今年仅五家美国科技巨头就将进行7000亿美元的资本开支，而油气行业去年在勘探与生产上的投资为5700亿美元，这意味着前者高出1300亿美元、约为后者的1.23倍（约高23%）。这一差距显示出人工智能数据中心投资正在把资本重心迅速从能源开采推向算力基础设施。

算力金融化仍处早期阶段：成立于2016年的OneChronos计划最早于6月推出算力捆绑拍卖市场，Ornn则已推出包含英伟达H100在内的芯片价格指数，并计划提供实物GPU看跌期权。若可信基准与流动性衍生品形成闭环，GPU篮子支持的债券可能像信用卡贷款和商业按揭支持证券一样被广泛分销。

核心障碍是折旧速度与区域割裂：摩根士丹利估计，Alphabet、Microsoft、Meta和Oracle未来四年可能确认6800亿美元折旧，规模几乎接近上述年度7000亿美元开支。由于算力受数据中心地理位置与时延约束、无法像燃料那样调运，区域价差显著；但若衍生品能有效转移“芯片快速贬值”风险，借贷成本与资本配置效率仍有望改善，并强化美国相对中国的金融工程优势。

Just five American tech giants are set to spend $700bn in capital expenditure this year, versus $570bn invested by the oil and gas industry in exploration and production last year, a $130bn gap that is about 1.23x (roughly 23%) higher. The spread signals that AI data-centre buildout is rapidly shifting capital intensity from energy extraction to compute infrastructure.

Compute financialisation is still early: OneChronos, founded in 2016, aims to launch a market to auction compute bundles by June, while Ornn has launched a chip-price index including Nvidia’s H100 and plans put options on physical GPUs. If trusted benchmarks and liquid derivatives emerge together, bonds collateralised by GPU baskets could be distributed more like securities backed by credit-card loans and commercial mortgages.

The biggest constraints are depreciation speed and regional fragmentation: Morgan Stanley estimates Alphabet, Microsoft, Meta, and Oracle could record $680bn of depreciation over the next four years, almost on the scale of that $700bn annual spending figure. Because compute is tied to data-centre proximity and latency and cannot be moved like fuel, regional prices diverge sharply; however, if derivatives can transfer sudden chip-obsolescence risk, borrowing costs and capital allocation could still improve and strengthen America’s financial-engineering edge versus China.

Source: The financialisation of AI is just beginning

Subtitle: Get ready for a new wave of securities, hedges and collateral

Dateline: 2月 19, 2026 04:44 上午