AI 驱动的稀土需求与地缘政治考量
本视频旨在深入解析美国为何对格陵兰表现出浓厚兴趣，以及这一兴趣如何与人工智能（AI）、中国以及硅谷的战略布局相互关联。在深入探讨之前，值得一提的是，本期节目由 Hrefs 的营销平台赞助，特别是其“Brand Raider”功能，该功能旨在帮助企业在 AI 时代实现增长。许多企业热衷于提升 Google 排名和 SEO 表现，但您是否曾核实过您的业务是否在 ChatGPT、Perplexity 或 Gemini 等 AI 搜索和问答系统中有所体现？您的品牌是否被提及于 YouTube 摘要或 Reddit 论坛中？如果答案是否定的，或者您不确定品牌在 AI 搜索中的表现，这便是一个严峻的问题，因为搜索行为已发生根本性改变。如今，人们几乎所有的事情都会通过 AI 进行研究，无论是寻求合作对象、消费决策、投资方向，还是其他任何领域。若您的品牌未出现在这些 AI 的搜索结果中，您就等于隐形了。Hrefs 的 Brand Raider 功能能够精确展示您的品牌在各类 AI 搜索引擎中的呈现方式，揭示用户在查询您时所看到的信息，统计 AI 聊天中提及您品牌的百分比，并找出 AI 关于您品牌的评价。最关键的是，它能指出您需要改进之处，以确保品牌在 AI 搜索中脱颖而出。您可以借此发现 AI 可见性上的差距，了解哪些内容被 AI 采纳，追踪 AI 的回答，并围绕提示词而非关键词来定位您的品牌。如果您正在经营一家企业或管理一个品牌，并希望 AI 能够了解您，不妨试试 Hrefs。顺带一提，Brand Raider 对所有 Hrefs 的免费和付费用户均可用。好了，让我们回到视频本身。
时间回到 1946 年，第二次世界大战刚刚结束。美国总统杜鲁门收到一份来自其国务卿的绝密备忘录，主题是“格陵兰”。备忘录直截了当：我们需要从丹麦手中购买它，价格为一亿美元，全现金交易。冷战即将开始，北极是连接美国与苏联导弹部署的最短航线，而格陵兰恰好位于关键的飞行路径上。谁控制了格陵兰，谁就掌握了核战争预警系统中一个至关重要的节点。然而，丹麦拒绝了。快进到 80 年后，特朗普总统再次提出对同一岛屿的兴趣。这一次，关注点不再是导弹，而是这些——磁铁。这些用于冷却 AI 数据中心的磁铁。因为在杜鲁门时代的冷战与当今的 AI 繁荣之间，美国已将维持这些机器运转的关键组件的整个供应链外包。而现在，中国几乎完全控制了它。然而，即使美国明天就获得了格陵兰，这也很可能无法解决中国带来的问题。今天，我们将聚焦格陵兰，从科技和 AI 的视角，探讨稀土以及下一场中美科技战争的前沿。
让我们深入了解。当您生成一张 AI 图像或向 Gemini 提问时，您的文字会被转化为向量，输入到一个庞大的概率模型中。但除此之外，您还在激活一个相当复杂的供应链，而这个供应链涉及稀土永磁体。一块单一的芯片大约消耗 0.3 瓦时（Wh）的能量，而一次典型的谷歌搜索消耗的能量大约是其一千倍。如此巨大的额外能量消耗会在数据中心内部转化为热量。这些热量不会自行消散。每一瓦特的计算能力都会产生热量，必须被积极地移除，而移除过程依赖于冷却系统。这些冷却系统大量使用了钕铁硼磁体。它们看起来就是这样。这些磁体被用于冷却风扇和高密度机架中，以实现快速的热响应，本质上是为了散热。问题在于，这些用于冷却数据中心的磁体几乎完全在中国制造，并且受到中国去年 12 月份大幅收紧的出口管制。现代 AI 依赖于模型、芯片和数据中心。数据中心要正常运行，尤其是在配备了高密度 GPU 时，需要庞大的冷却和电力系统。稀土元素存在于这个系统中的多个层面，从冷却风扇、水泵到存储、变压器和发电机。因此，当您听到德克萨斯州有新的 AI 数据中心项目时，您也必须明白，这意味着对这些磁体的需求也在同步增长。那么，格陵兰与这些磁体、冷却系统和 AI 有何关系？可以说，它们之间存在着千丝万缕的联系。
Original English
By the end of this video, you will understand exactly why the US wants Greenland, how it connects to AI, China, and Silicon Valley. Stick around. But before we start, this episode is brought to you by Href's marketing platform, and their feature called Brand Raider, which helps businesses grow in the AI era. You know how businesses are obsessed with Google rankings and SEO, but have you actually checked if your business is showing up in Chad GBT or Perplexity or Gemini? Is it being mentioned in YouTube summaries or Reddit threads? And if it doesn't or you're not sure how it shows up in the AI search, it's a problem because the search behavior has changed forever. People use AI to research everything these days, whether they want to work with you, who to buy from, what to invest in, and if you're not in those results, you're invisible. Brand Raider shows you exactly how your brand appears across AI search engines, what people are seeing when they ask about you, what percentage of AI chats mention your brand, find out what AI says about your brand, and this is the key part, what you need to fix to actually show up. You can spot AI visibility gaps, see what content is being picked up by AI, track AI answers, and position your brand around prompts instead of keywords. If you're running a business or a brand and want AI to know about you, check out HREFs. And by the way, Brand Raider is available to all free and paid ATFre users. All right, let's get back into the video. It's 1946. World War II just ended. President Truman receives a classified memo from his Secretary of State with the subject line, Greenland. The memo is blunt. We need to buy it from Denmark. For hund00 million, all cash. The Cold War is starting. The Arctic is the shortest route between the US and Soviet missiles. And Greenland sits directly in the flight path. Whoever controls Greenland controls one of the most critical pieces of the warning system for nuclear war. Denmark says no. Fast forward 80 years. President Trump. Same island, same interest. Except this time, it's not so much about the missiles. It's about this, the magnets. The magnets that cool AI data centers. Because somewhere between Truman's cold war and today's AI boom, the US outsourced the entire supply chain for the thing that keeps these machines running. And now China controls it, almost all of it. The only problem is that even if the US took Greenland tomorrow, it most likely would not fix the China problem at all. Today is about Greenland in the context of tech and AI, rare earths, and the next US China tech warfront. Let's dive in. What happens when you generate an AI image or ask Gemini a question? Your words are turned into vectors. that are fed into a giant probability machine. But what else? You're activating a pretty complex supply chain. And this supply chain involves rare earth permanent magnets. A single chip uses about.3 W hours. A typical Google search is roughly a thousand times less. All of that extra energy turns into heat inside a data center. And this heat doesn't dissipate on its own. Every watt of compute generates heat that must be actively removed and that removal depends on the cooling systems. The cooling systems are packed with neodymium iron boron magnets. And this is what they look like. The magnets are used in cooling fans and high density racks for rapid thermal response for basically cooling. And the problem is that these very magnets that cool down data centers are manufactured almost entirely in China and they're sitting under Chinese expert controls that tightened pretty drastically just last year in December. Modern AI depends on models, chips, and data centers. For a data center to function when it's got power dense GPUs, it needs a massive cooling and power stack. rare earths are inside that stack at multiple layers from cooling fans to pumps to storage to transformers to generators. So when you hear new AI data center project in Texas, you also have to understand that that means that the demand for the magnets is scaling as well. So what does Greenland have to do with these magnets, cooling systems, and AI? Well, just about everything.稀土供应链的变迁与中国的崛起
要理解这一切，我们必须回顾历史。在 1985 年，美国控制着全球约 60% 的稀土产量，其中大部分来自加利福尼亚莫哈韦沙漠的“山脉山口”（Mountain Pass）矿。而中国当时仅生产约 30%。美国曾是全球稀土开采、加工和磁体制造的领导者，从矿石到最终产品，他们主导了整个产业链。然而，形势发生了逆转。在 20 世纪 90 年代和 21 世纪初，中国投入了大量国家资金用于污染严重的稀土加工以及磁体工厂的建设，而西方公司则乐于关闭工厂并将生产转移到中国。因此，当人们意识到时，几乎整个供应链都已迁移至中国。
从 2006 年开始，连续九年，中国开始“武器化”其控制权，限制出口，并在 2010 年将供应量削减了 40%。随后，在与日本因一艘被扣押渔船船长发生的海洋争端中，中国更是长达两个月完全停止了对日本的稀土出口。日本制造商陷入恐慌，因为当时日本 90% 的稀土进口依赖于中国。日本的禁运是一个分水岭时刻，因为它第一次让世界看到中国愿意将稀土供应作为地缘政治武器。作为回应，日本积极推动供应链多元化，将对中国稀土的依赖从 90% 降至 60%，并将总稀土消耗量减半。2010 年的中日稀土禁运事件，对所有人来说都是一个警钟。
2014 年，世界贸易组织（WTO）裁定中国的出口配额违反了贸易规则，因为中国无法以保护环境或资源保护为由对其进行合理化解释。中国被迫取消了正式的配额，但保留了出口许可证制度，并继续完全控制稀土的加工。八年后，正如大家所知，磁体依赖性与 AI 的爆发发生了碰撞。2022 年，OpenAI 推出了 ChatGPT，正如大家所知，它成为了历史上增长最快的消费技术，并由此引发了当前 AI 基础设施的建设浪潮。AI 的繁荣创造了前所未有的数据中心容量需求，而这正是稀土再次成为焦点的原因，因为 AI 工作负载产生的热量远超传统计算。服务器机架的功耗从 10 千瓦（kW）飙升至 50 千瓦（kW），甚至超过 100 千瓦（kW），在这种情况下，每一瓦特的计算都需要冷却。现代冷却系统，包括高效风扇和液体循环泵，都依赖于钕铁硼磁体，这得益于其尺寸和能源效率。这意味着，在现有需求的基础上，我们还需要大量的磁体，也就意味着对中国的依赖进一步加深。我再次提醒您，这些磁体的 90% 都在中国生产和加工。如果您是美国，并且与中国处于直接对抗之中，您会希望尽早解决这个问题。在 AI 和科技领域，美国需要格陵兰，因为稀土的主导权对美国至关重要。它之所以重要，是因为当前局势赋予了中国对 AI 运行的物理硬件的议价能力。
Original English
It is important to understand the historical context here. In 1985, the United States controlled approximately 60% of global rare earth production and most of it came from Mountain Pass mine in California's Mojave Desert. China, on the other hand, produced about 30%. The United States was the world leader in rare earth mining, processing, and magnet manufacturing. They dominated the entire vertical from ore to finished product. But then the tables have turned. Throughout the '9s and 2000s, China poured a lot of state money into dirty rare earth processing as well as the magnet plants that western companies were happy to shut down and offshore to China. So by the time anyone noticed, nearly the entire supply chain had migrated to China. Starting 2006, for nine years in a row, China weaponized its control, they restricted exports and cut supply by 40% in 2010. And then during a maritime dispute with Japan over a detained fishing boat captain, they halted rare earth shipments to Japan entirely for 2 months. Japanese manufacturer panicked because Japan was 90% dependent on Chinese imports at the time. The Japan embargo was a watershed moment because for the first time the world saw China willing to use rare earth supply as a geopolitical weapon. In turn, Japan aggressively diversified and reduced Chinese dependence from 90% to 60%. And cut total rare earth consumption in half. The 2010 China Japan embargo was a wakeup call for everyone. And in 2014, the World Trade Organization ruled that China's export quotas violated trade rules because China couldn't justify them as conservation or environmental protection. China had to abolish formal quotas, but they retained export licensing and fully controlled processing of rare earths. And 8 years later, as you know, the magnet dependency collided with the AI boom. 2022 OpenAI launches Chad GBT that became as you all know the fastest growing consumer technology in history and that triggered the current AI infrastructure buildout. The AI boom created demand for data center capacity that we have never seen before and that is how the rare earths re-entered the chat because AI workloads generate vastly more heat than traditional computing. Server racks jumped from 10 kows to 50 kowatts, some exceeding 100 kow, where every watt of compute requires cooling. And modern cooling systems, high efficiency fans, liquid circulation pumps rely on neodymium iron boron magnets. And they do so for their size and energy efficiency. Meaning that on top of the existing demand, we're looking for a ton more magnets, which means a ton more dependency on China. And once again, I will remind you that 90% of these magnets are produced and processed in China. And if you're the US and you are in direct confrontation with China, you're going to want to deal with this sooner rather than later. In the context of AI and tech, the US needs Greenland because the rare earth dominance matters for the US. And it matters because the current situation gives China leverage over the physical hardware that AI runs on.AI 爆发下的稀土瓶颈与格陵兰的战略价值
您可能会问，格陵兰在这其中扮演什么角色？格陵兰拥有约 150 万公吨的稀土储量，这足以满足美国相当一部分的需求，并减少对中国的依赖。然而，格陵兰本身无法完全解决这个问题，因为问题的核心并非在于开采。
普遍的叙事将稀土问题归咎于开采环节，但事实并非如此。中国确实生产了全球大部分的稀土矿石，但这却是整个方程式中最重要的部分。中国拥有全球约 70% 的稀土矿业开采权，但其加工能力却占据了约 90% 的市场份额，而“加工”正是关键所在。
格陵兰无法解决此问题的原因有几个。首先，加工环节的模式无法仅凭陆地资源来解决。可以这样理解：美国或欧洲可以设计自己的服务器和数据中心，但如果其冷却风扇或水泵使用的是中国制造的磁体，那么根据定义，它们就暴露在中国出口许可之下。而中国的许可现在已成为默认的制约因素。特朗普的逻辑是：如果我们控制了更多的稀土矿石，我们对中国的依赖就会减少。但问题在于，矿石并非瓶颈。最大的壁垒在于矿石的加工能力。如果美国明天就收购格陵兰，它将无法解决稀土问题的 0%。因为格陵兰的矿藏需要 10 多年才能产出显著的矿石，即使开始开采，这些矿石仍需要中国的加工才能使用，而这正是格陵兰本应解决的依赖问题。
这就是为何瓶颈在于加工能力。加工是一项非常专业化的化学工程专长，例如稀土分离化学，这在全球可能只有十几所大学教授。它涉及危险的酸性处理、放射性废物处理以及有毒废物处理。每生产一吨稀土，就会产生约 2000 吨有毒废料。
当然，为了公平起见，如果格陵兰仅仅是提供矿石，美国也不会对其如此感兴趣。那么，美国为何仍想得到它呢？大量的土地资源。这为数据密集型科技和数据中心提供了广阔的物理空间，并能带来免费冷却的优势。此外，还有雷达站、卫星链路、地面站、空间态势感知、早期预警和导弹防御系统。随着太空竞争日益激烈，永久控制关键地面站点变得愈发重要。但悖论在于，公司已经可以在格陵兰进行建设和租赁。拥有该岛的土地和国家主权，只会改变华盛顿的政治风险。因此，在美国的这场博弈中，这更多地关乎西方科技和美国国防的“神经系统”，而非技术本身或思维方式。
第二个原因是，即使在极其有利的假设下，时间表也是不可实现的。如果格陵兰在 2026 年被收购或吞并，铀禁令被解除，获得全部资金，环境许可被快速审批，那么商业规模的稀土生产也需要等到 7 到 10 年后。而能够提炼这些矿石的加工能力，则需要在此基础上再增加 3 到 5 年。美国正在努力建设这种能力，但这需要再耗费十年时间，即便如此，其加工量也仅是中国产能的一小部分。与此同时，AI 基础设施的部署是以季度为单位进行的，而且就在当下。微软的“星门”（Stargate）项目正在进行中。OpenAI 公开表示，未来模型需要额外 10 吉瓦（GW）的电力，这大致相当于纽约市一个城市的电网规模，仅供一家公司使用，而且他们现在就需要。到今年年底，数据中心将消耗约 1000 太瓦时（TWh）的电力。格陵兰无法解决一个在 2025 年显现、但生产要到 2023 年才开始的问题。
Original English
You may ask why Greenland? Greenland holds 1.5 million metric tons of rare earth reserves which is enough to supply meaningful share of the US demand and reduce dependence on China. But Greenland alone can't fix this problem because the problem isn't in the mining. The common narrative frames rare earths as the mining problem, but it's not. China produces most of global rare earth ore, but that's the least important part of the equation. They own around 70% of global mining and they process around 90. Processing being the operative word. There are several reasons why Greenland cannot fix this problem. Reason number one, the processing mode is not solvable by land. Think of it this way. the US or Europe can design its own servers, can design their own data centers. But if the cooling fans or pumps use Chinese-made magnets, they're exposed to Chinese export licensing by definition. And Chinese licensing is now the default. Trump's logic goes like this. If we control more rare earth ore, we're less dependent on China. But the problem is that the ore isn't the bottleneck. The biggest moat is processing of the ore. If the United States acquired Greenland tomorrow, it would solve exactly 0% of the rare earth problem. Because Greenland's deposits won't produce significant ore for 10 plus years. And even after mining starts, the ore would still need Chinese processing, the very dependency Greenland was meant to fix. This is why the bottleneck is the processing capacity. Processing being a very specialized chemical engineering expertise. For example, rare earth separation chemistry, which is taught at maybe a dozen universities globally, hazardous acid processing, radioactive waste handling, toxic waste handling. There's about 2,000 tons of toxic waste per ton of rare earths. Now, in the interest of being fair, the US wouldn't be so interested in Greenland if there was nothing more to it than just the ore. So, why does the US want it anyway? A ton of land. It's a ton of physical location for data inensive tech and data centers and subsequently free cooling. On top of it, raiders, satellite links, ground stations, space awareness, early warning and missile defense. As space becomes more contested, permanent control of key ground sites becomes more valuable. But the paradox is that companies can already build and lease in Greenland. the ownership of the land, the ownership of the country would only change the political risk for Washington. So in this equation for the US, this is a lot less about technology or or minds themselves. It's a lot more about the nervous system of the western tech and American defense. Reason number two, the timeline is unachievable even under extremely favorable assumptions. If Greenland gets acquired or annexed in 2026, uranium ban repealed, full funding secured, environmental permits fasttracked, commercial scale rare earth production lies 7 to 10 years into the future, and processing capacity to refine that ore is another 3 to 5 years beyond mining. The US is attempting to build this capacity, but it's going to take another decade and even then they'll process just a fraction of the Chinese capacity. Meanwhile, AI infrastructure deployment happens in quarters and it's happening right now. Microsoft Stargate project is underway now. OpenAI is openly talking about another order of 10 gawatts of power for future models which is roughly a New York Citys sized grid just for one company and they need it now. Data centers will consume around a,000 terowatt hour by the end of the year. Greenland can't solve a problem that manifests in 2025 with production that starts in 2023.结论：战略价值与现实瓶颈
最后，我们来谈谈回收问题。回收问题是房间里的大象，但得到的关注度与格陵兰相比微乎其微。2025 年 4 月，包括微软在内的多家公司试点了一个硬盘回收项目，并宣布成功从约 50,000 磅（约 22.7 吨）的粉碎和退役硬盘驱动器以及从美国各地微软数据中心收集的安装硬件中回收了稀土。该项目回收了约 90% 的稀土材料，回收了约 80% 的进料（包括金、铜、铝和钢），并测试了一种无酸回收工艺，该工艺可将温室气体排放量与传统采矿相比降低约 90%。然而，讽刺的是，美国稀土回收率仅低于 10%。如果这一项目能在全球范围内推广，回收将能为美国带来大量的回收稀土，而无需开采新矿或获取新领土。而且，与采矿相比，该工艺的化学过程非常直接。磁体已经从矿石中分离出来，并已是纯化的金属形式。但我们选择不去思考回收，反而去考虑收购新领土。
那么，从技术角度来看，购买格陵兰是否值得？这取决于。购买该国能否解决磁体瓶颈问题？不能。但它能否提供对格陵兰长期技术优势（包括太空领域）的控制？是的。是否值得为此引发一场地缘政治风暴？不太确定。从技术上看，格陵兰作为一个关键矿产省、一个潜在的绿色 AI 中心以及一个太空节点，具有绝对的价值。但方法至关重要。到 2025 年，中国不仅主导稀土开采，还主导着整个供应链——矿石、加工、化学技术、磁体本身。您无法通过一笔房地产交易来摆脱这种局面。
我们希望这些信息对您有所帮助。请在评论区告诉我们您的看法。下次再见。