UK regulators set new rules for Google's AI search 英国监管机构给谷歌AI搜索“立新规”

Looking back to the domestic scene, the State Grid predicts that the maximum load this summer will exceed 1.3 billion kilowatts, an increase of 6%. This number serves as a cool reminder that beneath all the fervent narratives about large AI models and embodied intelligence, what still supports the operation of the modern digital society are these heavy, foundational, and unforgiving physical networks. The 1.3 billion kilowatts is the tangible manifestation of the energy devoured by steel, silicon chips, and algorithms combined. Every training session, every inference of an AI data center ultimately translates into the roar of a power plant and a curve on the grid dispatch screen. We are keen to discuss the "throne" of computing power, but often forget that beneath this throne flows electricity measured in billions of kilowatt-hours. While Silicon Valley elites argue heatedly over the energy efficiency of the next generation of GPUs, China's power grid engineers are diligently constructing 168 peak-shaving projects to ensure that this computing power does not vanish due to a localized overload during the scorching summer. This is a silent, foundational competition that determines the stability of all superstructures.

Meanwhile, the restless news on hot lists outlines another picture of power struggles. Intel's claim to "end Nvidia's computing power monopoly" is less a technological declaration than a carefully orchestrated market psychology war. The battlefield for GPUs is already thick with smoke, but the real revolution may lie in broadening the definition of "computing power." As specialized computing (like Nvidia's GPUs) and general-purpose computing (Intel's traditional stronghold) are reforged in the crucible of AI, players attempting to break the monopoly need not only better chips but also an entire "antifragile system" encompassing software ecosystems, developer relationships, and entirely new business models. At the same time, Microsoft pushing Agents to 1.6 billion Windows users represents another, more covert revolution. It is no longer content with being a tool but attempts to make the operating system itself an autonomous "agent." This is equivalent to pre-installing AI's "hands and feet" into the world's largest desktop ecosystem. If search engine AI is reconstructing "information access," then the agentification of operating systems is attempting to reconstruct the paradigm of "task completion." Users are gradually sliding from active operators to commissioners and supervisors. The power concession and security risks implied here may be more profound than Google's content scraping.

All threads—the drained content ecosystem, the foundational energy base, the white-hot competition for computing power, and the quietly reshaping human-machine interface—point to the same core: we are in a chaotic period where old rules are failing and a new order is yet to be established. The UK regulator's "blanket ban" is simple and crude; Intel's challenge is filled with the anxiety and ambition of a traditional giant; Microsoft's smooth rollout appears gentle but harbors a monumental ambition to reshape habits. Behind this lies a comprehensive renegotiation of content value, energy allocation, computing power control, and interaction sovereignty. This negotiation involves no cordial friendly discussions, only intense jockeying based on real-world interests, technological prowess, and regulatory loopholes. What AI brings is not an idyllic pasture, but an even more naked reconstruction of power. The temptation of winner-takes-all has never been greater, and every role swept into this—whether content creator, power company, chip manufacturer, or ordinary user—is unwittingly becoming a data point and variable in this grand experiment. We enjoy the convenience, but we are also paying a price not yet fully understood. In the end, we may find that the greatest regulatory challenge is not the delisting or launch of any specific product, but how to uphold the bottom line of fairness, sustainability, and human-centric values in an algorithm-dominated world.

英国竞争与市场管理局这纸针对谷歌“AI概览”的整改令，表面上是对一个产品功能的纠偏，骨子里却是内容创作者与科技巨头在AI时代第一次公开、激烈的“分赃谈判”破裂。谷歌的AI概览，本质上是一个终极形态的“内容抽水机”。它将互联网上无数网站辛勤创作的文章、攻略、评测、数据，在用户提问的瞬间，提炼成一段“免费午餐”式的摘要，然后轻飘飘地放在搜索结果顶部。用户满意地离开，谷歌巩固了流量入口的统治，而原始内容提供者？他们的网站点击率、广告收入、甚至品牌认知度，都被这一行AI文字悄然稀释、劫持。谷歌声称这是“提升用户体验”，但“体验”的代价由整个内容生态承担，而利润的绝大部分由谷歌独享。英国监管机构这次要求提供“退出选项”，就像给了内容农场主们一把锤子，可以砸碎那台吸走他们生计的抽水机的电源线。这迟来的正义，恰恰证明了此前商业模式的野蛮。但退出，或许是最消极的抵抗。真正的困境在于，当AI可以高效整合并呈现信息时，谁还愿意费力去搭建和维护那些被吸血的原始“信息矿场”？这不是一个功能开关能解决的生态危机。

视线拉回国内，国家电网预测今夏最大负荷将突破13亿千瓦，增长6%。这个数字冷静地提醒我们，在所有关于AI大模型、具身智能的狂热叙事之下，支撑现代数字社会运转的，依然是这些厚重、基础且不容有失的物理网络。13亿千瓦，是钢铁、硅片和算法共同吞噬能量的具象化。AI数据中心的每一次训练，每一个推理，最终都转化为发电厂的一次轰鸣和电网调度屏幕上的一条曲线。我们热衷于讨论算力的“王座”归属，却常常忽略这王座之下，流淌着的是以亿万度计的电力。当硅谷的精英们为下一代GPU的能效比争得面红耳赤时，中国的电网工程师们正埋头建设168项迎峰度夏工程，确保这些算力不会在酷暑中因一次局部过载而化为泡影。这是一种沉默的、基石性的竞争，决定了一切上层建筑的稳定性。

而热榜上那些躁动不安的消息，则勾勒出权力争夺的另一幅图景。英特尔宣称要“终结英伟达的算力垄断”，这与其说是技术宣言，不如说是一场精心策划的市场心理战。GPU的战场硝烟已浓，但真正的变革或许在于对“算力”定义的拓宽。当专用计算（如英伟达的GPU）和通用计算（英特尔的传统阵地）在AI的熔炉里被重新锻造，试图打破垄断的玩家，需要的不仅是更好的芯片，更是一整套包括软件生态、开发者关系和全新商业模式的“反脆弱体系”。与此同时，微软将Agent推送给16亿Windows用户，则是另一场更为隐蔽的革命。它不再满足于做工具，而是试图让操作系统本身成为一个能自主行动的“代理人”。这等于将AI的“手和脚”，预装进了全球最大的桌面生态。如果说搜索引擎AI是在重构“信息获取”，那么操作系统Agent化，就是在试图重构“任务完成”的范式。用户从主动操作者，逐渐滑向委托者与监督者。这其中蕴含的权力让渡与安全隐患，可能比谷歌的内容抓取更为深远。

所有的线索——被抽血的内容生态、托底的能源基础、白热化的算力争夺、以及悄然重塑的人机界面——都指向同一个核心：我们正处在一个旧规则失效、新秩序待立的混沌期。英国监管的“一禁了之”是简单粗暴的；英特尔的挑战充满了传统巨头的焦虑与野心；微软的平滑推广看似温和，却暗藏重塑习惯的磅礴企图。而在这背后，是内容价值、能源分配、算力控制与交互主权的全面重新谈判。这场谈判没有温情脉脉的友好协商，只有基于现实利益、技术实力和法规漏洞的激烈角力。AI带来的不是田园诗，而是一场更加赤裸的权力重构，赢家通吃的诱惑从未如此巨大，而每一个被卷入其中的角色，无论是内容创作者、电力公司、芯片商还是普通用户，都在无意间，成了这场宏大实验的数据点和变量。我们享受着便利，也正在付出尚未完全看清的代价。最终，我们或许会发现，最大的监管挑战并非某个具体产品的下架或上线，而是如何在一个算法主导的世界里，守住那条关于公平、可持续与人本价值的底线。