Risk remains elevated as financing deepens and China adds credible low-cost model and inference competition.

No rules-based signal changed. New evidence reinforces financing and infrastructure-execution risks: NVIDIA-backed capital platforms, the Ohio lease-and-guarantee structure, and expanding bespoke power-linked campuses increase the importance of utilization, residual-value and power-delivery assumptions. China-driven change was material: reported Zhipu domestic-chip scale and Chinese cloud monetisation evidence strengthen the possibility that cheaper capable supply accelerates adoption while shifting profits away from proprietary US model providers.

• NVIDIA’s proposed financing platforms could mobilize more than $500 billion for AI infrastructure, increasing the cycle’s reliance on third-party underwriting of compute assets and customer cash flows.
• OpenAI’s reported Ohio structure combines a 20-year lease, NVIDIA conditional guarantees and very large dedicated power plans; it illustrates contingent-financing and utilization risk at project scale.
• Moody’s identifies private-credit and infrastructure-style financing of specialized GPU assets, where future demand, technological relevance, power availability and utilization are central assumptions.
• Power, water, permitting and local political resistance remain practical constraints that can delay projects or raise delivered infrastructure costs despite strong demand signals.

• NVIDIA Data Center revenue growth remains exceptionally strong under the authoritative semiconductor-demand signal, providing direct evidence that current accelerator demand has not broken.
• Long-duration contracts, pre-leasing and dedicated financing can allocate construction risk to parties able to bear it, rather than necessarily indicating near-term credit stress.
• Alibaba reported strong AI-cloud and model-service growth, while Tencent expanded global access to its model through cloud and workflow products; these are signs that deployment and commercial use are broadening.
• Cheaper Chinese open models can improve enterprise economics and accelerate global adoption, potentially expanding total inference and infrastructure demand even if model-layer pricing falls.

• Whether large compute-financing platforms produce disclosed completed financings, tighter covenants, higher spreads, cancellations or revised offtake terms.
• Evidence on utilization and cash conversion at AI clouds, especially whether cloud and AI-service revenue closes the capex gap.
• Power-delivery milestones, permits and community constraints for gigawatt-scale campuses.
• Independent validation of Chinese domestic-chip performance, operating cost, reliability and ability to support frontier training—not only inference workloads.