Wall Street quantitative giant Jane Street plans to build its own data center 华尔街量化巨头简街计划自建数据中心

## Summary
Data is capital, and computing power is power. When Wall Street’s quantitative giants begin building their own data centers, the signal could not be clearer: in this battle fought in milliseconds, the ultimate control over computing power and data latency has shifted from being a supporting tool to a core battlefield.

## Deep Analysis
Data is capital, and computing power is power. When Wall Street’s quantitative giants begin building their own data centers, the signal could not be clearer: in this battle fought in milliseconds, the ultimate control over computing power and data latency has shifted from being a supporting tool to a core battlefield.

Jane Street is planning to build its own data center and seeking financing, in discussions with tech, crypto, and finance companies. The most intriguing aspect of this news lies precisely in its “early-stage” nature—specific capacity and location are yet to be determined. This is less a deliberate announcement and more a cautious “feasibility check” and display of strength. It implies that for top algorithmic traders, the standardized, shareable computing power and bandwidth provided by third-party cloud services or data centers are hitting performance ceilings. When your competitive strategy relies on modeling, trading, and risk management in nanoseconds, the physical-world variables—such as who your “neighbors” are (whether other tenants’ traffic interferes), peak power supply guarantees, and the physical distance of fiber-optic cable entry points—begin to erode the excess profits of the virtual world. Building in-house represents a leap from “renting computing power” to “defining the computing environment.” It’s no longer just about how many GPUs to buy, but about constructing a digital fortress tailored to your strategy.

Interestingly, at the same time, we see CITIC Securities’ research reports using extremely complex financial terminology—“capital contraction and siphon effects,” “divergence critical points of yield correlations”—to describe the current challenges and potential paths for China’s A-share market. The report suggests building a new “investment barbell” with “AI + energy and materials.” On one side, cutting-edge Wall Street players are using real capital to erect computing power walls in the physical world; on the other, the A-share market is discussing how to use “AI concepts” as a new narrative label for capital rotation. This contrast starkly reveals the current divide in understanding “AI”: for firms like Jane Street, AI is a “production tool” that must be internalized into every layer of hardware and algorithms; while for many market participants, AI is more of a “narrative tool” to carry capital, explain volatility, and tell new stories.

Jane Street’s move serves as a kind of disenchantment for all “AI+” stories. It exposes an underlying logic: any industry reliant on cutting-edge AI models for alpha will ultimately evolve into competition over infrastructure. No matter how sophisticated your model is, if data transmission is 0.1 microseconds slower than your competitor’s, or if GPU clusters jitter due to shared resources during volatile market conditions, your advantage evaporates. It’s like two top restaurants vying for three Michelin stars—one meticulously researching molecular gastronomy recipes, while the other is already purchasing exclusive ranches and water sources. The former is competing in “culinary technique,” while the latter is reconstructing the “ingredient supply chain.” When an industry reaches a fever pitch and competition spills over from the software layer to the hardware and infrastructure layers, it often marks a new phase of the game.

The hot-list news surrounding “AI”—the dividend track for 240 million single individuals, automation in chip design verification, the commercial story of Li-Ning and Curry—appears diverse on the surface, but actually highlights the current disconnect between AI’s application layer and infrastructure layer. The application layer is striving to find every capillary where AI can penetrate, trying to prove its commercial value; meanwhile, the most top-tier players truly reaping massive profits from AI are quietly engaging in “civil engineering.” This disconnect also explains why the market feels some fatigue toward AI: too many stories are built on the sandy foundation of “calling APIs” or “fine-tuning models,” while too few investments delve into the deep waters of computing power that truly constitute barriers.

The fact that Jane Street’s financing discussions include cryptocurrency enterprises is also worth pondering. The crypto world, especially in high-performance trading, shares the same demand for low latency and computing power as Wall Street quantitative funds. Their collaboration could give rise to a “hyper-computing power” infrastructure blending the rigor of traditional finance with the aggressiveness of the crypto world. This might blur the boundaries between finance and crypto, between centralized and decentralized trading infrastructure, creating new monsters.

Ultimately, Jane Street’s decision points to a more fundamental dimension of competition: data sovereignty and computing power sovereignty. When the external environment is filled with ambiguity and volatility—as CITIC’s report puts it, “during the blurry period before macro external shoes drop”—true giants choose not to entrust their lifeline to uncertain external supply chains. They opt to firmly hold core productive materials—data and computing power—within the physical entities they build themselves. This is not just a technological choice but a strategic risk hedge.

So, stop merely discussing what new breakthroughs have been made in AI algorithms. Pay attention to those actions that are “paving roads” and “building power plants” for AI. When firms like Jane Street start building their own data centers, they are not just purchasing servers—they are buying the right to define the rules of the game for years to come. For other participants, the question is no longer “Can I use AI well?” but rather “When computing power becomes the new oil, do I own my own oil wells and refineries?” The answer will determine who remains at the table in the next cycle.

数据即资本，算力即权力。当华尔街的量化巨头开始亲自下场修建自己的数据中心时，信号已经清晰得不能再清晰：在这场以毫秒计的战争中，算力和数据延迟的终极控制权，已从辅助工具变成了核心战场。

简街（Jane Street）计划自建数据中心并寻求融资，正与科技、加密、金融企业洽谈。这个消息最耐人寻味之处，恰恰是它的“早期”状态——具体容量、选址皆未定。这不像一次深思熟虑的宣布，更像一次谨慎的“摸底”和实力的展示。它意味着，对于顶级的算法交易者而言，租用第三方云服务或数据中心所提供的标准化、可共享的算力和带宽，正在触及性能的天花板。当你的竞争策略依赖于在纳秒内完成建模、交易和风控，那么“邻居”是谁（其他租户的流量是否干扰）、电力供应的峰值保障、光缆入口的物理距离，这些物理世界的变量，开始反噬虚拟世界的超额利润。自建，是从“租用算力”到“定义算力环境”的跃迁。这不再是关于买多少GPU的问题，而是关于建造一个如何为自己策略“量体裁衣”的数字堡垒。

有趣的是，与此同时，我们看到中信证券的研报用极其复杂的金融术语——“资金缩圈与虹吸”、“收益率相关系数背离临界点”——来描述A股当前的困境与出路。研报建议以“AI+能化”构建新的投资杠铃。一边是华尔街最前沿的玩家用真金白银在物理世界筑起算力高墙，另一边是A股市场在讨论如何用“AI概念”作为资金轮动的新叙事标签。这种对比，辛辣地揭示了当前阶段对“AI”理解的鸿沟：对简街这类公司，AI是必须内化于硬件与算法每一层肌肉的“生产工具”；而对许多市场参与者而言，AI更多是一个可以承载资金、解释波动、讲出新故事的“叙事工具”。

简街的动作，是对所有“AI+”故事的某种祛魅。它暴露了一个底层逻辑：任何依赖尖端AI模型产生阿尔法的行业，最终都会演变为基础设施的竞争。你的模型再精妙，如果数据传输比对手慢0.1微秒，或者GPU集群在行情剧烈波动时因共享资源出现抖动，那么优势将荡然无存。这就像两家顶级餐厅争夺米其林三星，一家在精心研究分子料理配方，而另一家已经开始购买专属的牧场和水源。前者在比拼“烹饪技艺”，后者在重构“食材供应链”。当行业进入白热化，从软件层的竞争溢出到硬件和基础设施层，往往是游戏进入新阶段的标志。

那些围绕“AI”的热榜新闻——2.4亿单身群体的红利赛道、芯片设计验证自动化、李宁与库里的商业故事——表面看琳琅满目，实则凸显了当下AI应用层与基础设施层的割裂。应用层在努力寻找每一个可以被AI渗透的毛细血管，试图证明其商业价值；而最顶尖的、真正从AI中攫取暴利的玩家，却在默默无闻地进行“土木工程”。这种割裂也解释了为何市场会对AI感到一丝疲惫：太多的故事建立在“调用API”或“微调模型”的沙滩上，而太少的投资深入到了真正构成壁垒的算力深水区。

简街的融资洽谈对象包括加密货币企业，这一点也值得玩味。加密世界，尤其是高性能交易领域，对低延迟和算力的需求，与华尔街量化基金同出一源。它们的联手，可能催生出一种混合了传统金融严谨性与加密世界激进性的“超算力”基础设施。这或许会模糊金融与加密、中心化与去中心化交易基础设施的边界，创造出新的怪物。

最终，简街的决策指向一个更本质的竞争维度：数据主权与算力主权。当外部环境充满模糊与波动，正如中信研报所言“宏观外部靴子落地前的模糊期”，真正的巨头选择不把命脉交托于不确定的外部供应链。他们选择将核心的生产资料——数据和算力——牢牢掌握在自己构建的物理实体中。这不仅是技术选择，更是一种战略性的风险对冲。

所以，别再仅仅讨论AI算法又有了什么新突破了。关注一下那些正在为AI“修路”和“建电厂”的行为吧。当简街们开始自建数据中心，他们购买的不是服务器，是未来数年内定义游戏规则的权利。对于其他参与者而言，问题已经不再是“我能否用好AI”，而是“当算力成为新的石油，我是否拥有自己的油井和炼油厂？” 答案将决定谁在下一个周期里，继续留在牌桌上。