核心要点
- OpenAI 正从软银旗下子公司 SB Energy 租用位于俄亥俄州的一座新数据中心，IT 容量约 8 吉瓦，租期 20 年。
- 英伟达为该项目提供最高 1050 亿美元的资金支持，并成为独家芯片供应商。英伟达 CEO 黄仁勋如今认为，土地、电力和建筑外壳是 AI 建设中最主要的瓶颈。
- 《华尔街日报》报道称，大型科技公司持有约 3 万亿美元的 AI 承诺，但这些并未体现在资产负债表中。分析师警告称，投资者已几乎无法评估这些公司的实际债务水平。
OpenAI 已签署租约，拿下迄今公布的最大数据中心项目，英伟达提供最高 1050 亿美元的担保。英伟达 CEO 黄仁勋将“LPS”称为下一项战略资源，而《华尔街日报》报道称，此类高达 3 万亿美元的 AI 承诺不会出现在任何资产负债表上。
OpenAI 与软银子公司 SB Energy 就俄亥俄州“PORTS-Pike”园区签署了 20 年租约。该协议为 OpenAI 提供约 8 吉瓦的 IT 容量，而《华尔街日报》指出，若将冷却和基础设施计算在内，该项目总规模达 10 吉瓦。该园区部分位于美国能源部前铀浓缩设施旧址上，电力来自一座 9.2 吉瓦的燃气电厂——该电厂由美国政府所有，日本作为贸易协定的一部分提供融资。
黄仁勋用一个新缩写宣布了这笔交易。“LPS”代表土地（land）、电力（power）和建筑外壳（shell），即场地、供电和建筑本身。黄仁勋表示，这些基础设施要素已取代芯片和网络设备，成为 AI 建设中的真正瓶颈。像 OpenAI 这样的 AI 实验室发展速度超过了其资产负债表对长期基础设施合同的支持能力，因此英伟达正凭借其资金实力介入。
《华尔街日报》描述了这一安排的运作方式：英伟达并非为OpenAI的租金支付提供担保，而是为第一阶段建设（涵盖4.25吉瓦的IT容量）中已完工数据中心资产的残值提供背书。如果OpenAI中途退出，SB Energy必须先找到替代租户，然后尝试出售这些设施。只有在这之后，英伟达才会弥补价值差额，上限为1050亿美元。作为回报，英伟达成为该园区前半部分的独家芯片供应商，并向SB Energy投资15亿美元。
黄仁勋预计每代系统约需150万块GPU，对应1500亿至2000亿美元的收入。就所有园区而言，他估计OpenAI到2030年的承诺总量约为12吉瓦的英伟达算力。如果英伟达行使对俄亥俄州剩余3.75吉瓦的选择权，整个方案将扩大到约16吉瓦，价值约6000亿美元。OpenAI表示，只为已完工的容量付费。首批800兆瓦预计于2028年上线。
3万亿美元的表外负债
《华尔街日报》的一项分析显示，这些承诺在整个行业中的规模已变得多么庞大。包括Alphabet、Meta、微软和英伟达在内的九家科技公司，持有约3万亿美元主要与AI相关的表外义务。租赁只有在付款开始后才被计入账目，采购承诺则只有在货物交付后才被确认。尚未开始的租赁合计达1.2万亿美元，是一年前的四倍。在Alphabet，采购承诺在三个月内从3320亿美元跃升至8110亿美元。
这些合同几乎无法取消，而Alphabet和亚马逊最近均报告了负自由现金流。摩根士丹利分析师警告称，投资者已几乎无法评估这些公司的实际债务水平。
《华尔街日报》此前曾报道，英伟达最初想以约2500亿美元为整个项目提供兜底，但在股价下跌5%及投资者施压后缩减了担保规模。最终1050亿美元的上限低于最近流传的1200亿美元数字，而且仅覆盖资产价值，而非全部租赁金额。
《华尔街日报》 / OpenAI 俄亥俄州
OpenAI / PORTS-Pike 项目
《华尔街日报》 / 大型科技公司支出
Key Points
- OpenAI is leasing a new data center with around 8 gigawatts of IT capacity from SoftBank subsidiary SB Energy in Ohio for 20 years.
- Nvidia is backing the project with up to $105 billion and becomes the exclusive chip supplier. Nvidia CEO Jensen Huang now sees land, power, and building shells as the biggest bottlenecks in the AI buildout.
- The Wall Street Journal reports that major tech companies hold around $3 trillion in AI commitments off their balance sheets. Analysts warn that investors can barely gauge these companies' actual debt levels anymore.
OpenAI has signed a lease for the largest data center project announced to date, with Nvidia guaranteeing up to $105 billion. Nvidia CEO Jensen Huang is calling "LPS" the next strategic resource, while the Wall Street Journal reports that $3 trillion in AI commitments like these don't show up on any balance sheet.
OpenAI has signed a 20-year lease with SoftBank subsidiary SB Energy for the "PORTS-Pike" campus in Ohio. The deal gives OpenAI around 8 gigawatts of IT capacity, while the WSJ puts the project's gross total at 10 gigawatts once cooling and infrastructure are factored in. The site sits partly on a former US Department of Energy uranium enrichment facility and draws power from a 9.2-gigawatt gas plant that the US government owns and Japan is financing as part of a trade agreement.
Huang announced the deal with a new acronym. "LPS" stands for land, power, and shell, meaning the site, the electricity supply, and the building itself. According to Huang, these basics have replaced chips and networking gear as the real bottleneck in the AI buildout. AI labs like OpenAI are growing faster than their balance sheets can support long-term infrastructure contracts, so Nvidia is stepping in with its financial muscle.
The WSJ describes how the arrangement works: Nvidia isn't guaranteeing OpenAI's rent payments. Instead, it's backing the residual value of the finished data centers in the first construction phase, which covers 4.25 gigawatts of IT capacity. If OpenAI walks away, SB Energy first has to find a replacement tenant, then try to sell the facilities. Only then does Nvidia cover the difference in value, capped at $105 billion. In return, Nvidia becomes the exclusive chip supplier for the first half of the site and is investing $1.5 billion in SB Energy.
Huang expects around 1.5 million GPUs per system generation, or $150 to $200 billion in revenue. Across all sites, he puts OpenAI's commitments through 2030 at roughly 12 gigawatts of Nvidia compute. If Nvidia exercises its option on the remaining 3.75 gigawatts in Ohio, the package grows to about 16 gigawatts worth roughly $600 billion. OpenAI says it only pays for finished capacity. The first 800 megawatts are slated to come online in 2028.
Three trillion dollars stay off the books
A Wall Street Journal analysis shows how large these commitments have become across the industry. Nine tech companies, including Alphabet, Meta, Microsoft, and Nvidia, hold around $3 trillion in mostly AI-related obligations that don't appear on their balance sheets. Leases only get recorded once payments begin, and purchase commitments only when goods are delivered. Leases that haven't started yet add up to $1.2 trillion, four times as much as a year earlier. At Alphabet, purchase commitments jumped from $332 billion to $811 billion within three months.
The contracts are nearly impossible to cancel, and both Alphabet and Amazon recently reported negative free cash flow. Morgan Stanley analysts warn that investors can barely gauge these companies' actual debt levels anymore.
The WSJ had already reported that Nvidia originally wanted to backstop the entire project with around $250 billion, but scaled back the guarantee after a 5 percent stock drop and pressure from investors. The final $105 billion cap comes in below the $120 billion figure that circulated most recently, and it only covers the asset value rather than the full lease.
Wall Street Journal / OpenAI Ohio
OpenAI / PORTS-Pike project
Wall Street Journal / Big Tech spending