AI demand is strong, but financing, permits and cheaper Chinese models keep cycle risk elevated.

No rules-based score changed. New evidence reinforces demand through NVIDIA, AWS, Alphabet and Microsoft, while financing is becoming more structured and permitting/community constraints are increasingly underwriting issues. China’s impact became more concrete: Alibaba raised equity for full-stack AI investment and reported stronger cloud economics, while Moonshot’s reported US-cloud talks test whether Chinese low-cost models can gain global distribution.

• Aggregate hyperscaler capex and weaker free-cash-flow conversion remain the dashboard’s primary pressure; strong cloud growth is not yet a direct measure of AI-specific return on invested capital.
• NVIDIA-backed platforms targeting more than $500 billion of third-party AI infrastructure capital increase dependence on long-duration utilisation and financing assumptions.
• Lenders cite permitting and community opposition as underwriting risks; Reuters reported at least 75 contested projects worth about $130 billion in the first quarter, including a terminated Virginia project.
• Regulatory restrictions on power- and water-intensive data centres could delay capacity, raise development costs and reduce the flexibility assumed in buildout plans. This is a supply constraint, not proof of demand weakness.','China’s cheaper capable models remain a rent-compression risk. Moonshot’s reported negotiations with US clouds, if concluded, could,

• NVIDIA’s reported data-centre revenue growth remains exceptionally strong, supporting the dashboard’s low semiconductor-demand risk reading.
• AWS sales grew 37% year over year, Alphabet reported Google Cloud revenue growth of 82%, and Microsoft reported $59.3 billion of cloud revenue, up 27%, alongside more than 30 million paid Copilot seats.
• China is not solely a competitive threat: Alibaba reported 45% AI Cloud and Compute Services growth, higher cloud profitability and triple-digit AI-related product-revenue growth, evidence that adoption can create local commercial returns.
• Cheaper Chinese models could broaden enterprise deployment and inference demand, potentially increasing global compute consumption even if model-layer margins narrow.

• Whether hyperscaler capex converts into disclosed cloud margins, free cash flow and enterprise AI revenue rather than merely higher usage.
• Terms, counterparties and credit allocation for NVIDIA-linked infrastructure financing platforms, including whether project-finance risk is shifted to lenders or equity sponsors.
• Data-centre permitting outcomes, grid-interconnection rules and cancellations in major US markets.
• Whether Moonshot reaches cloud-hosting agreements with Microsoft, AWS or Google, and whether export controls impede distribution or compute access in practice.