• 亿邦公众号
    • APP下载

2400亿 PE最大的一笔放贷出炉

作者:陶辉东 文章来源:投中网

2026-06-29 11:21

作者:陶辉东 文章来源:投中网

2026-06-29 11:21

邦小白快读

EN
全文速览

本文核心披露了PE行业跨界进军AI算力基建融资的标志性大交易,相关干货要点如下

1. 核心交易信息:全球PE巨头阿波罗全球资管、黑石联手敲定了350亿美元(约合2400亿人民币)的私募信贷融资,即大天空计划,支持AI企业Anthropic扩张算力,这是史上规模最大的私募信贷及芯片专项融资交易,项目规划算力容量1GW,整笔债务不进入Anthropic的资产负债表。

2. 交易运作逻辑:PE设立特殊目的载体SPV持有算力中心,PE出劣后资金,剩余部分通过私募信贷募资,采购芯片后租给AI企业,芯片厂商博通提供残值担保增信,PE再将信贷转卖给保险公司、养老金等各类投资者。

3. 行业整体趋势:当前AI算力需求爆发,股权融资和传统银行信贷都无法满足缺口,私募信贷已经成为填补缺口的主力,未来有望形成数万亿美元规模的新资产类别。

本文披露的AI算力产业发展新趋势,对布局AI相关业务的品牌商,有如下干货参考

1. 行业趋势参考:当前全球AI产业进入爆发期,算力是AI业务增长的核心制约,据预测2028年全球AI算力需求可达117GW,对应总投资额超4万亿美元,提前布局锁定算力资源,是品牌商做AI产品研发和用户服务的核心前提。

2. 运作模式参考:头部玩家已经形成资本+产业联盟化运作的模式,AI企业不需要自身重仓大额资产,可以通过绑定PE资本、芯片厂商实现轻资产扩张,这种模式值得布局AI的品牌商借鉴。

3. 风险警示:案例显示Anthropic曾因为算力不足,被迫收紧订阅、给模型降智,直接影响用户体验和品牌口碑,品牌商布局AI业务需要提前做好算力规划,避免因产能缺口损害品牌价值。

本文披露的AI基建融资新动态,给切入AI赛道的卖家带来了机会、风险和模式层面的干货参考

1. 市场机会判断:AI算力基建是当前增量空间极大的增长市场,总规模可达数万亿美元,传统商业银行受巴塞尔协议监管约束,无法提供足够资金,私募信贷成为主力,赛道缺口大机会多,卖家人可依托AI产业链寻找相关机会。

2. 扩张模式参考:轻资产杠杆扩张模式值得学习,卖家做AI相关业务不需要自身承担大额负债、占用现金,可以通过引入PE资本、联合芯片厂商做增信,将算力资产放在表外,实现轻装扩张。

3. 风险提示和融资机会:AI芯片折旧速度快,新一代芯片推出后旧芯片价值可能腰斩,重仓算力资产的卖家需要提前做好增信安排,规避贬值风险;当前高利率环境下传统私募股权投资压力大,PE纷纷布局私募信贷,卖家可以对接这类PE获得融资支持。

本文披露的AI算力基建爆发趋势,给芯片制造、数据中心配套等相关工厂,带来了商业机会和转型层面的干货启示

1. 产品需求增量明确:据预测2028年全球AI新增算力需求达117GW,对应需要数千万颗高端AI芯片,以及大量数据中心基建配套、电力供应相关硬件,相关工厂未来会迎来持续的订单增长,市场空间极大。

2. 商业合作模式可借鉴:本次交易中芯片厂商博通通过提供残值担保,实现“以保代销”,既获得了大额芯片订单,也绑定了核心客户,这种产融结合的模式值得各类硬件生产工厂参考。

3. 数字化转型启示:AI大基建的爆发催生了新的产业分工和融资模式,工厂可以主动对接PE资本,参与AI产业链分工,借助资本力量推进自身数字化升级,打开新的增长空间,抓住AI大基建的红利。

本文披露的AI基建融资行业新动向,给金融服务商、算力运营服务商带来了如下干货参考

1. 行业发展趋势明确:当前AI算力基建融资已经从传统的股权融资全面转向信用扩张,私募信贷成为绝对主力,未来整个市场规模可达数万亿美元,是服务商未来的核心蓝海市场。

2. 客户核心痛点清晰:AI企业扩张算力面临巨大资本缺口,股权融资规模有限无法满足需求,传统商业银行受巴塞尔协议限制,也不敢大规模介入长期、非流动性的芯片资产,AI企业普遍缺钱又不想承担表内负债。

3. 成熟可复制的解决方案已经成型:本次大天空计划形成了可复制的完整方案:由PE设立SPV持有算力资产,芯片厂商提供残值担保增信,发行私募信贷后转售给各类长线投资者,云服务商负责日常运营,既满足AI企业轻资产扩张需求,也控制了资金方的风险。

本文的AI算力融资创新案例,给布局AI相关业务的资本平台、产业平台,带来了运营和风险层面的干货参考

1. 市场需求明确:当前AI产业对算力基建融资的需求十分旺盛,原有股权融资、银行信贷都无法填满巨大缺口,平台可以围绕AI基建融资打造专项业务板块,对接资金方和AI企业,开辟新的增长曲线。

2. 可借鉴的最新运作模式:阿波罗和黑石打造的AI XPV平台,已经跑通了专项平台运作模式,以SPV载体运作单个项目,引入芯片厂商做增信分摊风险,再将信贷拆分卖给各类投资者,可以快速做大规模,该平台目标2028年前完成超7000亿美元投资,模式可复制性强。

3. 风险规避要点:AI芯片折旧速度快,资产贬值风险高,平台运作这类项目必须引入产业核心厂商做残值增信,才能拿到低利率的投资级资金,有效控制整体项目的风险,避免出现大规模亏损。

本文披露了全球PE行业与AI基建融合的最新产业动向,对相关领域研究者有如下干货价值

1. 产业新动向明确:AI大基建爆发后,融资结构发生根本性变化,私募信贷已经取代传统股权融资和银行信贷,成为AI基建融资的主力,目前美国非银与私人信贷发行的AI专项债存量已达1.2万亿美元,是传统银行对数据中心贷款余额的8倍,已经催生了一个数万亿美元规模的全新资产类别。

2. 创新商业模式已经成型:目前市场已经形成“PE提供资本+芯片厂商提供增信+AI企业轻资产租用+PE转售信贷给长线投资者”的全新商业模式,美国市场已经形成三大资本联盟,分别是贝莱德微软牵头的GAIIP、软银OpenAI牵头的星际之门、阿波罗黑石牵头的AI XPV。

3. 值得研究的新问题:PE充当影子基建银行,不受巴塞尔协议资本约束,扩张快风险隐患大,目前中国AI企业和PE行业在算力布局上和美国差距明显,这些都是值得深入研究的新课题。

声明:快读内容全程由AI生成,请注意甄别信息。如您发现问题,请发送邮件至 run@ebrun.com 。

我是 品牌商 卖家 工厂 服务商 平台商 研究者 帮我再读一遍。

Quick Summary

This article discloses a landmark mega-deal where private equity (PE) firms are entering the AI computing infrastructure financing space, with key takeaways as follows:

1. Core deal details: Global PE giants Apollo Global Management and Blackstone have jointly closed $350 billion (≈2.4 trillion RMB) in private credit financing for Project Big Sky, which will support AI firm Anthropic in expanding its computing capacity. This is the largest-ever private credit and chip-specific financing deal, with a planned total computing capacity of 1GW, and the entire debt will not appear on Anthropic's balance sheet.

2. Transaction mechanics: PE firms set up a special purpose vehicle (SPV) to own the computing center, with PE providing the junior tranche and raising the remaining capital via private credit. After procuring chips, the facility is leased to AI firms; chipmaker Broadcom provides residual value guarantees as credit enhancement, and PE then sells the credit on to investors including insurance companies and pension funds.

3. Broader industry trends: Amid the current explosive growth in demand for AI computing, equity financing and traditional bank lending have failed to fill the gap. Private credit has emerged as the main source of capital to cover the shortfall, and is expected to grow into a new trillion-dollar asset class in the coming years.

This article outlines emerging trends in the AI computing industry that offer key insights for brands planning AI-related initiatives:

1. Industry trend takeaway: The global AI industry is currently in a period of explosive growth, and computing capacity is the core bottleneck for AI business expansion. Global AI computing demand is projected to reach 117GW by 2028, corresponding to total investment exceeding $4 trillion. Securing long-term access to computing capacity in advance is a core prerequisite for brands developing AI products and delivering user services.

2. Operational model insight: Leading players have adopted a capital-industry alliance operating model, allowing AI companies to expand without holding large amounts of heavy assets on their own balance sheet by partnering with PE capital and chipmakers for asset-light expansion. This model is well worth adopting for brands entering the AI space.

3. Risk warning: As the case shows, Anthropic was once forced to restrict subscriptions and downgrade model performance due to insufficient computing capacity, which directly hurt user experience and brand reputation. Brands rolling out AI businesses should plan their computing capacity requirements far in advance to avoid brand value damage from capacity shortages.

New developments in AI infrastructure financing revealed in this article offer key insights on opportunities, risks and operational models for sellers entering the AI track:

1. Market opportunity assessment: AI computing infrastructure is currently a high-growth market with enormous incremental room, with a total projected size of tens of trillions of dollars. Constrained by Basel Accord regulations, traditional commercial banks cannot provide sufficient capital, leaving large gaps and abundant opportunities in the sector that sellers can tap by positioning along the AI industry chain.

2. Expansion model insight: The asset-light, leveraged expansion model is well worth learning. To grow AI-related businesses, sellers do not need to take on large debts or tie up their own cash: they can bring in PE capital and partner with chipmakers for credit enhancement, keep computing assets off-balance sheet, and expand with a light footprint.

3. Risk warning and financing opportunities: AI chips depreciate rapidly, with older generation chips often losing half their value once next-generation models launch. Sellers holding large computing assets should arrange credit enhancement in advance to hedge depreciation risk. In the current high-interest rate environment, traditional private equity faces mounting pressure, so more PE firms are shifting into private credit—sellers can partner with these PE players to access financing support.

The explosive growth of AI computing infrastructure revealed in this article brings clear commercial opportunities and transformation insights for chip makers, data center infrastructure suppliers and other related manufacturing players:

1. Clear incremental product demand: Global incremental AI computing demand is projected to hit 117GW by 2028, corresponding to demand for tens of millions of high-end AI chips, plus large amounts of data center infrastructure, power supply hardware and other supporting facilities. Relevant manufacturers will see sustained order growth and enormous market opportunities in the coming years.

2. Replicable commercial cooperation model: In this deal, chipmaker Broadcom used a residual value guarantee to enable "sales via guarantee", securing large chip orders while locking in long-term core customer relationships. This industry-finance integrated model is a strong reference for all hardware manufacturing players.

3. Digital transformation insight: The AI infrastructure boom has created new industry divisions of labor and financing models. Manufacturers can proactively partner with PE capital to participate in AI industry chain分工, leverage capital to advance their own digital upgrading, unlock new growth avenues and capture dividends from the AI infrastructure boom.

New industry developments in AI infrastructure financing disclosed in this article offer key insights for financial service providers and computing operation service providers:

1. Clear industry growth trend: AI infrastructure financing has shifted fully from traditional equity financing to credit expansion, with private credit now the undisputed dominant source of capital. The overall market will reach tens of trillions of dollars in size, making it a core growth market for service providers going forward.

2. Clear understanding of core customer pain points: AI firms face massive capital gaps to expand computing capacity; equity financing can only meet a small fraction of demand, while traditional commercial banks, constrained by Basel Accord rules, are unwilling to make large-scale investments in long-term, illiquid chip assets. Most AI firms need capital but do not want to carry on-balance sheet liabilities.

3. A mature, replicable solution already exists: Project Big Sky has produced a complete, replicable framework: PE firms establish an SPV to hold computing assets, chipmakers provide residual value guarantee credit enhancement, private credit is issued and sold to long-term investors, and cloud providers handle day-to-day operations. This framework meets AI firms' demand for asset-light expansion while controlling risk for capital providers.

This innovative AI computing financing case offers key operational and risk management insights for capital platforms and industry platforms building out AI-related businesses:

1. Clear market demand: Demand for AI infrastructure financing is extremely strong right now, and existing equity financing and bank lending cannot close the large gap. Platforms can build dedicated business segments focused on AI infrastructure financing, connecting capital providers with AI firms, and unlock new growth curves.

2. A replicable cutting-edge operating model: The AI XPV platform built by Apollo and Blackstone has already validated the dedicated platform operating model: it structures individual projects in SPVs, brings in chipmakers to provide credit enhancement and share risk, then splits and sells credit to a wide range of investors, allowing it to scale rapidly. The platform targets more than $700 billion in investment by 2028, and the model is highly replicable.

3. Key risk mitigation rules: AI chips depreciate quickly and carry high asset depreciation risk. For platforms running projects of this type, bringing in core industry players to provide residual value credit enhancement is mandatory to access low-cost investment-grade capital, effectively control overall project risk and avoid large-scale losses.

This article discloses the latest industry developments in the integration of the global PE industry and AI infrastructure, offering the following key value for researchers in related fields:

1. Clear new industry dynamics: Following the AI infrastructure boom, the financing structure has fundamentally changed: private credit has replaced traditional equity financing and bank lending as the dominant source of capital for AI infrastructure. Currently, the outstanding stock of AI-specific debt issued by non-bank lenders and private credit in the U.S. has reached $1.2 trillion, 8 times the outstanding balance of traditional bank loans for data centers, and has already spawned an entirely new asset class worth tens of trillions of dollars.

2. An innovative business model is now fully formed: The market has developed a new "PE provides capital + chipmakers provide credit enhancement + AI firms lease assets on a light-asset basis + PE resells credit to long-term investors" business model. Three major capital alliances have already formed in the U.S. market: GAIIP led by BlackRock and Microsoft, Stargate led by SoftBank and OpenAI, and AI XPV led by Apollo and Blackstone.

3. New research avenues worth exploring: PE firms are acting as shadow infrastructure banks, not bound by Basel Accord capital requirements, leading to rapid expansion but high latent risk. Currently, Chinese AI firms and the domestic PE industry lag significantly behind the U.S. in computing capacity布局. All of these are new topics that warrant in-depth research.

Disclaimer: The "Quick Summary" content is entirely generated by AI. Please exercise discretion when interpreting the information. For issues or corrections, please email run@ebrun.com .

I am a Brand Seller Factory Service Provider Marketplace Seller Researcher Read it again.

为PE行业带来一个数万亿美元规模的超大新资产类别。

作者 陶辉东

PE机构的放贷,已成为AI算力扩张的主要输血来源。

近日,阿波罗全球资管和黑石两家PE巨头,联手敲定了一笔高达350亿美元(约2400亿人民币)的私募信贷融资协议,用以支持Anthropic的算力扩张。这笔被内部命名为“大天空计划”(Project Big Sky)的交易,创下了有史以来规模最大的私募信贷及芯片专项融资交易纪录。

这笔交易的大背景是,当下各大AI巨头动辄抛出数百、乃至数千亿美元的算力基础设施建设计划,股权融资根本无力支撑,而商业银行由于巴塞尔协议的限制能提供的资金也有限。

于是,私募信贷成为了填补这一巨大的资本缺口的主力军。轰轰烈烈的AI大基建,或将为PE行业带来一个数万亿美元规模的超大新资产类别。

杠杆之上的算力

孟菲斯郊外,一座被SpaceX改造过的数据中心里躺着22万张英伟达GPU,运行容量高达300兆瓦,消耗的电力足够照亮半个孟菲斯,而它们全部为Anthropic一家运转。

这座代号Colossus 1的算力堡垒,每个月光电费和硬件租赁就要花掉Anthropic 12.5亿美元,年化150亿美元。这大概占Anthropic目前470亿年化营收的三分之一。听起来不少,但还远不够。

Anthropic的营收正在以每个季度翻倍的速度增长,而算力是营收增长的最大制约。因为算力不足,Anthropic不得不施加严厉的速率限制,收紧订阅计划,乃至偷偷地给模型“降智”。Anthropic需要更多算力,而这个任务最终落在了PE机构的头上。

据报道,“大天空计划”规划的算力容量高达1GW,相当于三个Colossus 1。更妙的是,如此庞大的算力中心建设项目,既不占用Anthropic的现金,也不增加它的负债。要知道,Anthropic未来几年的算力扩张目标超过10个GW,它必须找到一种巧妙的杠杆来让自己轻装上阵。

这是怎么实现的呢?“大天空计划”有五个直接参与方,其中,阿波罗全球资管和黑石两家PE巨头是资金提供者,Anthropic是客户,芯片巨头博通是担保方,云服务商Fluidstack负责算力中心的日常运营。谷歌作为算力芯片的供应商,也算是间接参与方。

具体操作是这样的:阿波罗旗下的Atlas SP Partners设立一个了SPV,作为算力中心的持有者。Atlas SP Partners自己在SPV中出资8亿美元普通股作为劣后资金,剩下的342亿美元来自私募信贷。SPV拿着这350亿美元向谷歌采购定制TPU芯片,然后租给Anthropic。因此整笔债务不进Anthropic的资产负债表。

但华尔街也不是做慈善的。要知道,芯片是高科技资产中折旧最快的品类之一,新一代芯片出来,上一代的价值可能腰斩。常规情况下,没有投资者会愿意给芯片放350亿的长期贷款。这笔交易能做成,关键在于博通跳出来当了“担保人”。

在342亿美元私募信贷中,有300亿美元是优先级。博通为这部分债务签了一份“残值担保协议”:如果Anthropic违约、SPV被迫折价处置TPU芯片,如果卖的钱不够还债,博通将补足差额。由于博通的增信,这300亿美元的债务拿到了投资级的评级,整体利率被大幅压低(仅比国债利率高1个点)。

博通之所以愿意做这个担保人,是因为它是这款定制TPU芯片的联合开发者和制造者,并且它还为TPU算力集群提供网络通信芯片与物理交换方案。总之,对博通来说这是“以保代销”的循环金融逻辑。

当然,交易中还一群看不见的参与者。阿波罗和黑石会把这342亿美元的私募信贷卖给广大投资者,包括保险公司、商业银行、养老金计划乃至高净值个人等。它们才是这次AI军备竞赛真正的背后金主。

PE机构已主导算力大扩招

Anthropic这笔350亿交易不是孤立事件,它是AI基础设施融资从“股权换算力”全面转向“信用扩张”的标志性案例。

目前美国市场上,各大科技巨头、投资机构围绕着AI基建已隐隐形成了三大资本联盟。

第一个是贝莱德、微软和MGX牵头的全球AI基础设施投资伙伴关系(GAIIP),计划撬动1000亿美元,专注美国本土数据中心和配套能源建设,英伟达和xAI提供技术支持。

第二个是OpenAI、软银、甲骨文的“星际之门”项目,四年目标5000亿美元,股权承诺约520亿,剩下4480亿则靠债务填。

第三个就是阿波罗、黑石和博通联手打造的AI XPV平台。实际上,Anthropic这笔350亿美元的交易只是首期启动项目,该平台的目标是2028年前在全球建设超过20吉瓦(GW)的算力基础设施,这意味着总投资额超过7000亿美元。

这些数字一个比一个夸张,但核心逻辑是一样的:AI公司自己扛不动算力债务,于是华尔街出马,把它们包装成了一种新资产类别。

而私募信贷,已成为这些债务最主要的买单者。传统商业银行受到巴塞尔协议限制,不敢大量持有这种长期、非流动性的芯片资产。于是私募信贷俨然成了“影子基建银行”。摩根大通的报告显示,非银与私人信贷发行的AI专项债存量已达1.2万亿美元,是传统银行对数据中心的贷款余额的8倍。这些钱没有巴塞尔协议的资本约束,卷起来更猛、更快。

在持续的高利率环境之下,传统的私募股权投资压力山大,但PE机构们在“放贷”上找到一片巨大的新“蓝海”,每月都有超大规模的交易。拿黑石来说,上月CoreWeave宣布了一项75亿美元的债务融资,领投方就是黑石。

这场盛宴才刚刚开始。

Anthropic发布的白皮书称,到2028年,整个美国的AI行业需要的新增算力是50GW。这只是美国。兰德公司预测,到2028年全球对AI算力的需求可达117GW。如果按照350亿美元-1GW的比例换算,总投资额超过4万亿美元。

Anthropic和OpenAI的算力都在今年进入了GW时代。相比之下,中国的AI公司们在算力上的差距依然很大。就连市值破万亿的智谱也未能解决算力的问题,coding plan只能限量卖。中国的PE们能否参与这场盛宴?

投中网原创

注:文/陶辉东,文章来源:投中网(公众号ID:China-Venture),本文为作者独立观点,不代表亿邦动力立场。

【版权提示】亿邦动力倡导尊重与保护知识产权。未经许可,任何人不得复制、转载、或以其他方式使用本网站的内容。如发现本站文章存在版权问题,烦请提供版权疑问、身份证明、版权证明、联系方式等发邮件至run@ebrun.com,我们将及时沟通与处理。

已点赞

+1

{{ praise_text }}

好文章,值得被更多人看到

广告

下一篇

跨境电商资源对接

《全球跨境电商服务商资源手册》第二期手册现已发布,触达更多资源,了解跨境服务商信息,欢迎点击链接下载。

了解详情

美客多广告重磅更新!接入TikTok生态、打通拉美最大达人网络、完成视频购物闭环

日前,美客多旗下数字广告业务Mercado Ads发布了三项全新广告解决方案,包括将广告能力扩展至TikTok生态系统、推出拉丁美洲规模最大的内容创作者网络,以及上线可购物视频广告。