内容来源：https://aibusiness.com/data-centers/prompt-ai-s-next-bottleneck-is-outside-data-center

内容总结：

AI基础设施扩张面临电力短缺与地方阻力等多重挑战。AI企业正投入数十亿美元扩大算力，但资金并非万能。电力短缺、环境问题及社区反对正使数据中心建设复杂化，且这些问题并非企业单凭自身力量所能解决。本周两则动态印证了上述困境：联邦法官驳回了xAI试图撤销《清洁空气法》诉讼的请求，该诉讼涉及为其田纳西州Colossus 2数据中心供电的密西西比州燃气轮机；OpenAI则在佐治亚州推进电力灵活性计划，通过在未来数据中心降低用电需求来缓解电网压力，但协议关键细节尚未公开。两则案例表明，算力建设日益需要就其对周边社区和电力系统的影响进行协商。核电虽重获关注，但新建电厂耗时数年，难以缓解当前电力短缺。科技公司可以采购GPU、获得融资并签署长期基础设施协议，但仍依赖公用事业公司供电、监管机构审批及社区接纳。随着AI基础设施向电力容量、水资源和本地基础设施已承压的地区扩展，这些依赖关系愈发关键。对企业而言，新增算力延迟上线可能影响AI服务的可用性、成本和时间安排。AI行业竞争的重点或将从“能建多少算力”转向“能实际上线多少算力”。本周AI其他动态：百度Apollo World模型基于数百万英里Robotaxi数据训练，凸显真实世界训练数据对物理AI的价值；AWS深入企业推动AI从试验走向日常运营；谷歌云推出数据与知识管理新功能以提升AI智能体准确性；麦当劳计划2027年在全国推出免下车AI点餐，但部分加盟商对客户接受度存疑；攻击者利用未修补的Flowise AI服务器漏洞窃取凭证并隐匿活动；AI接管常规IT任务引发人才梯队担忧，企业面临如何培养经验较少员工监督AI输出的问题。

中文翻译：

由谷歌云赞助
选择你的首批生成式人工智能应用场景
要开始使用生成式人工智能，首先要聚焦那些能够改善人类信息体验的领域。
人工智能公司正竞相建设数据中心，但电力短缺、监管障碍和当地反对正在延缓扩张步伐。
人工智能行业正投入数十亿美元扩大计算能力，但仅靠资金并不足以确保新数据中心得以建成。
电力短缺、环境挑战和当地反对正在使人工智能基础设施的扩张变得复杂。而且，与数据中心内部的技术不同，这些问题是人工智能公司未必能自行解决的。
本周的两项进展揭示了企业面临的挑战。
一名联邦法官驳回了xAI的请求，拒绝撤销一起涉及《清洁空气法》的诉讼。该诉讼涉及密西西比州的燃气轮机，这些燃气轮机用于为其位于田纳西州的Colossus 2数据中心供电。企业转而采用现场发电以避开公用事业公司的延误，但即便涡轮机安装在拖车上或被称为临时设施，仍须满足空气许可要求。此案表明，即便企业寻求替代传统电网接入的方案，为人工智能基础设施获取电力的努力仍可能引发法律和环境挑战。
与此同时，OpenAI正在采取不同的方式。该公司正在扩大一项电力灵活性倡议，旨在通过在电网承压时让未来的数据中心降低用电量，来缓解当地电力系统的压力。其佐治亚州园区是首个案例，不过该协议的重要细节尚未公开。
这两项进展反映出人工智能基础设施开发商面临的日益严峻的挑战。建设更多计算能力越来越需要就这些能力将如何影响附近社区和电力系统进行协商。
电力短缺和当地反对已经在拖延数据中心的开发，而现场发电也自带挑战。核能正重新受到关注，但建设新核电站需要数年时间，对当下的电力短缺缓解有限。
科技公司可以购买GPU、获得融资并签署长期基础设施协议。但它们仍然依赖公用事业公司输送电力、监管机构批准项目、社区接受正在建设的设施。
随着人工智能基础设施从成熟的数据中心市场扩展到电力容量、水资源可用性和当地基础设施可能已经承压的地区，这些依赖关系正变得更加重要。
对企业而言，其影响不仅限于数据中心建在哪里。新计算能力上线延迟可能影响企业日益期望使用的人工智能服务的可用性、成本和时机。
人工智能行业多年来竞相打造最强大的模型、获取最多的计算能力。人工智能扩张的下一阶段可能较少取决于企业有能力建设多少容量，而更多取决于它们实际能上线多少容量。
本周人工智能领域其他新闻：
百度新世界模型凸显物理人工智能的价值与挑战：百度利用数百万英里的自动驾驶出租车数据训练了其Apollo World模型，凸显了真实世界训练数据对物理人工智能的重要性。
AWS深入企业寻找人工智能的真正价值：AWS正在与客户更紧密地合作，推动人工智能从实验走向日常业务运营。
谷歌以最新一系列功能瞄准智能体人工智能的准确性：谷歌云新的数据和知识管理能力旨在为人工智能智能体提供更可靠的信息以供其工作。
麦当劳目标在2027年实现全国范围内的免下车人工智能点餐：麦当劳计划更广泛地推广自动点餐，尽管一些加盟商对顾客接受度仍持怀疑态度。
Greeter漏洞利用框架在人工智能服务器上隐藏云端窃取行为：攻击者正在利用未修补的Flowise人工智能服务器窃取凭证并探查云环境，同时伪装其活动。
人工智能重塑IT工作，却使人才管道面临风险：随着人工智能接管常规IT任务，企业面临的问题是如何让经验较少的员工培养出监督人工智能产出所需的技能。

英文来源：

Sponsored by Google Cloud
Choosing Your First Generative AI Use Cases
To get started with generative AI, first focus on areas that can improve human experiences with information.
AI companies are racing to build data centers, but power shortages, regulatory hurdles and local opposition are slowing expansion.
The AI industry is spending billions to expand computing capacity, but money alone isn't enough to ensure new data centers are built.
Power shortages, environmental challenges and local opposition are complicating AI infrastructure expansion. And unlike the technology inside a data center, these are problems AI companies can't necessarily solve on their own.
Two developments this week illustrate the challenges companies face.
A federal judge rejected xAI's attempt to dismiss a Clean Air Act lawsuit involving gas turbines in Mississippi used to power its Colossus 2 data center in Tennessee. Companies turning to on-site power generation to avoid utility delays still face air-permitting requirements, even when turbines are mounted on trailers or described as temporary. The case illustrates how efforts to secure electricity for AI infrastructure can create legal and environmental challenges, even when companies pursue alternatives to traditional grid connections.
OpenAI, meanwhile, is taking a different approach. The company is expanding a power-flexibility initiative intended to ease pressure on local power systems by enabling future data centers to reduce electricity consumption when grids are under strain. Its Georgia campus is the first example, although important details of the agreement have not been made public.
The two developments reflect a growing challenge for AI infrastructure developers. Building more computing capacity increasingly requires negotiating how that capacity will affect nearby communities and electrical systems.
Power shortages and local opposition are already delaying data center development, while onsite generation comes with its own challenges. Nuclear power is getting renewed attention, but building new plants takes years, leaving little relief for today's power shortages.
Technology companies can purchase GPUs, secure financing and sign long-term infrastructure agreements. But they still depend on utilities to deliver electricity, regulators to approve projects and communities to accept the facilities being built.
Those dependencies are becoming more consequential as AI infrastructure expands beyond established data center markets into areas where electrical capacity, water availability and local infrastructure might already be under pressure.
For enterprises, the implications extend beyond where a data center gets built. Delays in bringing new computing capacity online can affect the availability, cost and timing of AI services that businesses increasingly expect to use.
The AI industry has spent years competing to build the most powerful models and secure the most computing capacity. The next phase of AI expansion might depend less on how much capacity companies can afford to build and more on how much they can actually bring online.
Also in AI news this week:
Baidu's new world model highlights value, challenge of physical AI: Baidu trained its Apollo World model on millions of miles of robotaxi data, underscoring the importance of real-world training data for physical AI.
AWS goes inside the enterprise to find the real value of AI: AWS is working more closely with customers to move AI beyond experimentation and into everyday business operations.
Google targets agentic AI accuracy with latest flurry of features: New Google Cloud data and knowledge-management capabilities aim to give AI agents more reliable information to work with.
McDonald's aims for nationwide drive-thru AI in 2027: McDonald's is targeting a broader rollout of automated ordering, although some franchisees remain skeptical about customer acceptance.
Greeter exploit framework hides cloud theft on AI servers: Attackers are exploiting unpatched Flowise AI servers to steal credentials and investigate cloud environments while disguising their activity.
AI remakes IT work but puts the talent pipeline at risk: As AI takes over routine IT tasks, businesses face questions about how less-experienced workers will develop the skills needed to oversee AI output.