AI News AI资讯 12h ago Updated 2h ago 更新于 2小时前 66

How virtual power plants could provide energy for data centers 虚拟电厂如何为数据中心提供能源

Google is now paying ordinary Americans to turn down their smart thermostats so its data centers can keep humming. That’s the headline from its new deal with Voltus to create a virtual power plant across the PJM grid, the largest in the U.S. The tech giant will fund the setup, Voltus will recruit households with electric vehicles and smart devices to curtail usage or discharge stored energy during grid stress, and Google gets the extra capacity to power its servers. On the surface, it looks like 谷歌刚刚干了一件挺有意思的事:付钱让美国家庭少用电,好给自家数据中心腾出电力空间。这听起来像是某种未来主义的黑色幽默——科技巨头用真金白银购买普通人的“用电克制力”,来维持自己永不眠的算力引擎。这已经不是理论探讨,而是Google与Voltus公司签署的实在协议,将在覆盖美国东海岸的PJM电网内,整合电动汽车、智能恒温器等设备,构建一个容量可达每年100兆瓦的虚拟电厂。

70
Hot 热度
75
Quality 质量
65
Impact 影响力

Analysis 深度分析

Google is now paying ordinary Americans to turn down their smart thermostats so its data centers can keep humming. That’s the headline from its new deal with Voltus to create a virtual power plant across the PJM grid, the largest in the U.S. The tech giant will fund the setup, Voltus will recruit households with electric vehicles and smart devices to curtail usage or discharge stored energy during grid stress, and Google gets the extra capacity to power its servers. On the surface, it looks like a neat, market-based solution to the ballooning energy crisis created by the very industry Google leads. But beneath the slick press release lies a troubling admission: the digital economy is now so voracious that it must pay its neighbors to ration their own electricity to feed its growth.

This isn’t just a procurement deal; it’s a fundamental shift in who bears the cost and risk of our energy transition. For years, the narrative has been that tech giants are building the future with renewable energy credits and massive solar farms. Now, they’re discovering the hard reality that electrons are not just a commodity but a physical constraint. The grid, especially during peak summer or winter events, cannot simply be willed to deliver more. The elegant solution proposed by a Duke University study last year—that data centers could delay tasks for a mere 40 hours a year, unlocking 100 gigawatts of capacity—starts to look less like a clever optimization and more like a stopgap for a system straining under the weight of AI. The proposal presumes a flexibility that doesn’t always exist. When a user asks ChatGPT a question or streams a service hosted on Google Cloud, that demand is immediate. You can’t tell a customer, “Sorry, your query is queued behind a heatwave in Virginia.” Training a model can be scheduled; live inference cannot.

So, Google’s move into the VPP space feels less like innovation and more like an externalization of its operational costs. Instead of building more generation or paying for new transmission lines—long, expensive, and politically fraught processes—it’s creating a market to temporarily reduce local demand. This is “Bring your own capacity,” as Voltus calls it, but it’s really “Bring your neighbors’ capacity.” It’s a clever financial and logistical trick, but does it solve the core problem? 100 megawatts of aggregated response is a decent chunk for a single project, but it’s a rounding error compared to the multi-gigawatt demand of a growing cluster of data centers. This feels like a pilot, a way to secure positive PR and test regulatory waters, rather than a scalable blueprint for powering the AI boom.

The regulatory landscape is shifting to accommodate this reality. Texas has already mandated that large users switch to backup power during grid emergencies, a blunt-force approach to reliability. Some U.S. proposals would let data centers expedite permitting if they commit to demand flexibility. These are essentially governments saying, “You want to build your digital factory? Then you must also build—or buy—your own insurance policy against grid failure.” Google, with its deep pockets, is getting ahead of this curve. It’s buying flexibility now, shaping the market, and ensuring it has a seat at the table when the rules are rewritten. It’s a classic move of using capital to define the infrastructure of the future before public policy catches up.

But let’s be honest about the limits of this model. It relies on a sufficient number of homes and businesses with the right smart devices and a willingness to accept payments in exchange for a bit of discomfort—maybe a slightly warmer house on the hottest day of the year. How scalable is that, really? It depends on penetration of smart tech, on the generosity of payment, and on public tolerance. It also does nothing for the underlying transmission congestion that is the grid’s chronic disease. You can manage demand all you want, but if the wires from West Texas to Dallas or from Virginia’s data center alley to population centers are maxed out, you’re just playing a zero-sum game within a smaller area.

The deeper, unasked question is about fairness and socialization of costs. Data centers are private infrastructure generating private profit. Their energy crunch is now being softened by payments to residential consumers. In effect, Google is using its vast wealth to create a micro-market that mitigates its impact on a public good—the grid. This is a far cry from the era when big industry paid for dedicated power plants. Now, the strategy is to integrate parasitically but politely into the existing system, paying just enough to keep it from collapsing. It’s efficient, yes. But it’s also a profound retreat from the responsibility of building robust, dedicated infrastructure for growth.

Ultimately, this VPP deal is a symptom, not a cure. It shows that the breakneck expansion of AI and cloud computing has outpaced our physical infrastructure’s ability to adapt. Google is buying time and flexibility. It’s a smart, necessary move in the short term. But we shouldn’t mistake this tactical sidestep for a solution. The real test will come when every major tech company tries to replicate this. Who gets priority when thousands of megawatts are being bought and sold on the same local grid? What happens when the payments aren’t enough to entice curtailment during a multi-day grid crisis? This model pushes complexity and risk onto the distribution network and individual consumers. For now, Google gets to claim it’s part of the solution. But the hard truth is that no amount of virtual power plants can indefinitely hide the fact that we are trying to run a 21st-century digital economy on a 20th-century physical grid, and the tech giants at the heart of this boom are still looking for someone else to foot the bill for the upgrade.

谷歌刚刚干了一件挺有意思的事:付钱让美国家庭少用电,好给自家数据中心腾出电力空间。这听起来像是某种未来主义的黑色幽默——科技巨头用真金白银购买普通人的“用电克制力”,来维持自己永不眠的算力引擎。这已经不是理论探讨,而是Google与Voltus公司签署的实在协议,将在覆盖美国东海岸的PJM电网内,整合电动汽车、智能恒温器等设备,构建一个容量可达每年100兆瓦的虚拟电厂。

表面故事很美好:用分布式资源平衡电网,在用电高峰时调用千家万户的电池和调节能力,避免建造更多调峰电厂。这确实是能源互联网的炫酷应用。但剥开这层技术浪漫的外衣,内核却透出一股现实的无奈与狡黠。本质上,这是数据中心行业在为自己庞大而僵硬的电力需求,寻找外部“缓冲垫”的最新尝试。

数据中心的能耗问题,尤其是AI训练带来的峰值负荷,早已是房间里那头必须被喂食的巨兽。传统的解法简单粗暴:建更多电厂,铺更粗的电线。但环评漫长、建设缓慢、成本高昂。于是,像杜克大学那种“只要数据中心每年少用几十小时电,就能多塞进100吉瓦容量”的聪明研究就应运而生。谷歌这次不过是把学术概念落了地,而且选了一条最具美国特色的市场化路径:付费参与。

这里就暴露了第一个值得玩味的点:算力的刚性需求与电力的弹性供应之间的根本矛盾。 训练一个大模型可以推迟几小时,但全球用户点击搜索、使用服务是24小时不间断的实时需求。谷歌此举,等于承认了自身运营无法做到完全灵活,转而向电网和用户购买“灵活性”。这像极了一场精心设计的“用电版众筹”——谷歌出资,Voltus搭建平台,千万家庭提供调节能力,共同为谷歌的数据中心供电可靠性买单。

这种“外部性内部化”的操作,聪明,但也带着点甩锅的意味。数据中心作为能源消耗大户,其责任本应更多地落在自身能效提升和运营模式革新上。例如,是否可以通过更智能的算法调度,让计算负载本身跟随电网信号波动?是否愿意为了电网稳定而真正牺牲部分算力或响应延迟?目前的答案似乎是:我们愿意花钱让别人来解决我们引发的问题。

再看看参与激励机制。Voltus直接付钱给用户,这固然直接有效。但深想一层,这是否会扭曲普通家庭的正常用电行为?为了获得补贴,用户可能在需要舒适用电的时段强行忍耐。而更深层的问题是:这种由商业巨头资助的“灵活性”,是否在事实上让电网的稳定运行,越来越依赖于科技公司的预算和意愿? 当公共基础设施的弹性调节,需要靠私营企业的商业合同来维系,这究竟是能源系统的进步,还是其公共属性被悄然侵蚀的信号?

得州强制大用户在紧急情况下切换备用电源或削峰的法律,则提供了另一种强制性视角。它说明电网的脆弱性已上升到需要法律强制干预的地步。而谷歌的选择,更像一种市场化的、自愿的、但同样是针对自身需求而做的定制解决方案。它解决的是谷歌自己的问题,并顺带为电网提供了溢出价值。

目前这个虚拟电厂的规模,相对于整个PJM电网的峰值需求,依然是九牛一毛。它更像是一个概念验证,或一个漂亮的公关故事。但其示范效应不容小觑:它为科技巨头们指明了一条新路——当直接获取绿色电力或推动政策变革速度太慢时,可以直接下场购买和构建分布式灵活性资源。

最终,这个案例留给我们的疑问,远比答案清晰:我们是否正在走向一个“按需供电”的二元电网系统?在这个系统里,一部分(稳定的、基础的)用电需求由传统电网保障,而另一部分(弹性的、可中断的)需求则由这类商业化的虚拟电厂来满足。如果是,那么谷歌是先锋。但公共电网的公平性和普惠性又该如何保障?是否只有财大气粗的巨头才能玩得起这种“灵活性采购游戏”?

这不再仅仅是一个关于能源创新的故事,而是一个关于责任、成本和权力再分配的故事。科技巨头在扩张其算力帝国时,正在重新定义与基础设施之间的关系——从依赖、到合作、甚至到主导。谷歌付的每一分虚拟电厂费用,都是在为这种新关系支付定金。

Disclaimer: The above content is generated by AI and is for reference only. 免责声明:以上内容由 AI 生成,仅供参考。

芯片 芯片 部署 部署 训练 训练
Share: 分享到: