China adds credible AI demand and domestic-chip capacity, but returns and financing still lag the build-out.

No rules-based score changed. Recent reporting strengthens evidence of AI-cloud monetisation in China and of demand for data-centre capacity, while also highlighting capital-intensive, debt- and equity-supported infrastructure expansion. China is therefore a two-sided cycle influence: its commercial adoption supports global compute demand, but cheaper capable models and domestic hardware progress could compress model rents and redistribute returns.

• Authoritative inputs still show 72.29% hyperscaler capex growth alongside 16.11% aggregate free-cash-flow decline, and cloud-revenue growth remains well below the capex-growth proxy.
• IREN disclosed $7.24 billion of debt-financing proceeds and $4.74 billion of equity issuance in its fiscal year while warning that its AI-cloud expansion requires substantial additional capital.
• Riot's proposed 191 MW frontier-lab project assumes roughly $1.7 billion-$2.1 billion of debt financing against $2.1 billion-$2.3 billion of expected capex, illustrating reliance on project-finance execution.
• China's reported planned 160,000-chip Huawei deployment is a large domestic capacity commitment whose installation depends on Huawei manufacturing; it supports demand but raises execution and potential overbuild risk if deployment outruns profitable usage.

• The authoritative semiconductor-demand proxy remains strong: NVIDIA Data Center revenue grew 117% year over year in its latest reported quarter.
• Alibaba reported 45% year-over-year AI Cloud and Compute Services growth, triple-digit AI-related-product growth, and a 12% cloud adjusted-EBITA margin, evidence that some AI demand is converting into revenue and profit.
• Baidu reported 50% AI Cloud Infra growth and 283% GPU-cloud growth, with positive operating cash flow; this is direct evidence of Chinese AI-compute adoption.
• Digital Realty reported strong bookings and two post-quarter hyperscale leases, while C3.ai reported positive free cash flow and 73% sequential bookings growth; neither proves broad returns, but both counter a blanket demand-collapse thesis.

• Whether hyperscaler and neocloud capex produces disclosed cloud, AI-service and free-cash-flow conversion rather than only capacity commitments.
• Terms, leverage, customer prepayments and refinancing outcomes for data-centre and GPU-finance projects.
• Evidence that Chinese open-weight and domestic-chip deployments sustain capable low-cost inference at scale, versus production, software-stack and training limitations.
• Power interconnection timelines, equipment lead times and cancellations in major US data-centre markets.