AI demand and China’s commercial progress remain firm, but debt, cash burn and power constraints keep cycle risk elevated.

The underlying risk picture is broadly unchanged and the risk reading remains 3.125. Oracle reported stronger AI-cloud contracting and lower-than-feared cash burn, while HPE raised demand expectations; against that, data-centre borrowing continues to expand and projects remain exposed to permitting, labour and power constraints. China’s impact is mixed: Alibaba reported accelerating AI-cloud revenue and broader proprietary-chip use, but Chinese chip prices rose as high-bandwidth-memory supply tightened.

• Aggregate hyperscaler capex is still outrunning the cloud-revenue proxy and aggregate free cash flow is declining, leaving a high burden of future utilisation and pricing to validate investment.
• Data-centre finance is broadening beyond the largest balance sheets: Reuters Breakingviews estimated roughly $500 billion of data-centre debt issuance this year, with weaker terms and pricing for projects awaiting power or permits.
• Power, permitting, labour and component constraints can delay revenue conversion while carrying costs accumulate; Oracle itself flagged such constraints around Stargate.
• China’s capable, lower-cost open-weight models can pressure proprietary model pricing and shift value toward cloud distribution, applications and hardware rather than frontier-model rents. That is an inference from competitive and adoption evidence, not a separate risk measure.

• NVIDIA Data Center revenue growth remains exceptionally strong, indicating that accelerator demand has not yet broken.
• Oracle added more than $30 billion of AI-cloud contracts, reported a $664 billion backlog and disclosed customer prepayments that funded part of quarterly capex, providing evidence that some capacity is contracted rather than purely speculative.
• HPE raised fiscal 2026 and 2027 outlooks on AI server and networking demand, supporting enterprise and infrastructure demand beyond a narrow group of model developers.
• Chinese AI progress can be cycle-positive: lower inference costs and open deployment can accelerate adoption and compute usage. Alibaba reported 45% AI Cloud and Compute Services revenue growth and higher segment profitability, though its AI applications unit remained loss-making.

• Whether AI-cloud backlog converts into revenue and cash flow without further deterioration in capex intensity.
• Data-centre debt terms, power interconnection timelines, cancellations and the availability of project-level financing.
• Evidence that inference demand broadens across enterprises, including server, networking and optical supply constraints.
• Chinese HBM availability and domestic accelerator pricing: sustained shortages would limit domestic supply, while resolution could accelerate competitive capacity growth.