AI demand remains strong, but financing complexity and China’s cheaper competition keep cycle risk elevated.

New evidence strengthens the demand countercase: Nvidia reported $96.22 billion of quarterly revenue and guided roughly $108 billion. New evidence also tests financing and infrastructure risks: Nvidia-backed financing platforms target more than $500 billion, while its Ohio OpenAI project includes a contingent guarantee of up to $105 billion; lenders and regulators are increasingly focused on permitting, power and cancellation risk. China is a material incremental competitive development, not a score change: Moonshot is in early revenue-sharing talks with Microsoft, AWS and Google for Kimi K3, and Alibaba is funding chips, infrastructure and models through a $10.2 billion share placement.

• Capex and cash-return pressure remains authoritative: hyperscaler investment growth materially exceeds aggregate free-cash-flow growth, while the dashboard’s cloud-monetisation gap remains wide.
• Financing is becoming more structured and contingent. Nvidia’s planned $500 billion third-party financing platforms and up-to-$105 billion Ohio guarantee can sustain buildout, but also increase dependence on long-duration utilisation, leases and project-finance execution.
• Power, permitting and community opposition are becoming execution constraints. Reuters reports lender scrutiny of these risks, and multiple jurisdictions have imposed or considered restrictions on large data-centre development.
• China adds rent-compression risk. Moonshot’s Kimi K3 is reported to offer frontier-comparable performance and is seeking US-cloud distribution; if agreements close, cheaper Chinese models could intensify global price competition and reduce value capture by incumbent model vendors.

• Nvidia’s reported quarterly revenue more than doubled year on year to $96.22 billion, and its approximately $108 billion outlook indicates that infrastructure demand remains robust rather than showing a near-term collapse.
• Nvidia’s financing initiatives also demonstrate continued institutional willingness to fund AI infrastructure; the announced platforms may broaden access to compute beyond the largest hyperscalers.
• Alibaba reported 45% growth in AI cloud and compute-services revenue and stated that AI model-as-a-service annual recurring revenue exceeded 16 billion yuan, providing evidence of commercial demand in China.
• China’s model progress can be demand-positive as well as rent-negative: lower-cost capable models may expand enterprise deployment and token usage, raising overall inference and infrastructure demand even if model pricing falls.

• Whether Nvidia’s demand strength translates into sustained hyperscaler cloud revenue, enterprise adoption and free-cash-flow recovery rather than further capex acceleration.
• Final terms, investor appetite and risk allocation for Nvidia-backed financing, Broadcom/Anthropic-related debt discussions and the Ohio project’s eventual debt structure.
• Actual delays, cancellations or higher financing costs arising from grid audits, permits, local opposition and power-supply requirements.
• Whether Moonshot reaches cloud-hosting agreements, and whether Chinese open-weight models gain measurable international enterprise usage or trigger further pricing reductions.