AI demand remains robust, but financing and cash-return pressure persist as China advances without a verified cycle shift.

The underlying risk picture is broadly unchanged: Oracle reported rapid cloud-infrastructure growth and a larger contracted backlog while capital spending rose sharply, and a separate Oracle-linked project-finance loan came under pressure. NVIDIA and AMD results continue to support accelerator demand, while US power forecasts confirm data-centre load is becoming economically significant. China advanced domestic chip, cloud and deployment efforts, but available evidence still shows lower capital intensity and tighter financing than the US, so no material China-driven change to the global cycle was established.

• Oracle spent $28.5 billion on capital expenditures in its latest quarter, principally on data-centre expansion, while requiring substantial equity and prospective external financing.
• Reuters reported that roughly $18 billion of loans tied to an Oracle-leased New Mexico data-centre project traded at 89–91 cents on the dollar, testing project-finance appetite and execution risk.
• EIA identifies data-centre development as a driver of US commercial and industrial electricity-demand growth; power availability remains a binding infrastructure variable.
• China’s AI infrastructure spending is accelerating, but Rhodium estimates Chinese hyperscalers’ combined free cash flow turned negative in the first half of 2026, adding another capital-intensive buildout with uncertain returns.

• NVIDIA reported $89.0 billion of Data Center revenue, up 117% year over year, indicating no broad accelerator-demand break in the latest reported quarter.
• AMD reported record quarterly revenue of $11.5 billion, with Data Center representing 58% of sales, supporting breadth beyond a single accelerator supplier.
• Oracle reported cloud-infrastructure revenue growth of 121% and remaining performance obligations of $664 billion, evidence that some infrastructure spending is backed by contracted demand.
• Huawei reported enterprise AI-cloud deployments and a new cluster service; Chinese competition may broaden adoption and compute demand rather than simply displace US profits.

• Whether Oracle-linked project debt is distributed successfully or requires further price concessions.
• Conversion of contracted cloud backlog into revenue, operating cash flow and acceptable infrastructure margins.
• Evidence of data-centre cancellations, power delays or capacity constraints versus continued delivery.
• Whether Chinese domestic chips and model ecosystems achieve scalable price-performance and commercial deployment under export restrictions.