Risk stays elevated: demand is strong, financing grows more complex, and China adds competitive pressure but no verified cycle break.

No rules-based score changed. NVIDIA’s reported demand remains a counterweight to a downturn thesis, while disclosed credit support and AI-infrastructure financings reinforce scrutiny of financing quality and concentration. China-specific evidence showed commercial cloud growth, a new Alibaba flagship-model release, domestic inference progress and tighter platform oversight; this strengthens the competition/adoption lens but does not alter the authoritative open-model or other signal readings.

• Aggregate hyperscaler capex growth and declining aggregate free-cash-flow growth remain the primary pressure; reported cloud revenue growth still trails capex growth in the dashboard methodology.
• NVIDIA disclosed credit support for approximately 4.25 GW of IT-load leases at an Ohio campus, illustrating that supplier and partner guarantees are becoming part of the infrastructure funding stack.
• Oracle reported $43 billion of fiscal-2026 debt financing and expects roughly $40 billion more debt-and-equity financing in fiscal 2027, increasing reliance on external capital even where customer GPU prepayments reduce funding needs.
• IREN disclosed financing at both investment-grade and non-investment-grade customer levels, including a 9.0% fixed-rate financing structure; this is evidence that marginal AI capacity carries materially higher funding costs than top-tier contracted capacity in some cases.

• NVIDIA reported $89.0 billion of data-centre revenue, up 117% year on year, and guided to $108.0 billion of total revenue for the next quarter, supporting the dashboard’s low-risk semiconductor-demand reading.
• Customer prepayments and long-term contracts can reduce infrastructure providers’ upfront capital burden: Oracle cited $75 billion of prepaid or customer-supplied hardware in large AI contracts, and IREN reported prepayments plus committed financing covering most associated GPU capex.
• Baidu reported AI Cloud Infrastructure revenue up 50% year on year and GPU Cloud revenue up 283%, providing verified evidence of China-side commercial compute adoption rather than model progress alone.
• Cheaper capable Chinese and other open-weight models could expand the addressable market and inference volumes even if they compress frontier-model pricing and incumbents’ economic rents.

• Whether AI-cloud revenue, enterprise usage and pricing convert fast enough to close the authoritative capex-versus-monetisation gap.
• The credit quality, guarantees, collateral, prepayment levels and refinancing terms behind new data-centre projects, especially outside investment-grade counterparties.
• Evidence of cancellations, delayed energisation, power constraints or weaker accelerator orders that would challenge the semiconductor-demand proxy.
• Whether China’s model and inference advances translate into independently verifiable price-performance gains, exportable adoption and domestic-chip deployment at scale, rather than vendor claims or policy targets alone.