AI资本支出的博弈论：云巨头为何陷入囚徒困境

The False Dichotomy of AI Optimism vs. Pessimism (AI乐观与悲观的错误二分法)

"Will AI change the world" and "Are CapEx levels too high" are different questions.

"AI是否会改变世界"和"资本支出水平是否过高"是两个不同的问题。

Imagine you knew for certain that AI was going to be as transformational as the internet, and that you control the only AI company in the world. How fast would you build CapEx?

假设你确切知道AI将像互联网一样具有变革性，并且你控制着世界上唯一的AI公司。你会以多快的速度进行资本支出？

I believe the answer is: You would take your time. "AI CapEx" is a euphemism for building physical data centers with land, power, steel and industrial capacity. If you were the only company in AI, you'd wait to digest some AI revenues. You'd see how liquid cooling systems perform, and alter your data center designs as needed. You'd build new power generation assets in the right locations, and then build your data centers in proximity to fiber optic cables. What you would not do is immediately lock in multiple years worth of CapEx, because you'd know that as models and architectures shift, so too will your data centers need to evolve.

我认为答案是：你会从容不迫。"AI资本支出"是建设需要土地、电力、钢材和工业能力的实体数据中心的委婉说法。如果你是AI领域的唯一公司，你会等待消化一些AI收入。你会观察液冷系统的性能，并根据需要调整数据中心设计。你会在合适的位置建设新的发电资产，然后在靠近光纤电缆的地方建设数据中心。你不会立即锁定多年的资本支出，因为你知道随着模型和架构的变化，你的数据中心也需要进化。

Many market participants today would have you believe that there is a choice between being an "AI bull," who believes that infrastructure building is justified by AI's enormous potential or an "AI bear," who believes that overbuilding sets future expectations too high. The thought experiment above illustrates that this is a false dichotomy. The CapEx debate is a debate about speed, not about magnitude.

如今许多市场参与者会让你相信，要么成为"AI多头"（认为基础设施建设因AI的巨大潜力而合理），要么成为"AI空头"（认为过度建设会设定过高的未来预期）。上述思想实验表明这是一个错误的二分法。资本支出争论是关于速度的争论，而不是关于规模的争论。

The Infrastructure vs. Model Progress Dilemma (基础设施与模型进展的困境)

In fact, the more you believe in AI, the more you might be concerned that AI model progress will outpace physical infrastructure, leaving the latter outdated. For example, once everyone has 100k clusters, big tech companies will need to figure out what to do with their 50k and 25k clusters. We've heard a few industry experts make comments along the lines of: No one will ever train a frontier model on the same data center twice—by the time the model has been trained, the GPUs will have become outdated, and frontier cluster sizes will have grown. There is also the issue of how much power you need for a given real estate footprint and how dense to pack your GPU racks, decisions that are dependent on GPU power efficiency—a moving target.

事实上，你越相信AI，你可能越担心AI模型进展会超过物理基础设施，使后者过时。例如，一旦每个人都拥有10万个集群，大型科技公司将需要弄清楚如何处理他们的5万和2.5万个集群。我们听到一些行业专家评论说：没有人会在同一个数据中心训练两次前沿模型——当模型训练完成时，GPU已经过时，前沿集群规模已经增长。还有给定房地产占地面积需要多少电力以及GPU机架密度的问题，这些决策取决于GPU能效——一个移动的目标。

The Competitive Dynamics of Cloud Oligopoly (云寡头垄断的竞争动态)

The key to understanding the pace of today's infrastructure buildout is to recognize that while AI optimism is certainly a driver of AI CapEx, it is not the only one. The cloud players exist in a ruthless oligopoly with intense competition. This is no small prize to defend—the cloud business today is a $250B market, roughly the same size as the entire SaaS sector, combined. The cloud giants see AI as both a threat and an opportunity and do not have the luxury to wait and see how the technology evolves. They must act now.

理解当今基础设施建设速度的关键是认识到，虽然AI乐观主义确实是AI资本支出的驱动因素，但不是唯一的驱动因素。云服务参与者存在于竞争激烈的残酷寡头垄断中。这不是一个小奖赏需要捍卫——今天的云业务是一个2500亿美元的市场，大约相当于整个SaaS行业的总和。云巨头将AI视为威胁和机遇，没有等待技术发展的奢侈。他们必须立即行动。

The arms race between Microsoft, Amazon and Google is thus game theoretic. Every time Microsoft escalates, Amazon is motivated to escalate to keep up. And vice versa. We are now in a cycle of competitive escalation between three of the biggest companies in the history of the world, collectively worth more than $7T. At each cycle of the escalation, there is an easy justification—we have plenty of money to afford this. With more commitment comes more confidence, and this loop becomes self-reinforcing. Supply constraints turbocharge this dynamic: If you don't acquire land, power and labor now, someone else will.

因此，微软、亚马逊和谷歌之间的军备竞赛是博弈论的。每次微软升级，亚马逊就有动力升级以跟上。反之亦然。我们现在正处于世界上三家最大公司之间的竞争升级循环中，总价值超过7万亿美元。在每次升级循环中，都有一个简单的理由——我们有足够的钱来承担这个。随着更多承诺而来的是更多信心，这个循环变得自我强化。供应限制加剧了这种动态：如果你现在不获取土地、电力和劳动力，别人就会。

The Desperation of Smaller Players (较小参与者的绝望)

For smaller players, the urgency is even higher. If Microsoft and Amazon buy up all the land and power, buy up all the diesel generators and buy up all the liquid cooling systems, then how will you compete? When you look one notch below the scale of Amazon, Google and Microsoft, there is a sense of desperation. If you do not move now, you will never get another chance.

对于较小的参与者来说，紧迫性甚至更高。如果微软和亚马逊买下所有土地和电力，买下所有柴油发电机和液冷系统，那么你将如何竞争？当你看到亚马逊、谷歌和微软规模下一级的公司时，有一种绝望感。如果你现在不行动，你将永远不会再有机会。

This helps explain another potential motive for the aggressive behavior we're seeing from the cloud providers: Defense. There is a real-sense in which only companies that have the balance sheets to withstand big write-offs can now afford to play in the AI infrastructure race. From this perspective, overbuilding may be perfectly rational.

这有助于解释我们看到云提供商激进行为的另一个潜在动机：防御。在某种意义上，只有那些拥有能够承受大额减记的资产负债表的公司现在才能负担得起参与AI基础设施竞赛。从这个角度来看，过度建设可能是完全合理的。

Positive Externalities for the AI Ecosystem (对AI生态系统的正外部性)

Whether for reasons of optimism or reasons of competition, today's rapid construction of new AI data centers should have a big positive effect on startups going forward. Much of the risk in AI today is being borne by infrastructure providers. This is effectively a subsidy for the startups who are building on top of them. Competition between Microsoft, Amazon and Google should assure lower API pricing in the future. It's also good for the AI ecosystem—these CapEx investments will enable us to test scaling laws and learn more about AI's future potential.

无论是出于乐观还是竞争的原因，今天快速建设新的AI数据中心应该对未来初创企业产生巨大的积极影响。今天AI的大部分风险由基础设施提供商承担。这实际上是对在其基础上构建的初创企业的补贴。微软、亚马逊和谷歌之间的竞争应该确保未来更低的API定价。这对AI生态系统也有好处——这些资本支出投资将使我们能够测试扩展定律并更多地了解AI的未来潜力。

Like governments did in the past, big tech companies are making high upfront infrastructure investments that will spur innovation. Whether or not these investments end up being profitable before they depreciate, they are on the critical path to AI's long-term impact.

就像政府过去所做的那样，大型科技公司正在进行高额的前期基础设施投资，这将刺激创新。无论这些投资在贬值前是否最终盈利，它们都在AI长期影响的关键路径上。

Frequently Asked Questions (常见问题)

什么是AI资本支出博弈论？

AI资本支出博弈论分析云服务巨头（微软、亚马逊、谷歌）在AI基础设施投资中的战略互动，解释为何竞争压力而非单纯技术乐观驱动当前投资热潮。
为什么科技巨头加速AI基础设施投资？

主要驱动因素是寡头竞争博弈：任何一方的投资升级都会迫使其他方跟进，形成"囚徒困境"。根据行业报告，2500亿美元的云市场价值使防御性投资成为必要策略。
AI模型进展如何影响基础设施投资？

AI模型迭代速度可能超过硬件生命周期，导致数据中心快速过时。行业专家指出，前沿模型很少在同一数据中心训练两次，因为训练完成后GPU可能已落后。
较小公司如何参与AI基础设施竞争？

较小公司面临资源约束，但可通过专业化服务、边缘计算或与巨头合作参与。当前投资热潮为初创企业提供了基础设施补贴，降低了进入门槛。
AI基础设施投资对生态系统有何影响？

竞争将降低API服务价格，使更多企业能够使用AI能力。基础设施投资还支持扩展定律测试，加速AI技术突破，类似政府主导的基础设施建设对创新的促进作用。