Risk remains elevated: demand is strong, while financing and China’s lower-cost competition test returns.

No rules-based signal changed. Recent reporting strengthens both sides of the cycle: NVIDIA’s demand outlook and July US construction data support the semiconductor and infrastructure case, while large lease commitments and financing platforms test credit and return discipline. China produced no single verified shock today, but recent evidence of Alibaba fundraising, Chinese chip revenue growth and model releases reinforces its role as a source of cheaper supply, domestic capacity and potentially faster adoption.

• Aggregate hyperscaler capex growth and falling free cash flow remain the dashboard’s primary pressure; AI-specific revenue and profitability remain incompletely disclosed.
• Big Tech’s disclosed pipeline of uncommenced data-centre lease payments is about $1.09 trillion, creating substantial fixed commitments before facilities enter reported lease liabilities.
• NVIDIA is extending beyond chip supply into infrastructure guarantees and third-party financing arrangements, increasing interdependence among vendors, developers and AI-model customers.
• Credit-market supply is rising: JPMorgan estimates hyperscaler issuance reached $194 billion in the first half and expects technology spreads to retain supply pressure even without a credit-crisis thesis.','Chinese AI-chip challengers are reporting rapid revenue growth but remain unprofitable and working-capital constrained, illustrating that domestic supply

• NVIDIA reported data-centre revenue of $89 billion, up more than 100% year on year, and forecast 70% revenue growth in its next fiscal year; this is strong evidence that accelerator demand has not broken.
• NVIDIA says Vera Rubin systems have begun shipping, while memory shortages constrain fulfilment; supply limits, rather than cancelled demand, currently appear to be the near-term bottleneck.
• US data-centre construction spending exceeded a $75 billion annual pace in July and was nearly 60% above a year earlier, corroborating continued physical buildout.
• Chinese AI investment also shows commercial traction: Alibaba reported 45% growth in AI cloud and compute-services revenue, which is counter-evidence to a purely speculative adoption narrative.

• Whether hyperscaler capex converts into disclosed cloud revenue, operating cash flow and durable enterprise demand rather than capacity additions alone.
• Terms, final debt sizing and risk allocation for NVIDIA-backed infrastructure guarantees and financing platforms.
• Memory availability, power interconnection, land and construction bottlenecks that could defer deployments or raise delivered compute costs.
• Whether Chinese open-weight models and domestic accelerators gain exportable price-performance advantages, compressing global model pricing but expanding token demand.