← 返回 Avalaches

伦敦信贷避险基金 Sona Asset Management 的分析指出,当前人工智慧(AI)生态圈中的各大科技巨头与核心实验室之间,存在著极其错综复杂的资金与资本承诺网络。数据涵盖截至 2026 年 8 月的三年半内、202 个实体间共 176 笔、总额达 3.6 兆美元的交易。核心实验室如 OpenAI 和 Anthropic 虽然缺乏稳定的未调整获利,却已累积了巨额的资本承诺,使整体生态系统的潜在脆弱性备受关注。

该研究揭示了 AI 融资网络高度呈现「闭环系统」的特征,在分析的交易中约有 120 笔带有高度循环性标记。例如亚马逊和微软大举投资 OpenAI,而 OpenAI 又反向向其承诺庞大的云端运算采购合约;同时辉达注资 OpenAI,OpenAI 向云端巨头采购运算服务,这些云端业者又成为辉达晶片的最大买家,资金在彼此之间循环流动,此种循环融资模式已引发标普等评级机构对超大规模业者信用风险的警惕。

除了复杂的股权与债务关系外,Sona 亦透过供应链营收依赖度与杠杆率(总债务对 EBITDA)绘制出高度集中的 AI 网络,显示部分厂商的生计几乎完全受制于少数几家巨头的资本支出决策,且如 Oracle 与 CoreWeave 等节点展现出显著偏高的杠杆。这种相互交织的营收与资本支出链条意味著任何单一环节的违约或失败都可能引发系统性冲击,评估个别企业的信用体质必须综观整个生态系。







A paper by Sona Asset Management reveals that major tech giants and core labs across the AI ecosystem are bound by an intricate web of capital commitments, covering 176 transactions worth at least $3.6 trillion across 202 entities in the three and a half years leading to August 2026. Central players such as OpenAI and Anthropic have amassed vast financial commitments despite their lack of positive unadjusted earnings, raising questions about potential systemically important vulnerabilities within the network.

The analysis highlights that the AI financing landscape increasingly resembles a closed-loop system, with roughly 120 out of 176 deals exhibiting high circularity. Significant bilateral transactions exemplify this dynamic: Amazon and Microsoft make multibillion-dollar investments in OpenAI, which OpenAI counters with massive compute and cloud purchase commitments, while Nvidia backs OpenAI, whose spending with cloud providers directly fuels purchases of Nvidia silicon—a circular financing cycle that has caught the attention of credit rating agencies like S&P Global Ratings.

Beyond financing ties, Sona mapped AI supply-chain revenue dependencies and financial leverage, finding that multiple companies depend existentially on the capital expenditures of just one or two dominant players, with certain firms like Oracle and CoreWeave exhibiting high debt-to-EBITDA ratios. Because the fortunes of chipmakers, neoclouds, hyperscalers, and AI labs are deeply interwoven through revenue, capex, and financing loops, understanding credit risk or the fallout of a single point of failure requires analyzing the AI ecosystem as an interdependent whole.
2026-09-20 (Sunday) · 4e835d6f81b74429e6b031a1f8455babc1314fa8