimates a net power shortfall of about 32 gigawatts by 2028 — 34% of the electricity those sites will need. The crunch is traveling down the chip supply chain.

美国 AI 数据中心的建设热潮撞上了一堵墙：摩根士丹利估算，到 2028 年净电力缺口约为 32 吉瓦 —— 相当于这些机房所需电量的 34%。这场电力紧张正在顺着芯片供应链向下传导。

32 GW
Net power gap — the shortfall US data-center developers face by 2028, even after counting on-site generation and fuel cells.
净电力缺口 —— 即便把现场发电与燃料电池算进去，美国数据中心开发商到 2028 年仍面临的缺口。
34%
Share of demand — the shortfall as a slice of all the electricity those sites will need.
占需求比重 —— 这个缺口相当于这些机房所需全部电量的多大一块。
9 → 32 GW
Growth path — the gap widens from about 9GW this year to 15GW in 2027 and 32GW by 2028.
扩大路径 —— 缺口从今年的约 9 吉瓦，扩大到 2027 年的 15 吉瓦、2028 年的 32 吉瓦。
$238.90
Nvidia's close — on October 5, up 2.1% and a second straight record, leaving the company worth about $5.76tn.
英伟达收盘价 —— 10 月 5 日收于 238.90 美元、涨 2.1%，连续第二个交易日创新高，市值约 5.76 万亿美元。
$329.17
Capacity price — PJM's clearing price per megawatt-day, up from $28.92 a year earlier.
容量价格 —— PJM 每兆瓦日的结算价，一年前还只是 28.92 美元。
7 in 10
Public opposition — Americans who told Gallup they oppose a data center in their own area.
公众反对 —— 盖洛普民调中反对在自己所在地区建数据中心的美国人比例。

原文精读 · Full Text

33 段（含标题）。点每段右上角的「译」展开中文翻译与结构配色；配图已就近排在相关段落之后。
颜色 = 句子成分，中英同色表示对应
主语 Subject 谓语 / 动词 Verb 宾语 · 表语 Object 从句 · 非谓语 Clause 状语 Adverbial 插入语 · 引述 Attribution
TITLE
US AI data centers face a 34% power shortfall by 2028
美国 AI 数据中心面临34% 的电力缺口到 2028 年
PARA 01
The boom in artificial intelligence data-center construction across the United States has run headlong into a critical obstacle: a severe electricity supply shortage.
美国各地人工智能数据中心建设的这轮热潮一头撞上了一个关键障碍：严重的电力供给短缺。
High-voltage transmission lines: planning and energizing a new line takes five to ten years, while a data center is built in 18 to 36 months.
High-voltage transmission lines: planning and energizing a new line takes five to ten years, while a data center is built in 18 to 36 months. 高压输电线路：规划并让一条新线路真正通电，要五到十年；而一座数据中心 18 到 36 个月就能建好 —— 这个时间差就是缺口本身。 配图：Unsplash 无版权图库素材（非新闻原图），仅作示意。
PARA 02
In a report released on October 5, investment bank Morgan Stanley said US data-center developers will face a net power shortfall of approximately 32 gigawatts by 2028.1
在 10 月 5 日发布的一份报告中，投行摩根士丹利表示美国数据中心开发商到 2028 年将面临约 32 吉瓦的净电力缺口。
PARA 03
The gap amounts to 34% of the electricity those sites will require, and it would be difficult to close even after accounting for stopgap measures such as behind-the-meter generation and fuel cells.2
这个缺口相当于这些机房所需电量的 34%，而且很难补上即便把表后发电、燃料电池这类权宜之计算进来，也是如此。
PARA 04
The shortfall is expected to widen every year, from 9 gigawatts this year to 15 gigawatts next year3 before reaching 32 gigawatts by 2028.
这个缺口预计每年都会扩大，从今年的 9 吉瓦到明年的 15 吉瓦再一路扩大到 2028 年的 32 吉瓦。
The net power gap keeps widening US data-center shortfall still uncovered in 2026-2028, after behind-the-meter generation and fuel cells 2026 gap 9 GW 2027 gap 15 GW 2028 gap 32 GW
Chart 1 | The net power shortfall keeps widening: about 9GW in 2026, 15GW in 2027 and 32GW by 2028 How to read the bars: these are the net shortfalls US data-center developers still face after counting behind-the-meter generation, fuel cells and other stopgap supply - roughly 9GW this year, 15GW next year and 32GW by 2028, which the report puts at 34% of the electricity those sites will require. Source: Morgan Stanley report of 5 October 2026, as reported by Reuters.
PARA 05
Generative AI has driven billions of dollars into data-center investment, but the vast computing and electricity needs of the technology are increasingly running up against the limits of the US power supply.
生成式 AI已经把数十亿美元投进数据中心建设，但这项技术庞大的算力与用电需求正日益撞上美国电力供给的天花板。
PARA 06
A hyperscale AI data center4 uses more than 100 megawatts of electricity, roughly the amount consumed by a mid-sized city.
一座超大规模 AI 数据中心要用超过 100 兆瓦的电，大约相当于一座中等城市的总用电量。
PARA 07
Inside those buildings, the newest racks draw 120 to 140 kilowatts each, and they run flat out around the clock.
在这些建筑内部，最新的机柜每个就要吃掉120 到 140 千瓦，而且 24 小时满负荷运转。
Data-center racks: the newest ones draw 120 to 140 kilowatts each, around the clock.
Data-center racks: the newest ones draw 120 to 140 kilowatts each, around the clock. 数据中心机柜：最新一代机柜每个要吃掉 120 到 140 千瓦，而且 24 小时满负荷运转。 配图：Unsplash 无版权图库素材（非新闻原图），仅作示意。
PARA 08
That demand lands on a system that moves slowly: planning and energizing a new transmission line takes five to ten years, while a data center can be built in 18 to 36 months.
这样的用电需求落在了一张运转缓慢的电网之上：规划并给一条新输电线路通电要花五到十年，而一座数据中心 18 到 36 个月就能建好。
PARA 09
Morgan Stanley does not see the bottleneck putting the 2027 forecasts of Nvidia or Broadcom at risk.
摩根士丹利并不认为这一瓶颈会让英伟达或博通 2027 年的业绩预测面临风险。
PARA 10
The two chipmakers know where their products actually get installed, sell into several regions, and can coordinate early with data-center operators and power suppliers.
这两家芯片厂商清楚自己的产品最终装在哪里，业务分布在多个地区，并且能提早与数据中心运营商、供电方协调。
PARA 11
If chip capacity cannot be deployed because the power is not there, customers may push out deliveries or cancel orders outright.
如果芯片产能因为没电而无法部署，客户可能会推迟交货甚至直接取消订单。
PARA 12
Memory, optical, power-management and analog components face the greatest risk of inventory disruption in that scenario. Memory chips made by Micron, SK hynix and Samsung are especially exposed, because their order schedules can be adjusted fairly easily to match the moment a server is actually switched on.
内存、光模块、电源管理与模拟元件面临最大的库存错配风险在这种情形下。美光、SK 海力士与三星生产的内存芯片尤其容易受冲击，因为它们的订单节奏可以相当容易地调整，以匹配服务器真正通电的那一刻。
PARA 13
Economists have a name for this: the Jevons paradox,5 in which cheaper, more efficient computing pushes total demand higher rather than lower.
经济学家对这种现象有个说法：杰文斯悖论，即更便宜、更高效的算力反而把总需求推得更高，而不是更低。
PARA 14
Power scarcity also changes what is worth buying: when electricity is the binding constraint, the most energy-efficient chips win, and that is a market Nvidia is happy to own.
电力稀缺也改变了「什么才值得买」：当电成为硬约束时，最省电的芯片胜出，而这正是英伟达乐于独占的市场。
PARA 15
Nvidia closed at $238.90 on October 5, up 2.1% and a second consecutive record high, which left the company worth about $5.76 trillion.
英伟达收于 238.90 美元在 10 月 5 日，涨 2.1%，连续第二个交易日创下新高，这让公司市值达到约 5.76 万亿美元。
PARA 16
Nvidia's chief executive, Jensen Huang, has made a similar point: before any computing gear goes into a facility, developers must line up real estate, electricity and the building itself a sequence that runs two to three years.
英伟达首席执行官黄仁勋说过类似的判断：在任何算力设备进机房之前，开发商必须先备齐土地、电力和厂房本身这个流程要走两到三年。
PARA 17
Goldman Sachs also flags mounting constraints on the US buildout, although it expects the near-term impact of political opposition to stay limited.
高盛也指出美国数据中心扩建正面临越来越多限制，不过它预计政治反对的短期影响仍然有限。
PARA 18
Data centers have been a political issue since last year, because of their enormous power use. Ahead of the November midterm elections, campaigns have sprung up across the country arguing that the sites drain power while creating few jobs for local communities.
数据中心从去年起就一直是一个政治议题，自去年以来，原因是它们耗电惊人。在 11 月中期选举之前，全国各地的反对运动已经冒出来，主张这些机房耗掉电力，却只为当地社区创造很少的就业。
PARA 19
The fight is not abstract, and the first people to feel it are the ratepayers, ordinary households on the same grid.6
这场争执并不抽象，最先感受到它的是按表付费的普通住户 —— 也就是同一张电网上的一般家庭。
PARA 20
In Manassas, Virginia, one homeowner watched his monthly electricity bill jump to $281 in January from roughly $100 the month before.
在弗吉尼亚州的马纳萨斯，一位屋主眼看着自己的月度电费跳涨1 月要交 281 美元，而上个月才约 100 美元。
PARA 21
The mechanism behind that jump is the capacity auction7 run by PJM Interconnection, the grid operator for 65 million people across 13 states.
这次跳涨背后的机制是容量拍卖由 PJM 互联电网运营，它是覆盖 13 个州、6500 万人的电网运营商。
PARA 22
The clearing price went from $28.92 per megawatt-day in 2024-25 to $329.17 in 2025-26, and PJM put 63% of that increase down to data-center demand.
结算价从2024–25 年度的每兆瓦日 28.92 美元，涨到 2025–26 年度的 329.17 美元，而 PJM 把其中 63% 的涨幅归因于数据中心需求。
Load curves and prices on screen: PJM's capacity clearing price rose from $28.92 to $329.17 per megawatt-day, and those costs reach household bills.
Load curves and prices on screen: PJM's capacity clearing price rose from $28.92 to $329.17 per megawatt-day, and those costs reach household bills. 屏幕上的负荷曲线与报价：PJM 容量拍卖的结算价一年内从每兆瓦日 28.92 美元涨到 329.17 美元，成本最终摊到用户电费上。 配图：Unsplash 无版权图库素材（非新闻原图），仅作示意。
PARA 23
Virginia residential electricity prices rose about 13% over 12 months, and nearly three-quarters of the state's voters blamed data centers for the spike.
弗吉尼亚州的居民电价12 个月内上涨约 13%，而该州近四分之三的选民把这次跳涨归咎于数据中心。
PARA 24
Nationally, a March 2026 Gallup poll found that seven in 10 Americans opposed a data center in their own area, with 48% strongly opposed and more opposed than to a nuclear plant, which 53% rejected.
在全国层面，2026 年 3 月的一项盖洛普民调发现每 10 个美国人中有 7 个反对在自己所在地区建数据中心，其中 48% 强烈反对反对比例甚至高过核电站（53% 反对）。
PARA 25
Roughly 379 US jurisdictions have imposed moratoriums or outright bans,8 including Indianapolis, which voted 23-1 in August to bar new construction through 2027, and Charlotte, which paused approvals for 150 days.
美国约有 379 个辖区已经出台暂停令或彻底禁令，包括印第安纳波利斯 —— 它在 8 月以 23 比 1 通过决议，禁止新项目在 2027 年前开工，以及夏洛特 —— 它把审批暂停了 150 天。
PARA 26
Local opposition disrupted about 120 projects in the first half of 2026, including at least 45 worth roughly $68 billion stalled in the second quarter alone.
地方反对打断了约 120 个项目在 2026 年上半年，其中仅第二季度就有至少 45 个、价值约 680 亿美元的项目被卡住。
PARA 27
Developers promise jobs, but a typical 250,000-square-foot data center employs only about 50 full-time staff, nearly half of them contractors. Water has also become a flashpoint: a large facility can use up to 5 million gallons a day, and in Georgia's Newton County residents were told water bills would rise 33% over two years.
开发商承诺带来就业，但一座典型的 25 万平方英尺数据中心只雇约 50 名全职员工，其中近半还是外包人员。水同样成了引爆点：一座大型设施每天最多要用 500 万加仑水，而在佐治亚州牛顿县，居民已被告知水费将在两年内上涨 33%。
PARA 28
Some developers are no longer waiting for utility approvals: xAI, now owned by SpaceX, built temporary gas turbines to power its operations while bypassing the grid interconnection queue,9 a move that has triggered lawsuits over air pollution.
一些开发商已经不再等公用事业公司批准：如今归属 SpaceX 的 xAI 就建起了临时燃气轮机在绕开电网并网排队的同时给自己的运营供电，此举已因空气污染引来诉讼。
PARA 29
Bloom Energy expects US data-center computing load to nearly double from about 80 gigawatts in 2025 to 150 gigawatts by 2028, and developers now expect roughly a third of sites to run entirely on on-site power by 2030.
Bloom Energy预计美国数据中心的算力负载将接近翻倍从 2025 年的约 80 吉瓦升到 2028 年的 150 吉瓦，而开发商如今预计，到 2030 年约有三分之一的项目会完全靠自备电源运行。
PARA 30
The limit on AI infrastructure is no longer whether chips can be made, but whether they can be powered and actually switched on.
AI 基建的限制不再是芯片能不能造出来，而是它们能不能拿到电、真正开机运行。
PARA 31
If the power bottleneck persists, pressure in the AI supply chain may travel from the chip designers at the top toward the component makers below.
如果电力瓶颈持续存在，AI 供应链上的压力可能会移动从顶端的芯片设计商，转向下游的零部件供应商。
The power wallHow an electricity bottleneck ends up as an inventory problem1AI demand surgesefficiency lifts total load2Chip orders risememory, optics, power3Power outruns grid100MW+ per campus4Interconnection queue5-10 years to energize5Projects delayedthen orders cancelled6Inventory builds uprisk moves downstreamSix steps: demand surges, chip orders rise, power outruns the grid, the queue delays projects, orders are cancelled, inventory builds up.The dashed line is the Jevons paradox: cheaper computing per watt raises total demand instead of lowering it.
Chart 2 | The power wall: how an electricity bottleneck becomes an inventory problem Editor's schematic, not a chart from the original article. Six steps trace the transmission chain: AI computing demand surges, chip orders rise, electricity demand outruns the grid, the five-to-ten-year interconnection queue delays projects, customers defer and then cancel orders, and component inventory builds up downstream. The dashed line back to step 1 is the Jevons paradox: cheaper computing per watt raises total demand instead of lowering it.
PARA 32
For now, the market is treating the makers of the most efficient chips as the safest place to hide but the constraint has become physical, and physical constraints do not negotiate.
眼下，市场仍把最省电的芯片制造商当作最安全的避风港但约束已经变成物理性的，而物理约束不会跟你谈判。

核心术语 · Glossary

编号与正文上标一一对应，点正文里的数字可跳到对应卡片。默认只显示英文解释（练阅读）；想看中文，点卡片标题。
1power shortfall/ˈpaʊə ˈʃɔːtfɔːl/ 点击看中文解释 ↓

A power shortfall is the gap between the electricity a site needs and the electricity it can actually get. A net shortfall is what is left over after every available stopgap supply is counted.

Why it matters here ▸ Morgan Stanley puts the US net shortfall at about 32 gigawatts by 2028, or 34% of what data centers will need.

电力缺口 —— 一座设施「需要的电」与「实际能拿到的电」之间的差额。所谓「净」缺口，是指在把现场发电、燃料电池等所有能想到的补充手段都算进去之后，仍然供不上的那部分。

2026≈ 9 GW→2027≈ 15 GW→2028≈ 32 GWNet shortfall = power still missing in 2028 after on-site generation and fuel cells; 34% of demand.
拓展知识：读这类数字要先分清口径。毛缺口只算「需求减去在建项目能贡献的容量再减去电网可交付容量」；净缺口会进一步扣掉燃气轮机、燃料电池、核电重启、加密货币矿场改造这些「即插即用」方案。本文的 32 吉瓦是净口径，34% 也是净口径下占总需求的比重。更关键的是它逐年放大（今年约 9 吉瓦、明年约 15 吉瓦、2028 年 32 吉瓦）—— 说明这不是等一等就过去的短缺，而是越拖越大的结构性缺口。
Extended note: Watch gross versus net. A gross gap counts only demand minus capacity already under construction and grid deliverability. A net gap also subtracts plug-and-play fixes such as gas turbines, fuel cells, restarted nuclear and converted crypto-mining sites. The 32GW here is net, and the 34% is a share of net demand. Because the number grows every year, this is a structural gap rather than a one-off shortfall.
2behind-the-meter/bɪˈhaɪnd ðə ˈmiːtə/ 点击看中文解释 ↓

Power generated behind the meter is produced on the customer's own side of the utility meter, so it never travels across the public grid.

Why it matters here ▸ The report's 32GW net gap is what remains even after counting behind-the-meter generation and fuel cells.

表后发电 —— 在用户自己的电表「之后」发电，也就是现场自备电源；这部分电力不经过公共电网，因此不占用并网容量。

拓展知识：为什么这个词重要：美国新建高压线路要五到十年，而「表后」自建燃气轮机、燃料电池、小型核电机组，理论上几个月就能发电。它既是云厂商绕开排队的办法，也是电力板块新的生意来源。代价是环保与监管争议 —— 绕过并网审批，意味着绕过了其他用电户必须遵守的那套规则。
Extended note: Why it matters: a new high-voltage line in the US takes five to ten years, while a behind-the-meter gas turbine or fuel cell can be generating within months. It is both an escape hatch from the queue and a new source of revenue for the power industry. The trade-off is regulatory and environmental pushback, because bypassing interconnection also bypasses the process other customers must follow.
3gigawatt/ˈɡɪɡəwɒt/ 点击看中文解释 ↓

A gigawatt is a billion watts — roughly the output of a large power plant, and enough to supply several hundred thousand homes.

Why it matters here ▸ The 2028 shortfall is measured in gigawatts: about 32 of them.

吉瓦 —— 10 亿瓦，约等于一座大型发电厂的出力，够几十万户家庭用电。

拓展知识：单位阶梯要记牢：1 吉瓦 = 1000 兆瓦（MW）= 100 万瓦。一座超大规模数据中心要用 100 兆瓦以上，也就是 0.1 吉瓦；换句话说，1 吉瓦的缺口约等于十座这样的机房拿不到电。把这两个数字摆在一起，就能直观感受到 32 吉瓦是什么概念 —— 那是三百多座超大规模机房。「算力够不够」的问题，第一次被换算成了「电厂够不够」。
Extended note: Keep the unit ladder in mind: 1GW = 1,000MW. A single hyperscale campus needs 100MW or more, which is 0.1GW, so a 1GW gap means roughly ten such campuses with no power. Put that next to a 32GW national figure and the scale becomes concrete: the discussion has shifted from whether chips can be made to whether power plants can be built.
4hyperscale data center/ˈhaɪpəskeɪl ˈdeɪtə ˈsentə/ 点击看中文解释 ↓

A hyperscale data center is a very large, standardized computing campus built by one of the big cloud operators — Amazon, Microsoft, Google, Meta and their peers.

Why it matters here ▸ Each one draws more than 100 megawatts, roughly the load of a mid-sized city.

超大规模数据中心 —— 由大型云厂商（亚马逊、微软、谷歌、Meta 等）建设的、体量极大且高度标准化的算力园区。

拓展知识：行业把这类客户叫 hyperscaler（超大规模云厂商）。它们在全球新增数据中心产能中占绝对主导，也因此在抢电这件事上最有优势：信用评级高、采购量大、愿意签 15 到 25 年的长期供电合同或租约。反过来，中小云厂商和能效偏低的芯片用户，在电力紧张时最容易被挤出 —— 这是本期「谁受益、谁受损」的底层逻辑。
Extended note: The customers are usually called hyperscalers. They dominate new capacity and therefore win the competition for scarce electricity: strong credit, enormous order books, and appetite for 15- to 25-year power contracts. Smaller cloud operators and low-efficiency chip users get squeezed out, which is the underlying logic behind who wins and who loses in this story.
5Jevons paradox/ˈdʒevənz ˈpærədɒks/ 点击看中文解释 ↓

The Jevons paradox says that when a resource becomes cheaper or more efficient to use, people end up consuming more of it, not less.

Why it matters here ▸ Morgan Stanley notes that computing output per watt is rising fast, yet total power demand keeps climbing anyway.

杰文斯悖论 —— 一种资源的使用效率提高、单位成本下降之后，人们反而消费得更多，而不是更少。

拓展知识：经典例子：更省油的发动机没有降低总油耗，反而让开车变便宜、大家开得更多。放到 AI 上：芯片每瓦算力几年间大幅提升，但单位算力变便宜之后，Agent 这类高负载应用爆发，总用电量继续上升。所以「芯片更省电」并不等于「数据中心更省电」，这是这篇稿子最反直觉、也最值得记住的一层。
Extended note: A classic example: fuel-efficient engines did not cut total petrol use; they made driving cheaper and people drove more. In AI, better performance per watt makes compute cheaper, which unlocks heavier workloads such as agents and pushes total electricity demand up. So more efficient chips do not automatically mean less thirsty data centers.
6ratepayer/ˈreɪtpeɪə/ 点击看中文解释 ↓

A ratepayer is anyone who pays an electricity bill — in practice, the households and small businesses sharing the same grid.

Why it matters here ▸ When a grid upgrades lines for data centers, the cost is spread across all ratepayers, not billed only to the site that caused it.

电费承担者 —— 按电表付费的用电户，实务上主要指与数据中心共用同一张电网的家庭与小商户。

拓展知识：在美国多数地区，电网升级的成本是通过受监管的「费率」回收的：输电线路与变电站的投入会均摊进所有用户的电价，而不是只由引发需求的数据中心买单。容量市场里竞价上涨，最终也会层层传导到零售电价。正因为账单最终落在普通家庭身上，「谁为 AI 买单」才会从技术问题变成选举议题。
Extended note: In most US regions, grid upgrades are recovered through regulated rates, so the cost of new transmission is spread across every customer on the system rather than billed only to the data center that triggered it. Higher capacity-auction prices flow through to retail bills. Because the bill lands on ordinary households, 'who pays for AI' turned into a political question.
7capacity auction/kəˈpæsəti ˈɔːkʃn/ 点击看中文解释 ↓

A capacity auction is a market in which a grid operator buys commitments to be able to generate electricity in the future, so that enough plants stay available.

Why it matters here ▸ PJM's clearing price went from $28.92 to $329.17 per megawatt-day, and PJM put 63% of the increase down to data centers.

容量拍卖 —— 电网运营商通过竞价，购买「未来随时能发电」的承诺，以保证有足够的机组待命。

拓展知识：注意它买的不是电，而是「可用性」。PJM 是覆盖 13 个州、6500 万人的电网运营商，每年举行一次容量拍卖；结算价从每兆瓦日 28.92 美元飙到 329.17 美元，等于为「备用能力」付的钱涨了十倍以上。这类成本最终由用电户分摊，所以数据中心越密集的地区，居民电价压力越大 —— 这也是第 20 段那位屋主账单跳涨的直接原因。
Extended note: Note that a capacity auction buys availability, not electrons. PJM runs one every year for a 13-state, 65-million-person footprint, and its clearing price rose from $28.92 to $329.17 per megawatt-day — more than a tenfold jump in the price of standby capability. Those costs reach households through regulated rates, which is why bills rise fastest where data centers cluster.
8moratorium/ˌmɒrəˈtɔːriəm/ 点击看中文解释 ↓

A moratorium is an official temporary ban — here, a local government barring new data centers for a fixed period.

Why it matters here ▸ About 379 US jurisdictions have imposed moratoriums or outright bans.

暂停令 —— 官方发布的临时禁令；在这里指地方政府在规定期限内不允许新建数据中心。

拓展知识：暂停令通常会写明时长和适用范围，例如「150 天」或「一直到 2027 年」，名义上与永久禁令不同，实际效果一样：项目进不了审批流程，资金只能原地等。对开发商来说，真正的成本不是罚款而是时间 —— 已买的土地、已排队的电力、已签约的客户都在计息。所以 379 个辖区、120 个项目、680 亿美元这组数字，其实讲的是同一件事。
Extended note: A moratorium usually carries a stated length and scope — 150 days, or a set year — and differs from an outright ban in name only: the project simply cannot enter permitting. For a developer the real cost is time, because land, queued power and signed customers all keep costing money. That is why 379 jurisdictions, 120 disrupted projects and $68 billion describe one single problem.
9interconnection queue/ˌɪntəˌkəˈnekʃn kjuː/ 点击看中文解释 ↓

The interconnection queue is the line of projects waiting for a grid operator's permission and engineering work to connect to the network.

Why it matters here ▸ Some developers stopped waiting altogether: xAI built temporary gas turbines instead.

并网排队 —— 等待电网运营商批准、并完成接入工程才连得上电网的项目队伍。

拓展知识：在美国，新建高压输电线路从规划到通电常要五到十年，审批、土地、跨州协调层层叠加；而一座数据中心 18 到 36 个月就能建成。于是「排队时间」成了项目能否落地、何时开始收钱的决定变量，也催生了新生意：手里已有土地和电网接入权的公司，把「快速通电」当作商品卖给云厂商。谁先通电，谁就先收钱 —— 时间本身变成了一种可交易的资产。
Extended note: Planning and energizing a new high-voltage line in the US routinely takes five to ten years, while a data center goes up in 18 to 36 months. Queue time therefore decides whether, and when, a project earns money, which created a new business: selling time-to-power to cloud operators off existing land and grid rights. Whoever energizes first collects revenue first, so time itself became a tradable asset.

读后测验 · Check Your Understanding

3 道题。点选项立即判分；解析需要单独点开。
Q1What exactly does Morgan Stanley estimate for US data centers?
正确答案：B
答案在第 2、3 段：摩根士丹利估算，美国数据中心开发商到 2028 年将面临约 32 吉瓦的净电力缺口，占所需电量的 34%。A 项把缺口说成盈余，方向反了；C 项的「完全覆盖」与原文相反 —— 表后发电与燃料电池是已经被计入之后仍剩 32 吉瓦；D 项把 34% 错当成 34 吉瓦，也漏了「净缺口」这个口径。
US data-center developers will face a net power shortfall of approximately 32 gigawatts by 2028. Option B matches the report's own wording: the figure is net, and it is expressed as a share of demand. A reverses the direction, C turns a stopgap into a solution, and D misreads a percentage as a quantity.
Q2Why can Nvidia and Broadcom ride out the power crunch while memory makers cannot?
正确答案：C
答案在第 9、10、12 段：摩根士丹利认为瓶颈不会危及英伟达与博通 2027 年的预测，理由是两家公司清楚芯片最终装在哪里、业务分布在多个地区，并能提早与数据中心与供电方协调；而美光、SK 海力士、三星的内存订单节奏「可以相当容易地调整」，以匹配服务器真正通电的那一刻，因此最容易出现库存错配。A 项原文没有；B 项属臆造；D 项是常见误解 —— 原文讲的是订单节奏的灵活性，不是单位耗电。
Memory chips made by Micron, SK hynix and Samsung are especially exposed, because their order schedules can be adjusted fairly easily to match the moment a server is actually switched on. Option C mirrors the report: visibility over chip placement plus geographic flexibility on one side, adjustable order schedules tied to server installation on the other. A is unsupported, B is invented, and D confuses power efficiency with scheduling flexibility.
Q3Which statement best captures the argument of the piece as a whole?
正确答案：D
答案在最后两段：文章的核心判断是「限制不再是芯片能不能造出来，而是它们能不能拿到电、真正开机运行」，并指出压力可能从顶端的芯片设计商转向下游零部件供应商。A 项与全文相反 —— 需求仍在快速增长；B 项曲解历史，芯片短缺曾是真实约束；C 项只对了一半，电费既冲击家庭，也决定厂商能否按期交付、能否收到钱。
The limit on AI infrastructure is no longer whether chips can be made, but whether they can be powered and actually switched on. Option D is the thesis, restated in the closing lines. A contradicts the report, B rewrites history, and C collapses a two-sided problem into one side: the same power crunch moves both household bills and corporate delivery schedules.
财经外刊精读 · 互动版 · 2026-10-08 · 由 WorkBuddy 自动生成 · 仅供个人英语精读学习使用，版权归原作者与出版方所有
英文原文：Morgan Stanley 研报（2026-10-05）经 Reuters 报道的英文稿（BusinessKorea 全文转载），并综合 Gallup、Data Center Watch、PJM、Bloom Energy 等公开数据的英文报道整理。原文按精读节奏切分为 33 段，引用的原句文字未作改动。
数字交叉核对：Reuters 报道的摩根士丹利研报（2026-10-05）、The Economic Times（Reuters 通稿）、Financial News 英文版、BusinessKorea（英文全文，含 34%／32 吉瓦／9-15-32 吉瓦的逐年口径，以及英伟达 10 月 5 日收盘 238.90 美元、市值约 5.76 万亿美元）。
背景数据：Gallup 民调（2026-03，七成反对／48% 强烈反对）、Data Center Watch（379 个辖区、约 120 个项目受阻）、PJM Interconnection 容量拍卖结算价（每兆瓦日 28.92 → 329.17 美元）、Bloom Energy 数据中心负载预测（80 → 150 吉瓦）。
配图：Unsplash 无版权图库素材（非新闻原图），图表按公开数据自绘；本材料不构成任何投资建议。
← 全部精读