未在直播

政治 · 数据中心电力成本

参议院搁置数据中心电力法案，谁来买单仍无定论

即使有57–43的多数票，《电价支付者保护法》仍未能推进。争议的核心是：该法案会要求各司法管辖区审议是否让大型计算客户承担电网成本，但不会强制采纳。

播出 文字 阅读需 8 分钟244 个来源

新闻追踪第 7 天

1 篇报道144 家媒体244 个来源4 个视频

  • 暂无标注日期的事件
  1. 第 1 次更新参议院搁置数据中心电力法案，谁来买单仍无定论当前位置

译自英文原文。

原语言视频

文字实录 · 正在加载

众议院有417名议员投票支持数据中心电力法案，3人反对。两周后，57名参议员支持推进该法案——但这仍然不够。《电价支付者保护法》于9月30日受阻，其拟议的联邦电力用户保障措施未能成为法律。100 5

实际问题足够简单：当一个大型计算园区需要新建发电厂、输电线路或配电设备时，要求扩建的公司是否应该承担费用？还是部分成本可能落到从未要求扩建的家庭和企业身上？

参议院争议的焦点在于，这项法案能否可靠地实现第一种结果。拟议标准要求受涵盖的计算客户承担全部增量升级成本。但司法管辖区只须审议该标准，无须采纳。保护原则与具有约束力的保护措施之间的落差，比两党取得进展和遭遇政治阻挠这两种相互对立的说法，更能解释这场争议。99

9月30日的结果是一次程序性投票失败，并非在最终表决中否决法案。东部时间下午12:51，参议院第254号唱名表决记录显示，57票赞成对推进H.R. 9340的动议进行终止辩论，43票反对；法案需要60票。推进动议随后被撤回。4名民主党人加入共和党人投赞成票：新罕布什尔州的玛吉·哈桑、明尼苏达州的艾米·克洛布彻，以及佐治亚州的乔恩·奥索夫和拉斐尔·沃诺克。5 55

大多数参议院民主党人反对推进该法案，因为他们认为其中的消费者保护措施不足。尽管如此，众议院的支持压倒性明显——而且支持的正是同一项法案，不仅仅是类似提案。众议院书记官记录显示，该法案于9月16日在暂停议事规则下经修正后通过，票数为417–3，另有12名议员未投票。NBC新闻9月30日的报道说众议院是在前一周通过法案；官方记录显示，众议院投票是在参议院投票前两周。3 100 1

标准不要求强制采纳

科罗拉多州共和党众议员加布·埃文斯于6月18日提出H.R. 9340，佛罗里达州民主党众议员凯西·卡斯特作为原始共同提案人。俄亥俄州共和党参议员乔恩·赫斯特德主导了参议院的推动工作。他提出的配套法案S. 5028于7月16日提交，但9月30日的投票针对的是众议院通过的法案。101 102 6

该提案针对的是超大型计算负荷，而非所有被称为数据中心的建筑。受涵盖客户是指非住宅用户，在法案生效当日或之后签约使用主要运行信息技术基础设施以及相关数据存储和计算系统的设施。其总峰值需求须在单一地点或园区达到至少100兆瓦。99

根据拟议标准，费率将向这些客户收取为向其供电所必需的发电、输电或配电升级的全部增量成本。即使客户终止合同或停止购电，剩余成本也仍由其承担。公用事业公司将在进行升级之前取得财务担保或费用分摊款。99

最后这一部分与最初的账单同样重要。基础设施的使用寿命可能超过为其建设的客户。园区项目缩减、离开，或用电量低于预期，都可能留下昂贵的闲置容量。要求在建设前提供财务保障，是应对这一风险的办法，而不是等投资完成后再争论。

但H.R. 9340将通过“审议并作出决定”框架修订1978年《公共事业监管政策法》（PURPA）。州监管机构和不受监管的公用事业公司将在法案生效后一年内开始审议该标准，并在两年内完成审议、作出决定。此前州政府采取的可比行动可以满足这一程序。是否采纳仍属自愿。由于法案未能成为法律，这两个期限均未开始计算。99

参议院民主党领袖查克·舒默称其为“一项毫无实效的宣传法案”。他的反对有明确的立法依据：某个司法管辖区可以完成规定程序，却拒绝拟议的保护措施。这并不意味着文本中没有有意义的成本分摊标准，而是意味着联邦要求不会将该标准变成普遍适用的规定。12 99

赫斯特德的主张是，国会应采取两党共同支持的措施，避免大型负荷的成本落到其他用户的账单上。在9月17日早些时候一次未能通过的尝试中，他曾寻求以一致同意方式通过法案，并表示：7

《电价支付者保护法》是国会能够采取的、保护美国民众免受电价上涨影响的最有意义的两党共同行动。

乔恩·赫斯特德 7

众议院的投票结果让两党共同行动的论点颇具分量，但没有回答关于执行力度的质疑。认同企业应自行承担成本，不等于要求每个司法管辖区都让企业这么做。

预测背后的风险

潜在电力需求的规模使这一区别十分重要。国会研究处援引能源部/劳伦斯伯克利国家实验室的一项历史估算：2024年美国数据中心耗电约192太瓦时，约占美国用电量的4.7%。该机构的2030年情景预测为521–843太瓦时，即9.5–15.3%。这些是预测，并非必然实现的需求。80

另一项由EPRI建立的模型估计，数据中心2030年用电占比约为9–17%，其情景下用电量约为380–790太瓦时。由于假设不同，这些区间不应合并为一组数字。园区项目延期或取消、开发项目储备规模被高估，以及负荷预测下调，都可能导致预测偏高。这种不确定性更凸显了审慎分摊成本的必要性：公用事业公司可能会为最终并未完全出现的需求进行建设。80

穆迪9月的融资情景说明了潜在投资规模。该情景以2030年数据中心耗电约426太瓦时为前提，估计到当年需要新增约45吉瓦发电能力、资本成本约1100亿美元，并使电力系统年度成本额外增加250亿–300亿美元。预计的建设项目主要使用天然气，也包括一些太阳能和储能项目，核电机组重启则很少。79

这些数字既不是已经实现的成本，也不意味着家庭将支付所有未覆盖的费用。分析预计，数据中心可以通过现场或电表后项目直接承担最高约150亿美元；这些项目约占新增发电建设规模的30%，而非年度系统成本的30%。最终由谁付费取决于电价机制、监管规定和实际开发情况。79

俄亥俄州让这场争论切身相关

投票前，路透社9月28日报道称，共和党人希望在中期选举前展示其对电价问题采取了行动。原定投票现已举行，但法案未能推进。俄亥俄州是一个具体的竞选考验：赫斯特德将在11月3日的特别选举中对阵前民主党参议员谢罗德·布朗，争夺J.D.万斯参议员任期剩余部分的席位。4 142

布朗曾批评赫斯特德此前对数据中心和税收激励措施的支持。两位候选人都反对把大型新增负荷的基础设施成本转嫁给家庭，但他们的立场并不相同。赫斯特德推动的是联邦层面的审议提案。布朗则支持让地方决定选址、使用工会劳工、结束过度税收优惠，并防止电费上涨。这些竞选立场本身并不能证明另有一项已生效的联邦成本强制分摊规定。200 198 203 99

民调反映的是政治上的不安，而非选举预测。NBC新闻原始调查发现，64%的人表示较不可能支持“支持在你所在社区建设数据中心的候选人”：其中51%表示极不可能，13%表示有些不可能。11%表示更可能支持，22%表示这不会影响他们。NBC在9月30日播报时简略称51%的人较不可能支持；这指的是强烈反对者，而非全部反对者。213 16 1

这项问题于9月14–15日向A卷的397名登记选民提问，误差范围为±4.92个百分点。NBC委托哈特研究公司/公共意见战略公司开展的总体调查于9月11–15日访问了1,000名登记选民，但其较小的总体误差范围并不能说明这一具体问题的精确度。在九项全国性议题中，只有2%的人把数据中心列为最重要议题。某个议题在全国范围内优先级不高，仍可能在当地成为不受欢迎的立场。213 134

唐纳德·特朗普总统继续以经济和对华竞争为由主张扩大建设。9月29日，他表示，数据中心所在地区“已经变得非常富裕”，而且“他们的税收下降了，而不是上升了”。这些是他的说法，并非经核实、普遍适用的好处。国家竞争和繁荣承诺并不能决定由谁为电网买单。74

已知

  • 该法案未能推进，也未成为法律。其拟议的联邦审议要求和采纳期限均未生效。 55599

未知

  • 目前没有确定的下一次国会投票日期或替代法案。这场争议对选举的影响仍不得而知。

后续

  • 11月3日将举行俄亥俄州特别选举，但仅凭选举结果无法证明数据中心左右了选民的选择。

参议院的投票结果既没有禁止数据中心，也没有为受涵盖客户新设一项全国性义务，要求他们承担电力和基础设施的增量成本。现有州和公用事业公司规则仍是另一回事。对消费者而言，眼下的后果是：国会尚未解决成本分摊问题；即使这项未能通过的提案，也只是要求作出是否保护的决定，而非保证保护本身。99 172

播出原貌 24 行
  1. The Senate blocked a data-center cost bill that 417 House members supported. The vote leaves a proposed federal ratepayer standard unenacted, with families and businesses facing the question of who pays for power plants and grid upgrades built for giant computing campuses. Would this bill have guaranteed that data centers pay those costs—and what protections remain now?
  2. Fifty-seven senators supported advancing the measure—not rejecting it. But they fell three votes short of the sixty needed to overcome the procedural hurdle. On September thirtieth, the Senate rejected cloture on the motion to proceed to H.R. 9340, the Ratepayer Protection Act. That prevented senators from taking up the House-passed bill; it was not a final-passage vote. The motion to proceed was subsequently withdrawn. Four Democrats joined Republicans in voting yes: Maggie Hassan of New Hampshire, Amy Klobuchar of Minnesota, and Georgia's Jon Ossoff and Raphael Warnock. Most Senate Democrats opposed moving ahead because they considered the consumer protections inadequate.
  3. The House had passed this same measure two weeks earlier, on September sixteenth—not during the week immediately before the Senate vote. The tally was 417 to three, with twelve members not voting. Passage came under suspension of the rules, requiring a two-thirds majority. That establishes overwhelming bipartisan passage, despite older bill-status descriptions that still reflected an earlier legislative stage.
  4. Colorado Republican Representative Gabe Evans introduced H.R. 9340 on June eighteenth. Florida Democrat Kathy Castor was an original cosponsor, and the measure was ordered reported from committee on July twenty-first. Ohio Republican Senator Jon Husted championed the Senate effort. His companion, S. 5028, was introduced July sixteenth and referred to the Energy and Natural Resources Committee. But the September thirtieth vote concerned the House bill, not final passage of Husted's Senate companion.
  5. The bill draws its coverage line at one hundred megawatts of aggregate peak demand at a single site or campus. It targets nonresidential customers contracting on or after enactment for facilities principally operating information-technology infrastructure and related data-storage and computing systems. It would not cover every building called a data center.
  6. Its proposed standard would recover the full incremental costs of generation, transmission and distribution upgrades from the large-load customer needing them. That responsibility would continue for remaining costs after the customer terminated its contract or stopped buying electricity. Utilities would obtain financial assurances or contributions before undertaking the upgrades. The purpose is to protect other customers from infrastructure costs left behind when a campus shrinks, leaves, or uses less capacity than planned.
  7. States would have to consider the protection, but would not have to adopt it. That is the decisive distinction. The proposal would amend the Public Utility Regulatory Policies Act of 1978, known as PURPA. State regulators and nonregulated utilities would begin consideration within one year of enactment, then finish considering the standard and make a determination within two years. Comparable prior state action could satisfy that process. The binding federal obligation would be to consider and decide—not necessarily to implement the cost protection. Because the bill did not become law, neither deadline has begun.
  8. The bill contains detailed proposed cost protections, but no nationwide requirement to adopt them. Senate Democratic leader Chuck Schumer attacked that optional structure as insufficient to ensure data centers pay their costs. His criticism does not mean the text contains no protective standard. It means jurisdictions could complete the required process and decline to put that standard into effect.
  9. Husted's goal was to make large data centers bear their electricity and infrastructure costs rather than shift them to other ratepayers. An earlier attempt to obtain passage by unanimous consent was blocked on September seventeenth, separate from the later cloture vote.
  10. His argument rests on taking a bipartisan step toward protection. After the September thirtieth vote, he accused Senate Democrats of rejecting a policy conservative and liberal House members had supported. The House tally supports his claim of bipartisan agreement. It does not settle whether requiring state consideration would deliver the protection households need.
  11. Data centers used an estimated 192 terawatt-hours of electricity in 2024—about 4.7 percent of U.S. consumption. That is a historical estimate, not a measurement of their 2026 use. One set of scenarios projects 521 to 843 terawatt-hours in 2030, or 9.5 to 15.3 percent of U.S. electricity. A separate model projects roughly 380 to 790 terawatt-hours, representing about nine to seventeen percent. Those models use different assumptions and should not be combined into one forecast. Delayed or canceled campuses, overstated development pipelines and downward revisions by grid operators can all make demand forecasts too high.
  12. National electricity sales are projected to reach 4,135 billion kilowatt-hours in 2026 and 4,211 billion in 2027. The 2026 forecast represents about two percent growth. Commercial-sector sales are projected to grow 3.3 percent in 2026 and 2.7 percent in 2027, with data centers and manufacturing among the drivers. These are sales forecasts across the economy—not generation totals, not data-center-only consumption, and not a measurement of how much a family's bill has increased because of a nearby campus.
  13. A separate financing scenario puts the generation buildout at about forty-five gigawatts through 2030, costing roughly 110 billion dollars. It anticipates mostly gas, some solar and storage, and little nuclear restart. Its calculation uses approximately 426 terawatt-hours of data-center consumption in 2030—a different planning assumption from the other demand scenarios.
  14. That scenario estimates twenty-five billion to thirty billion dollars in additional electricity-system costs each year. Data centers might directly cover up to about fifteen billion dollars through on-site or behind-the-meter projects representing roughly thirty percent of the generation buildout. Thirty percent describes the buildout—not a share of annual system costs. These figures are not enacted spending or realized nationwide costs, and they do not establish that all remaining costs would fall on residential customers. The allocation depends on actual development, utility tariffs and regulation.
  15. Sixty-four percent of respondents said they would be less likely to support a candidate who backed building a data center in their community. Fifty-one percent were much less likely, and thirteen percent somewhat less likely. So fifty-one percent is the strongly negative subgroup—not the total negative response. Eleven percent were more likely to support that candidate, and twenty-two percent said it would make no difference. The remaining response category is not established and cannot be labeled. Respondents were asked how information about a candidate would affect their likelihood of voting for that person, then whether a positive or negative effect would be much or somewhat.
  16. The data-center item was asked only on Form A, September fourteenth and fifteenth, to 397 registered voters. Its margin of error is plus or minus 4.92 percentage points. The broader survey ran September eleventh through fifteenth with one thousand registered voters: 724 live telephone interviews and 276 text-to-web interviews. Its overall margin was 3.10 points; the Form A and B half-samples had a 4.38-point margin. Neither smaller margin describes the precision of this particular data-center question.
  17. A separate poll published September sixth found sixty-nine percent opposed building AI data centers in their areas, including forty-five percent strongly opposed. That measures construction opposition, not a candidate's electoral appeal. Its methodology is not established here, so it cannot be merged with the candidate-support results.
  18. Forty-four percent chose neither party on the question of trust over data centers. Thirty-two percent chose Democrats and twenty percent Republicans. The remaining response category and item-level methodology are not established. Those are issue-trust responses, not predicted votes. And only two percent selected data centers as their leading issue among nine national issues in the September survey. Low national priority can coexist with a negative reaction to a candidate backing a project close to home.
  19. Ohio voters will choose between Husted and former Democratic Senator Sherrod Brown on November third, in a special election for the remainder of J.D. Vance's Senate term. Brown has criticized Husted's earlier promotion of data centers and their tax incentives. Before the Senate vote, Republicans were seeking to demonstrate action on electricity-rate concerns ahead of the midterms. That planned vote has now happened, and the measure failed to advance.
  20. Both candidates say major new loads should not shift infrastructure costs onto households. But their records and proposals are not interchangeable. Husted advanced the federal consider-and-determine proposal. Brown calls for local say over siting, union labor, an end to excessive tax breaks and protection against higher electricity bills. He has not produced a separate enacted federal mandate resolving the cost-allocation dispute. Minimum-billing percentages and multi-year exit provisions in Ohio utility tariffs are utility rules—not Brown's personal legislative proposal, and not protections enacted by this failed federal bill.
  21. President Donald Trump continues to promote data-center expansion as economically beneficial and necessary for technological competition with China. His September twenty-ninth remarks asserted that companies would satisfy communities, that host areas had become wealthy, and that their taxes had fallen. Those are political claims—not verified findings that every host community benefits or pays lower taxes. His assertion that China would welcome projects rejected in the United States does not establish a Chinese commitment to accept any particular project.
  22. His comparisons with the Industrial Revolution also varied in certainty. In one September twenty-ninth statement, he pledged not to stifle a technology he described as larger than that transformation. In another, he attributed the comparison to others and explicitly acknowledged uncertainty. Those remarks belong to September twenty-ninth, not the previous day's coverage of his support. Neither formulation answers who pays for expanding the grid or proves household costs will fall.
  23. The answer to the central question is no: this bill would not have guaranteed nationwide that covered data centers pay their full incremental electricity and infrastructure costs. It would have required consideration of that protection, with adoption left optional. The failed vote neither bans new data centers nor sets a national electricity price or construction cap. Existing state and utility rules remain distinct, and actual household exposure depends on those rules and what gets built.
  24. No dated next congressional vote or replacement legislation is established. November third is the next concrete test in the Ohio campaign, but national hypothetical-candidate polling cannot determine how Ohio residents will vote or which issue will decide the race. The proposed federal consideration requirement is not law, and this vote secured no nationwide obligation for covered data centers to pay their incremental electricity and infrastructure costs.

来源

244 个来源144 家媒体4 个视频3 篇通稿

来源地图

追溯本篇报道的依据。图谱展示其关联；下方链接可打开本文引用的来源。

来源图谱将在您阅读时加载。

探索完整的新闻编辑室图谱 ↗

译文更新于

修订记录

  1. r1首次发布。