AI financing and permitting risks persist, while China’s cloud monetisation and domestic stack progress broaden demand and competition.

No rules-based signal changed. New evidence tests the credit, macro, capex, semis and monetisation signals: Nvidia-supported financing and OpenAI’s Ohio lease increase the cycle’s contingent-financing exposure; Pennsylvania and Texas policy actions raise siting and power-execution risk; Core Scientific reported delivered capacity and rising colocation revenue; and Alibaba reported strong China AI-cloud monetisation and domestic-chip adoption. China therefore broadened both the demand case and competitive pressure, rather than creating a one-way bearish signal.

• Nvidia’s proposed capital pools and residual-value support, plus its conditional guarantees on the Ohio project, shift more AI-buildout risk toward structured financing and supplier-linked commitments.
• Pennsylvania removed AI data-center proposals from fast-track permitting and imposed binding development requirements; Texas standards require developers to secure or provide power and one project was ended before construction.
• The dashboard already shows weak aggregate hyperscaler free-cash-flow growth and a large capex-versus-cloud-revenue gap; recent financing structures do not resolve that cash-return test.
• Chinese full-stack models, chips and cloud services are progressing in commercial deployment. If price-performance improves further, this could pressure proprietary-model and cloud economics, especially in inference.

• Core Scientific reported 395 MW of billable customer power capacity, sharply higher colocation revenue, and new long-term AMD and neocloud leases, indicating that some buildout is converting into contracted service revenue.
• Alibaba reported 45% year-on-year AI Cloud and Compute Services revenue growth, higher cloud profitability, and sustained triple-digit growth in AI-related product revenue—evidence that monetisation can accompany investment.
• China’s domestic hardware progress remains uneven: reporting indicates frontier training still depends materially on Nvidia hardware, and independently comparable benchmarks, pricing and power-efficiency data for domestic alternatives remain limited.
• Dedicated power commitments and stricter permitting may reduce ratepayer conflict and improve project discipline, even if they slow near-term capacity additions.

• Whether Nvidia-linked financing is disclosed with firm project-level collateral, guarantees, pricing and termination protections rather than headline commitments.
• Evidence of cancelled, deferred or repriced AI data-center projects following Pennsylvania, Texas and local permitting requirements.
• Nvidia’s next results and customer commentary for demand durability, supply constraints, gross margins and signs of capacity digestion.
• Chinese model price-performance and verified deployment data, plus whether domestic accelerators move from inference and post-training into repeatable frontier pre-training.