Anthropic 把 2000 亿塞回谷歌口袋:AI 时代最体面的左手倒右手

marsbitPublished on 2026-05-06Last updated on 2026-05-06

作者:Ada,深潮 TechFlow

5 月 5 日,据 The Information 报道,Anthropic 承诺未来五年向谷歌云支付 2000 亿美元。

这笔多年期协议从 2027 年开始,将占谷歌云收入积压的 40% 以上,该指标反映的是企业客户的合同承诺。

一家五年前还不存在的 AI 公司,凭一纸合同,吃掉了谷歌云未来收入的近一半。

消息出来当天,Alphabet 盘后涨 2%。

但更值得玩味的是另一个数字。Alphabet 同时反向投资 Anthropic 最高 400 亿美元。

钱从谷歌账上出去,转一圈,又回到谷歌账上。中间多了一行“Anthropic 算力支出”的会计科目。

那么,这究竟是史上最大的云计算订单,还是史上最体面的财务魔术?

一份不止给谷歌的“独家承诺”

要理解这笔交易的本质,先看一组并不孤立的数据。

Anthropic 4 月 20 日宣布与 Amazon 扩大合作,承诺未来 10 年在 AWS 技术上花费超过 1000 亿美元,换取最多 5 吉瓦的算力。作为交换,亚马逊在已有 80 亿美元投资基础上,再追加最高 250 亿美元。

而去年 11 月,微软同意向 Anthropic 投资最高 50 亿美元,Anthropic 则承诺购买 300 亿美元的 Azure 算力。

也就是说,谷歌:投 400 亿 ,收 2000 亿。亚马逊:投 330 亿 ,收 1000 亿+。微软:投 50 亿 ,收 300 亿。

三家云巨头加起来掏出约 780 亿美元,换回 3300 亿美元的“合同承诺”,账面净流入 2500 亿。

这套玩法的本质,就是把资本支出洗成营收。投资 Anthropic 的钱计入投资活动现金流,Anthropic 付的算力费计入主营收入。同一笔钱,左口袋出右口袋进,财报上多了一笔漂亮的 backlog。

Alphabet 一边给 Anthropic 输血,一边把 Anthropic 的算力采购计入未来收入,AI 基建繁荣由此呈现出一种自我强化的闭环。

华尔街是这场游戏里的真正赢家,只要 backlog 数字够大,市盈率就能撑得住。

飞轮的高级版本

Strategy 高位加仓的故事还没讲完,AI 圈就把同一套飞轮放大了一千倍。

Strategy 的逻辑是,发股票筹钱,买比特币,币涨推高市值,发更多股票,买更多币。

而云厂商的逻辑是,投资 AI 公司,AI 公司付费买算力,营收增长,股价上涨,资本市场加注,继续投资 AI 公司。

不同的是,比特币是稀缺资产,每一枚都对应链上的真实供给。算力不是。2027 年才会上线的“多吉瓦级 TPU 产能”,今天甚至连机柜都没装。

也就是说,2000 亿美元里有相当一部分,是 Anthropic 提前承诺购买一批尚未制造的芯片,谷歌则用这份承诺去说服资本市场。

这不就是远期合约吗?区别在于商品期货有交割日和保证金,但这份合同没有。Anthropic 真到 2027 年付不出这笔钱怎么办,违约成本由谁承担?

不会是谷歌。它已经把 backlog 写进了财报电话会的 PPT。Alphabet 在 4 月 29 日的财报电话会上披露,谷歌云收入同比增长 63%,超过 200 亿美元,云业务积压达到约 4620 亿美元。这个数字撑起了 Alphabet 现在的市值。

也不会是 Anthropic。它只要继续融资就行,反正下一轮估值还在涨。

最后埋单的,可能还是以为自己买的是"AI 卖铲人"故事的散户。

50 亿撬动 3300 亿

Anthropic 自己的体量配得上这个数字吗?

据媒体报道,Anthropic 的年化收入在 2025 年从 10 亿美元涨到了 50 亿美元。

一家年化收入仅 50 亿的公司,签了一份 5 年 2000 亿、10 年 1000 亿、外加 300 亿美元的合同,三份合同加总 3300 亿。

哪怕 Anthropic 收入再翻 10 倍,5 年累计也撑不到 3300 亿。

那么,钱从哪来?

只有一条路,继续融资。

而最大的潜在投资人,恰恰是这三家云厂商本身。

这就是循环的全部秘密。Anthropic 不需要真的赚到钱,它只需要保持“一直在融资”的状态,把每一轮新融来的钱当作下一年的算力账单。融资估值上去了,又能融到更多。

听上去像谁?

Strategy. 它也不需要比特币真的产生现金流,只需要保持“一直能发股发债”的状态。唯一的区别是,Strategy 的资产负债表上还有比特币这种全球公开定价的资产。

AI 公司的估值逻辑,已经和 很像 2021 年的 SaaS 公司了。那时候大家拼 ARR,今天拼的是算力承诺。本质都是用未来贴现现在,唯一的问题是,未来会不会兑现。"

OpenAI 在做什么

就在 Amazon 给 Anthropic 加注的同一份 8-K 文件里,OpenAI 也承诺通过 AWS 基础设施消耗约 2 吉瓦 Trainium 算力,从 2027 年开始爬坡。

两个月前,亚马逊向 OpenAI 投资 500 亿美元,并签下 1000 亿美元的云计算合同。

剧本完全一样。

也就是说,三大云厂、两大模型公司,五个玩家把同一个游戏玩了好几遍。每一次都伴随"史上最大""战略合作""算力革命"的标题。

每一次背后都是同一笔钱在打转。

那么,谁会先停下来?

不会是云厂商,它们现在的市值靠的就是这套叙事。Alphabet 把 2026 年的资本支出指引上调到最多 1900 亿美元,这种规模的支出必须有 Anthropic、有 OpenAI 来"对冲"成营收,否则华尔街第一个不答应。

也不会是模型公司,停下来意味着拿不到下一轮融资,也意味着死亡。

最先被踢出局的,可能会是那些没站好队的二线玩家。

音乐会停吗

这一切的脆弱性,藏在“兑现”这两个字里。

2027 年 TPU 上线。如果届时 Claude 的商业化没能跟上算力扩张的速度,Anthropic 用什么去消化这 2000 亿?

如果一笔合同被重谈、被砍单、被分摊,谷歌云那张 4620 亿美元的 backlog 表立刻露馅。

但今天,没人愿意第一个戳破。CFO 们在写指引,分析师在写买入评级,CEO 们在财报电话会上字斟句酌。所有人都在赌,在音乐停下之前,自己已经站到了离椅子最近的位置。

现在不是泡沫不泡沫的问题,是泡沫怎么拆的问题。所有人都知道这是循环交易,但所有人也都知道,只要 AI 这个故事还在继续,就没人敢做空 backlog。

合同写在纸上,钱在三家公司之间之间流转,估值在一级二级之间打转。每个人都拿了一份“未来的承诺”,每个人都把这份承诺当作“现在的资产”。

直到未来的某一天,某家公司的财报跟不上预期。那一刻,2000 亿美元会突然有了另一个名字,或有负债。

而在这一天到来之前,狂欢会继续。

Related Reads

The Shutdown of Claude Mythos Revealed the True Cost of Renting AI to Me

The sudden shutdown of Claude Mythos this week starkly highlights a critical, often overlooked risk for founders: when your core capability relies entirely on someone else's platform, your fate is not in your own hands. The key question becomes: who truly owns the intelligence your product depends on? For years, the debate around open-source models focused on cost. Now, the evidence is clear: fine-tuned open-source models can achieve frontier-level quality for specific, mission-critical tasks at a fraction of the cost. However, the deeper issue is control. Relying on a third-party API is like renting; it works until the landlord changes the rules, raises the rent, or asks you to leave—as Mythos experienced. The lesson is not to stop using frontier models—they are incredible infrastructure. The goal is ownership. Ownership means starting with a powerful open-source model and shaping it around what makes your company unique: your data, workflows, domain expertise, and definition of "good." Over time, the model becomes less generic and more reflective of your business, creating durable value. The optimistic conclusion is that AI's future doesn't hinge on one superior model. There is no single frontier. The frontier includes proprietary models, models fine-tuned on company-specific knowledge, specialized models for narrow problems, and intelligent routers orchestrating model ensembles. The most interesting development is not models getting smarter, but intelligence becoming increasingly customizable. The winning companies will be those that transform intelligence into a unique, owned asset. Looking ahead, the vision is not one model dominating all, but many teams owning the part of the frontier that matters most to them.

marsbit1m ago

The Shutdown of Claude Mythos Revealed the True Cost of Renting AI to Me

marsbit1m ago

Tiger Research: U.S. Strategic Bitcoin Reserve - Should the Market Be Happy or Disappointed?

Tiger Research analyzes the evolution of U.S. legislative efforts regarding a strategic Bitcoin reserve, concluding the market impact is limited in the short term but potentially positive long-term. The core event was a March 2025 executive order by former President Trump, which designated confiscated Bitcoin as a strategic reserve and promised not to sell existing holdings (approx. 190k BTC). As it contained no mandate to purchase new Bitcoin, the market reacted negatively, with prices dropping 5.7%. Legislative history shows a significant retreat from initial ambitions. The 2024 "BITCOIN Act" proposed mandatory purchases of 1 million BTC over five years. Reintroduced in 2025, it stalled due to high fiscal costs, concerns over dollar hegemony, and opposition from the Treasury Secretary. The current frontrunner, the 2026 "American Retirement and Monetary Advancement (ARMA) Act," is a compromise. It lacks any purchase requirement, instead focusing on consolidating existing government-held Bitcoin and legally prohibiting its sale for at least 20 years. While ARMA has higher passage odds due to bipartisan support and no purchase mandate, its immediate market effect is neutral. It eliminates potential government selling pressure but creates no new demand. The long-term significance is that formally establishing Bitcoin as a national reserve asset in law could later reignite debates on mandatory purchases. Therefore, the path to a government buyer is longer than initially priced by the market, but the directional narrative remains intact.

marsbit4m ago

Tiger Research: U.S. Strategic Bitcoin Reserve - Should the Market Be Happy or Disappointed?

marsbit4m ago

US Stock Market Trend (June 16): SpaceX Rises 42% in Two Days, New Fed Chairman Takes Office Today

**U.S. Stocks Trend (June 16): SpaceX Soars 42% in Two Days, New Fed Chair Takes Office Today** Markets surged on Monday following former President Trump's social media announcement of a completed U.S.-Iran deal to reopen the Strait of Hormuz, pending a June 19 signing. The news triggered a broad risk-on rally: oil prices crashed, tech stocks soared, bond yields fell, and defensive sectors lagged. **Market Performance:** The Nasdaq jumped 3.07%, led by semiconductor stocks like Micron (+9.2%). The S&P 500 gained 1.65%, and the Dow rose 0.92% to a record high. However, the Russell 2000 small-cap index underperformed (+0.72%). SpaceX continued its hot streak, rising another 5% pre-market after disclosures of large buys by an Australian billionaire and Cathie Wood's ARK. Boeing also rallied on the transportation optimism. Conversely, energy stocks like Chevron fell over 3% on the oil price plunge, with other defensive sectors also selling off. The day's action showed a clear rotation of funds from energy/defensive plays into AI and tech narratives. **Macro & Outlook:** The VIX fear index fell 8.37%. Treasury yields declined, and WTI crude dropped over 5%. Attention now shifts to a packed schedule: the Bank of Japan is widely expected to hike rates to 1.0% on Tuesday. The Fed's June meeting concludes Wednesday, marking new Chair Wash's debut. While rates are expected to hold, his tone on stubborn inflation and the "dot plot" will be crucial for gauging the 2024 rate path. The formal Iran deal signing is set for Friday. **Trend Perspective:** While the peace deal is a genuine positive, Monday's explosive rally may have gotten ahead of itself, pricing in a swift resolution to inflation concerns. The shortened trading week faces a triple test: BoJ tightening, the Fed's policy stance, and deal implementation details. Tech and semiconductors, which led the surge, remain vulnerable to any disappointment from these key events. The real price discovery begins with the central banks' communications this week.

marsbit25m ago

US Stock Market Trend (June 16): SpaceX Rises 42% in Two Days, New Fed Chairman Takes Office Today

marsbit25m ago

Xiaohongshu's Second Great Voyage, This Time Sailing Towards AI

Xiaohongshu's Second Voyage: Navigating Towards AI Since ChatGPT's emergence, Xiaohongshu's founder Mao Wenchao has been acutely aware of AI's potential threat, recognizing that the life advice people seek from chatbots overlaps directly with his platform's core business. Founded in 2013 as a PDF shopping guide for Chinese tourists, Xiaohongshu evolved into a massive community where millions share authentic, personal experiences—from product reviews to travel tips. This vast repository of "I've tried this" human judgment became its most valuable asset. However, the rise of AI, which delivers instant answers, challenges the very need for users to sift through numerous personal notes. Fearing its treasure trove of lived experience could become mere training data for others, Xiaohongshu is proactively adapting. In 2026, it established a dedicated AI division (Dots), launched RED Skill to turn user experiences into usable AI tools, and acquired the AI search product "Diandian." Its investments now extend to AI firms like MiniMax and hardware startups, moving upstream to address needs before they even become search queries. The platform's commercialization strategy is also evolving. With a newly acquired payment license and tools like the AIPS model to track consumer decision journeys, Xiaohongshu aims to seamlessly integrate recommendations with transactions, embedding commerce within AI-generated answers. Yet, a critical tension remains. While building smarter machines to organize and leverage its human experiences, Xiaohongshu must prevent AI from drowning out the authentic, flawed, and trustworthy "I've tried this" voices that built its community. Its core challenge is to harness AI's power without letting the map—the machine's perfect, synthesized answer—replace the territory of genuine human experience. This balance between technological advancement and preserving human trust defines its current journey and its future.

marsbit58m ago

Xiaohongshu's Second Great Voyage, This Time Sailing Towards AI

marsbit58m ago

Trading

Spot
Futures

Hot Articles

Discussions

Welcome to the HTX Community. Here, you can stay informed about the latest platform developments and gain access to professional market insights. Users' opinions on the price of AI (AI) are presented below.

活动图片